I'm not going to show you every nitty-gritty detail on how to build this. What I am going to do is show you what I built, and hopefully it gives you all the ideas you need to build your own. Mine has a lot of proprietary things built into it that wouldn't really apply to you, and when you build your own, you're going to have your own things you build in. That's why I think we should all build our own rather than use one that's pre-built for us.

The three problems it solves

Let's talk about what this thing does, because it is a beast. Prospecting is a grind, and it comes with three problems this skill is built to solve:

• It eats hours. It takes a lot of digging to find the right opportunities.
• It leaves blind spots. When someone's doing this exercise by hand, there can be a lot they miss.
• It's easy to duplicate effort. Once you've put together a list and done all your signing up and outreach, the next run needs really accurate record keeping so you don't repeat what you already did.

What the skill does is use four sub-skills, and it's just like having four specialists aimed at this task. They're all hunting in parallel, and then they report up to a master orchestrator skill that dedupes the information and puts everything together in a nice sequenced outreach program.

It takes about 10 minutes to run, and what you get out of it is incredible.

You could have months, or even more than a year, of outreach work to do just from that one 10-minute run.

Like I said, we're building citations here. Most of the opportunities are probably going to result in a link, which is great. Links still matter, and we still want them. But we'd be okay with some of them ending up as mentions.

What you get: an explainer doc and a workbook

The two things this master skill outputs are an explainer doc and a workbook.

The explainer doc. I love this one because Claude reports to me on what happened: here's what I did, here's what I found, these are the opportunities I chose to focus on and why. I like it to report to me just like a human direct report, and it tells me how much I need to look into the research exercise. Is there anything I need to go back and check? Did Claude go off the rails? That kind of thing.

The workbook. You get two tabs. Your working tab has all the opportunities and tons of columns that give you really relevant data on each one. The second tab is a "how to use this workbook" explanation: best practices, color coding, useful links, and an outreach template. It's got everything, and it's there to make the whole thing handoff ready.

Because that's really what we're doing: all the upfront research with AI, and then it's up to a human to vet each opportunity to make sure it makes sense for the client, and then perform the actual outreach, sign up for a profile, or whatever the opportunity requires.

The four sub-skills

Let's talk about the four skills that go into the master skill. These were things my team and I were doing anyway, and that's where I got the idea. I built the skill, tested it, and it's been working incredibly well.

1. The Sixth City Marketing Link Prospector. Internally, we have a spreadsheet with more than a decade's worth of learnings from link building. We've got the sites that have linked to us, the clients they've linked to, wins, playbooks, failures, and insights. It's essentially a diary of all the crazy efforts we've tried over the years. This is really the secret sauce of the skill. You might not have something like this, and I'll talk about how you can build one in a minute. After I give the master skill a client, it researches that client a little, then goes into this proprietary data and grabs the bits that are most relevant.

2. The Ahrefs Link Intersect. If you're not familiar with it, this gives you other websites that link to two or more competitors, but not your client. You get a lot of really good opportunities out of this. All you have to do is hook into the Ahrefs MCP. I do it with Claude Code, and it's super easy: I just ask Claude Code how to hook into it, and it gives me a step-by-step.

3. The SparkToro prospector. This uses the SparkToro MCP that was just newly released. With the information about your client, this skill develops an audience report within SparkToro, then looks at websites, podcasts, and even Reddit threads and plucks out the most relevant opportunities. This is from the audience perspective, so you can see why each of these four activities comes at prospecting from a slightly different angle. That's what makes the final report so incredible.

4. The Google SERP prospector. To round things out, this one does a bunch of site: searches and relevant searches trying to surface opportunities. If you're a local business, that might be chambers of commerce or local roundups. It might even try to surface scholarship opportunities. There are a lot of different searches this thing does, and I worked with Claude to nail down which ones made the most sense.

How the master skill reconciles everything

The master skill reconciles all this information from all four sub-skills. Yes, it dedupes. It also focuses on high relevance only. I really drilled Claude on this one, because we want to make sure, as best we can, that AI is making decisions based on relevance.

The other thing it's really good at: if it sees something in a competitor's profile, like a leaning toward PR or press releases, it's going to name it and report that to you in the explainer doc.

Sequenced into four modules

I've also sequenced the output into four modules. If you follow this channel, you know I build my SEO strategies around modules, and I do this to focus clients on progress toward a goal rather than time, which is arbitrary. So the link opportunities get grouped the same way:

• Foundations — directories and things that are easier to get and land. They're foundational, and maybe not the strongest, but they matter.
• Velocity — guest posting, getting into roundups, and doing some PR on platforms like Qwoted and Featured.
• Growth — growing beyond those tactics into things like podcasts or an awards page.
• Expansion — social media and expanding into other platforms.

Always vet the "already links to us" flag

One thing I have the tool do is try to see if it can find a link to the client already, because what we don't want to do is reach out to a site that already links to them. It does try to do this. But I tell our employees: no matter what the spreadsheet says, whether it says they already link to the client or they don't, you need to vet that yourself.

Turning a decade-old spreadsheet into a knowledge base

I want to talk a little more about that internal resource we have for link building. When I got this thing, it was a spreadsheet with well over 100 tabs, because it's over a decade of info. And it was causing my master skill to grind to a halt. The runs took almost 30 minutes, and I had to constantly revisit Claude Code to approve the next step and keep hitting yes.

So I worked with Claude to turn that monstrosity into something lighter, something easier for AI to read and use, and something more collaborative. I had an agent run through each tab methodically and pull out the best stuff, the best insights, and de-duplicate everything, because there was a lot of repeated information in the sheet.

Now we've got markdown files and CSVs, everything broken up nicely: client-by-client and niche-by-niche insights, all cleaned up and way lighter. I host it on GitHub in a private repo and make my team collaborators. The great thing is that not only can they tap into it through Claude Code, since you just point them to the repo, but we can make updates. If I win a link somewhere, I can add that domain, say who I won it for and what niche it's good for, and update the repo. Now we've got a brand-new insight inside our proprietary resource.

A look at the output

Let me show you an example from the workbook, because this is what you're going to get. For each row, I've got the module the opportunity falls under. This one is foundational, because it's a directory. Then there's where it came from, in this case the link intersect sub-skill, and the target, MacRAE'S Blue Book. I've got DR, and then great notes: why it fits, what makes it relevant, and how to pursue it. A lot of that information comes from the Sixth City knowledge base.

Then there's a score, which is really cool, and Claude Code did this itself. If you build this and yours doesn't, I'd ask it to. The score is based on domain rating, relevance, the module, and how easy the opportunity is perceived to get. There are 95 opportunities in this output, all ordered by Claude's suggested order.

After that I've got a couple of columns for checking the listing and seeing if it exists, a spot for the analyst to leave a note or a question, and, as you check opportunities off, a couple of formulas that calculate how many we've completed and what percentage is done. That can be nice to track.

Running it again later

So what do you do when you want to do another run? The cool thing is I built the tool to first check in on any existing workbooks. It goes through the workbook, spots the opportunities that are already found, and then when it does its new run, it ignores those. Which is really nice.

Where I draw the line with AI

I'm finding that Claude Code is getting so much better and more reliable at these research tasks. Not too long ago I tried this kind of thing, where it went out onto the web and accessed tools, and I wasn't really confident in what it returned. But now, with all these MCPs out there and Claude's own research capabilities improving, these kinds of master skills are much more viable and reliable.

Of course, we still have to have guardrails. I check Claude's work. I always give it the human eye and make sure the information I'm working with is accurate. This master skill follows my philosophy for AI and content: I use AI up to a point, and then I draw the line, and that's where I want human beings coming in.

If you haven't built a link prospector with AI, I hope this inspires you to do so. I suggest Claude Code, and I suggest using a number of sub-skills that feed up into one master skill. That gives you great output and a working list, and then I'd draw the line there and have human beings vet the output and do the outreach themselves for the greatest chance of success.

What Microsoft Clarity actually is

If you're not familiar with Clarity, it's a heat mapping tool. You can actually watch real recordings of people using your website. It's from Microsoft, it's a hundred percent free, and it's really easy to install.

The announcement: AI search data comes into Clarity

The big announcement is that they're pulling AI search data into the tool and marrying it with the UX and user engagement data you already get from the heat maps and recordings. That's an unbelievable pairing, and this is real data from Microsoft and its partners.

The AI overview dashboard gives you a look at queries cited, your share of authority, some grounding queries, and even the pages that are cited in AI responses. An earlier announcement brought this AI search information into the tool. Today's announcement pushes it even further.

First-party data, not third-party

One of the first things Microsoft talks about in the announcement is the data itself, and this is Microsoft calling out other GEO tools. The first point they make is that this data is from Microsoft AI and select partner grounding, whereas other GEO tools use third-party data, and we know that can never compare to first party. From there, they lay out what makes it different:

• Real query-level data rather than estimates.
• Actual AI-generated citations, nothing synthetic, which third-party tools have to rely on.
• Integrated CDN-level data that gives you info on bot traffic (more on that later, because it's got some useful stuff too).

The full path from discovery to engagement

What's so cool about this tool is that now you can see the full path from discovery to engagement. Someone searches something within an AI platform, the assistant gives them an answer and cites your page, they click on it, and they visit your site.

You can see what they searched, what page got cited, and then when they clicked through, you can literally watch them interact with it.

All of that gives you actionable information, and that's one of the things I love the most about this announcement. It's highly actionable. Microsoft is telling you how to use the tool to make improvements to your site and to your strategy.

See where your brand is being cited

The first bit is seeing where your brand is being cited in AI responses. The data lets you compare how often your content is cited against competitors across Microsoft experiences. It's not Google data, but it's still really solid. From there, you can:

• Find the questions that trigger citations so you know what content to focus on next.
• Improve the pages that are cited most often to get more value from what's already working.

We know improving what's already working is always a great strategy. In fact, the way I typically start isn't even with SEO, it's with CRO. Improve what you already have, then look to build new assets.

Topics: rolling queries up to spot strengths and gaps

The second part of the announcement focuses on something called topics, and I want to lean in on this one because it's a fantastic update. It's rolling out gradually, so you might not see it in all your accounts right away, but you will in the near future.

I love looking at AI search queries this way. Rolling them up and aggregating them into topics lets you spot where a site is stronger or weaker, and that's exactly what this helps you do. In their example, they've clicked into the topic electric bicycles, and you get your competitive share and citation rate for it.

But if you scroll down further, the real gold is top content opportunities: the topic areas where you're weaker and not covering very well.

In the example, that's battery lifespan and replacement expectations for electric bicycles, a subtopic where you might need more content.

On the right side it also gives you your top content to AI responses, so you can see how your top pages are contributing. Going back to the announcement, Microsoft frames the topic tool as turning AI citations into a competitive advantage: compare how often your content is cited against competitors, find the questions that bring up your content, and improve the pages that get cited most to drive bigger gains faster.

The only thing I'd add is that gap feature I called out as probably the best one. It helps you spot where you're weaker so you can create content and hopefully close some of the citation and authority gaps you have with competitors.

Which AI bots are crawling your site

The next part is really interesting: you can understand which AI bots are crawling your website. Here's what that gets you:

• See which bots visit your site and how often they show up.
• Tell the difference between crawling and search traffic so you know how your content is being picked up.
• Find the pages bots hit the most so you can focus fixes and improvements where they count.

That last one is my favorite. Microsoft is using your CDN and your log files to understand bot behavior on your site. In the tool you'll see your bot operators, like OpenAI, Anthropic, Meta, and Google, each with a percentage, plus the page paths and resources those bots are visiting.

An amazing update to an already amazing tool

All told, this is an incredible update to an already amazing tool. Clarity on its own, before any of the AI stuff, was already fantastic, and we use it at our agency regularly to watch people use our clients' websites and pull insights out of that exercise.

Now we have AI search data coming in too, giving us insight on how we're doing against competitors, which pages or topics to lean into, and what the bots are doing when they visit. It's great data to add to your strategies and your keyword research, with a ton of uses.

I love seeing Microsoft push the envelope on the kinds of AI data they're giving us, and honestly, they're outpacing Google on this one.

We've seen a lot of great updates in Bing Webmaster Tools, and to see it here in Clarity is awesome.

What this data actually shows

The first thing to understand is exactly what data this is going to show. Right from the announcement: site owners and creators will now understand how their social and video posts perform on Google Search and Discover. That line is really important, because this is not data coming in from your YouTube channel telling you how many times your video got clicked on YouTube.

This is your video showing up in a Google search, someone clicking it, and that data coming into Google Search Console.

To make it clear, if I search something like how to shell a lobster and look at the results, there's going to be a videos section. If your video shows up there and you get a click, or even an impression, it's going to come into Google Search Console. You've got your AI overview, and you've got short videos as a section as well. Any one of these, if you get an impression or a click, it's going to show up.

It only covers four platforms for now

The other thing to keep in mind is that this only applies to four social platforms for now, as per the announcement:

• Instagram
• TikTok
• X
• YouTube

The reports you're going to get

There are going to be a few reports associated with this within Google Search Console:

• A performance report. This is very similar to what you already see in Google Search Console and are used to. It shows your clicks, your impressions, and additional metrics.
• Query data. This is going to be highly actionable, because you'll be able to see the search terms that associate with your different social assets.
• An insights report. This one helps you make sense of all the data coming in. Google's example screenshot shows a report noting that YouTube content had 17,800 clicks in the last 28 days on Google Search, and I'm sure there are plenty of areas you can click into for further insights.

They also say you can export the data. Personally, I'm hoping the Google Search Console MCP that hooks into Claude Code will give us access to this, because it would be really nice to just talk to Claude Code and get insights on the data coming in here.

How to sign up, and when you'll see it

Here's how you sign up. In the top left of Google Search Console, where you'd click to see a dropdown of the properties in your account, you go to the bottom and click Add Property like you normally would. From there you should get the option to add any one of those four social platforms, or maybe multiple.

When will you actually see it? That's kind of unpredictable. I'm not seeing it for any clients at my agency yet, and I'm not seeing it in any of my personal accounts yet either. Some people I follow on LinkedIn do see it already, and they have pretty sizable social followings, so maybe that's why they got it first. Google says it's rolling out over the next few weeks, so be on the lookout.

The marketing pinball machine

On that topic, I want to talk a little more about socials and how important these are. We've kind of gone away from the traditional marketing funnel with top, middle, and bottom. People don't follow such a rigid, structured path anymore.

The folks at SparkToro refer to this more as a marketing pinball machine, where people bounce around from social platforms to websites, all over the place. Here's one way it might play out:

• Someone notices something while scrolling on Instagram.
• They hop over to YouTube and watch a couple of videos to learn more.
• They go to Reddit to see what people are saying and get that human-to-human trust building.
• They land on a brand, do a Google search, and convert.

That's just one way it can play out. If you really focus on your own path toward buying a product or booking a service, it is all over the place. So it's becoming more and more important for businesses to get on social platforms.

It's a heavy lift, but it pays off

That's really prohibitive. It's difficult, especially the video platforms. It's a heavy lift, and it's hard for businesses and brands to be consistent with it. But it helps a lot.

If I search something like build a retaining wall uphill and look at the AI overview and the first traditional search result, I see a video from SRM Contracting. Click into that channel and this guy has well over 63,000 subscribers on YouTube, and his most popular videos have millions of views. A good portion of those millions won't become converting customers, but some portion will, and it's certainly helping his brand exposure.

His website has a domain rating from Ahrefs of just one, and yet that YouTube channel is one of his primary assets on digital.

The site is nothing to write home about. We can assume his PageRank isn't very strong and his authority isn't really there, so the website isn't doing him any favors. The channel is carrying the load.

Look at a giant in the local space, Roger Wakefield Plumbing, and he's got 725,000 subscribers on YouTube alone, plus one heck of a platform spread. He's on Facebook, Instagram, pretty much anywhere you look. But he's been working on this. His oldest video is nine years old, so he's been putting in the effort for almost a decade, and he really understands the social game.

This isn't just for local businesses

Something else that's really interesting is that this doesn't just apply to local businesses. Ten months ago I made a post about the Edward Jones YouTube channel and how I thought it was really under-nurtured. In another post a year ago, I compared them to one of their competitors, Schwab, and showed the difference in YouTube as an initiative.

At the time I posted, Edward Jones had around 11,000 subscribers. They now have 12,600. Charles Schwab, in that same timeframe, went from a little under 500,000 subscribers to 545,000. Schwab was way bigger to begin with, they've been investing much more heavily into YouTube, and it's paying off: tons of brand exposure, a lot of trust, and way more growth in that period.

Do what we did as young SEOs

I'm really bullish on YouTube, and this Google Search Console update just got me thinking about social media and how, as marketers, we need to find ways to convince our clients to get on social platforms, make consistency as easy as possible, and guide them through the whole thing.

We also need to challenge ourselves, because social is a different game. There's some overlap between traditional search on a website and social, but there's a whole lot that is completely different. So if you haven't yet, I'd advise doing the same thing we did as younger SEOs in years past. We started our own websites, we experimented, and we gained our own learnings.

This is an exciting update from Google, and they're recognizing the importance of social media. As marketers, I think it's important to help our clients expand far beyond just websites and Google Business Profiles so they can reach their full potential and pull ahead of the competition.

But let's be real: that range is pretty arbitrary. Maybe it's the average, but a lot of context and accuracy gets lost inside an average. So instead of guessing, here is a quick, simple way to grab some real numbers, the kind of measurable benchmarks you can use as goals, and forecast far more accurately.

Start by benchmarking what you are up against

This isn't about producing one absolute score. It's about understanding the shape of the gap between you and the competition. Is it a small gap, a huge gap, or are you neck and neck, where a few improvements would let you pull away from the pack?

That read changes how you talk to a client and how you set the strategy. A small gap is something you attack right away. A big gap just means a particular goal needs more runway and a longer track.

You do not have to catch up to compete

Here's the mindset that matters most. Say you're looking at a local business, and a couple of competitors have around 200 reviews while your client has 20. They have ten times what you do. It can feel hopeless, and it's easy to accidentally make it sound hopeless to a client.

Don't. You do not need to hit exactly 200 reviews to become competitive. We fight this on multiple fronts at once. The big gaps become long-term projects, and the smaller gaps are ones we can often close entirely, and quickly.

The five things to measure

These are the quick, grabbable numbers that tell you where you stand. This example focuses on a local business, but the method works for a national brand too: just drop the local-only metrics and swap in others.

• Authority. Keep it simple with Ahrefs Domain Rating, plus the number of referring domains. If you want to go one step further, look at link velocity too.
• Reviews. Count Google Reviews. For a local client, it's also worth checking other third-party review platforms and any high-relevance industry sites.
• Social strength. Look at follower counts. You can track the single most popular platform per business, or loosely aggregate across platforms to get a directional number.
• Footprint. Grab the number of ranking keywords and the number of pages on the site, to see how wide a net each competitor is casting.

One caution on footprint: more pages and keywords don't automatically mean a competitor is winning more business. A lot of those pages could be off-topic and doing nothing for them. These numbers are directionally useful, not a verdict.

Turn the gaps into a timeline

Once you have the numbers, you size each gap and do a little math. To do that, you need a realistic monthly pace for each metric. Here are reasonable starting estimates you can adjust to your own experience:

• Reviews: about 4 per month on the lower end.
• Referring domains: about 2 per month with an active backlink program.
• Domain Rating: roughly 1 point per month (honestly a lofty target, so adjust down if needed).
• Content pages: at least 2 published per month.
• Ranking keywords: about 20 per month if you're publishing twice a month and earning links.

With a pace attached to each metric, the projection is simple arithmetic. I use a small spreadsheet that runs this for every metric automatically:

Months to close = (your target − where you are now) ÷ monthly pace.

The target is usually the "pack floor," the lowest competitor number, because you don't have to beat the leader to get competitive. That single equation turns a vague "three to six months" into a specific, defensible set of timelines per front.

A word of caution on timelines

I'd much rather focus on progress than the calendar. Have we been publishing consistently? Has the client been quick on approvals? Those actions are what make the graph go up and to the right. Use the timeline as a planning tool, and always pair it with the work required to hit it.

When a gap looks impossible

Sometimes this exercise surfaces a number that looks daunting, the kind that makes you think "I could never close that." When that happens, two moves help:

• Lower the target. Go after longer-tail, niche keywords where the competition isn't as fierce. Almost every site has a pocket of keyword territory where it can win. Get in there first, then build out.
• Lead with the fronts you can win now. Attack the smaller gaps and shorter timelines first, and frame the bigger gaps as a longer parallel track.

A live example: a plumber in Austin

Let's walk it. I searched the client's primary keyword (plumber in Austin, Texas), picked a fictional client called Beyond Wow, and used the other two businesses in the map pack as competitors. Here's how the benchmarks came out.

Authority. The client's Domain Rating was 25, against competitors at 34 and 35. Measured to the pack floor of 34, that's a gap of 9. On referring domains, the client had 461 versus 626 for the competitor floor, a gap of 165 links to build.

Reviews. The client had about 2,500 reviews. Competitors had roughly 17,000 and 2,900. Because we aim at the pack floor of 2,900, the real gap is only about 400 reviews, significant but very catchable with a dedicated review campaign.

Footprint. The client ranked for 237 keywords across roughly 40 pages. One competitor had 523 keywords on 128 pages. The other had 1,300 keywords on just 117 pages.

That last number is exactly the kind of thread this exercise hands you. A competitor ranking for far more keywords with fewer pages is worth investigating. How are they being so efficient? That's a lead to chase, not just a stat to record.

Reading it all together, the plan of attack writes itself: backlinks and reviews are the long fronts, and content is a clear push given the keyword-and-page comparison. The shorter gaps, like Domain Rating, are where you show early wins.

The bottom line

That's a market evaluation you can run in minutes. It tells you where the client really stands, which lines of attack to prioritize, and lets you make far more reasonable projections than a guessed-at "three to six months."

Purna Virji: AI Has to Prove Its ROI

The opening speaker was Purna Virji, and my takeaway from her speech was that AI has to prove its ROI. She said time saved is the ultimate vanity metric, and that's definitely something we give AI a lot of credit for, being really fast. But is it actually making things better? And is it actually impacting the bottom line? That's really what you should be focusing on, not just speed gains.

One of her great quotes was, "Better comes before faster." It's fine to be fast, but make sure it's actually making things better. And how should it be making things better? She gave three areas: it should be improving the quality of output, it should be increasing your scope or your competitive advantage, and it should be increasing capability, but sustainably. That was the kickoff of SMX, and it was a great mindset speech.

Kelsey Libert: Become a Source, Not a Summary

The next speaker I attended was Kelsey Libert. Search is expanding across channels. We know this is happening, and marketing is going to have to get more expansive and go beyond just websites. So yes, SEOs, we're going to have to crack the code on socials. This is in line with everything Rand Fishkin has been talking about for a while.

AI search had trust early on but has since lost a lot of it. People want disclosures if you're using it. On my LinkedIn, I've talked a lot about human sensitivity to AI and how many people are out there rejecting it when they identify it. So if we're using it for our clients and the brands we work for, we have to be really careful. You have to govern it. The best thing you can do with AI is give it a lot of context so the outputs aren't average or generic, and you want to supervise it every step of the way so it's not getting off course, hallucinating, and degrading trust in the brand.

She had some stats here. There's a quality problem with AI: 53% of marketing work uses it, but only 26% say it's faster and better, and 48% say it's making the work more average. That's right in line with what Purna was talking about. Adoption pressure is really high for analytics and SEO, so we actually have a lot of responsibility to use it appropriately.

If you're thinking about Google's AI Overviews, don't just let your content dissolve into the answer without your brand mentioned. You really want to be doing the long-term hard work to build a moat: build a brand, build yourself as an authoritative entity on the web. That's tough, but she gave some tips on how to do it: lean into original data, proprietary studies, and digital PR. We do a lot of this at the agency I work at, where everything we do is expert-led. We schedule regular Q&A sessions with our subject matter experts to keep things as unique as possible, make it hard for competitors to copy us, and make it more likely we're going to be selected by AI. Because if you're writing to the average, what compels AI or any other system to select you?

She talked a lot about repurposing to help build yourself as an authoritative entity, and that's something our agency is big on too. We love starting with video, because video can become long and short-form video, and you also have the transcript to turn into long and short-form text. All of those formats can go to a lot of different platforms, so it really helps with your spread across the web.

Her framework for success in AI search:

• Monitor platforms and choose the ones relevant to your audience.
• Build entity authority by getting your subject matter expertise out there.
• Triangulate visibility with a good spread of platforms where you're publishing.
• Govern for trust when it comes to AI.

Dave Davies: Can This Text Stand on Its Own?

Dave Davies was a more technical speaker, but there were some great takeaways we could apply to our work. When it comes to copywriting, his question was, "Can this block of text stand on its own and answer a specific question?" I just had an example of this the other day with one of my copywriters who wrote a key takeaway that was really dependent on the context from the paragraph that preceded it. You have to be careful with that, because it was a great key takeaway, but it couldn't stand on its own, so it might have a little more trouble being selected for an AI answer.

He also talked a lot about how powerful third-party mentions are for AI visibility, and that was a through line with a lot of speakers. It's not surprising, but there was a lot more talk of mentions being just as powerful, if not more powerful in some cases, than links. And he emphasized that freshness is important when it comes to AI, but if you're going to update your content, make sure the updates are significant. You can't just go in and change the last updated date and think that's enough.

Andrew Beckman and Andrew Shotland: Extracting the Gold From Local Owners

Andrew Beckman and Andrew Shotland held a local SEO seminar that was like an open forum to share your issues. It was really good. I actually went up to Andrew Shotland afterward and talked to him about how local business owners often have content that could grow their business locked away in their mind, and how our job is to extract it and get it out onto the web.

I shared some of the ways I do that. Our monthly meeting is sixty minutes: the front thirty minutes is reporting and strategy, and the back thirty minutes is an SME Q&A session. Another way we try to extract subject matter expertise is to send questionnaires, whether there's a new page we want to create, a page we want to update, or a PR request for expertise. Then I asked him, "What other ways can we extract this gold from our busy local business owners?" And he said, "Why don't you ask if they record sales calls?"

I thought that was a great idea and wanted to share it. Sales call transcripts and recordings: you can mine those for subject matter expertise and customer pain points. There's a lot of gold in there. You can also start to understand the language customers use, and that might alter your copywriting a little bit.

Amanda Farley: Conversational Queries and a Local Video Tip

Amanda Farley leaned more into the e-commerce side, but she had a local tip I'll share too. She had a system where she took in data on search behavior, product intent, and audience intelligence, and gained insights she could then apply to product pages.

What we're finding with conversational queries is that there's a lot more detail about what people are looking for. They average about 24 words, from what I heard in another presentation, whereas traditional SEO keywords average about four words. So there's a lot more data in these prompt-based queries, and we have to account for that on our pages. One example I've given is someone might prompt for men's Nike shoes, size 11, that are good for both running and hiking and also for narrow feet. There's a lot of detail in there, and if our data is telling us that's important to people, we want to update our pages accordingly so the AI system can make those connections. It doesn't mean we have to make our pages a mess of fine details, but you do want to do this research so you can gain those insights and then intelligently and tastefully update your page content.

Her one great local tip, which harkens back to what I talked about with Andrew Shotland, is to capture local video in the field using Meta glasses and build a video bank for your clients. Then again, you've got video you can use on multiple platforms, and if there's any talking, it can become text. Meta glasses are not as expensive as I thought they were, so this is one I'm going to try.

Dawn Anderson: Lost in the Middle

Dawn Anderson gave quite a technical talk breaking down how AI systems and large language models read text. One of the big things was a paper called "Lost in the Middle": if you have an enormous amount of text or context, the large language model will tend to read what's at the top and the bottom of it. So the takeaway is to put your most important stuff at the top of the page so it doesn't get missed.

Then she came in with a great point: a lot of this is just good journalism and common sense, isn't it? That's definitely something for SEOs and marketers to take away. With AI coming into the fray, things are getting really technical. We're looking under the hood and trying to understand what's behind these systems, and I think that's great to do. But at the end of the day, it boils down to principles of good copywriting and SEO fundamentals. Get into the detail and understand it, but don't get lost in it and lose the bigger picture.

Monica Ho: Treat Reviews, Search, and Social as One System

Monica Ho shared a presentation based on a large data study of businesses with multiple locations. Here are some cool takeaways:

• Reviews are a gate. With AI especially, she found you don't even get in unless you have 4.3 stars.
• One in four people use social as primary search. That harkens back to what Kelsey and Rand Fishkin have been talking about: search is fragmenting, and you want to be everywhere your target audience is. SparkToro is a great tool for that. You type in some information about your target audience, and it shows you the social platforms they use, the AI platforms they use, and a ton of other great data, so you can get your content in front of them.
• One in three people use generative AI in over half their searches. A lot of people are using AI search.

She emphasized auditing your data, and this is tough to do. I think we're going to have to be more thorough as SEOs.

This is the kind of stuff we have to monitor, because the AI systems are grabbing onto it, putting it in front of people, and confidently telling them it's current information.

She said to treat reviews, search, and social as one system, not as separate fragmented teams, and I think that's a great mindset.

Beth Nunnington: Build Something Worth Finding

For AI search, pretty much everything Beth Nunnington said was great. First of all, stop chasing GEO or LLM summaries and build something worth finding. This is sort of what I was getting at with not getting lost in the detail or pulled into the hype of getting into an AI response. Think bigger than that.

So lean into your experts. From Beth: "Your experts are your strongest asset." I totally agree, especially in the AI era. They always have been, but now especially. It's not about manipulating outputs. Do the long-term hard work. Quick wins are fine for SEO, but they tend not to be lasting. You want to do things that are going to be sustainable despite how many Google algorithm updates are happening. Build a strong foundation, then build on top of it with good SEO fundamentals and strong PR.

This SMX definitely focused a lot on third party: the wider web pointing to you with something, a link or a mention. That is really how to build yourself as an entity, how to build authority, and how to get into AI outputs. Nothing new, but good to reinforce, and I think it's even more important when it comes to AI. She also emphasized that true brand building needs multiple teams working together, which is the same cohesion point Monica was making.

Grant Simmons: A Quick Word on EntityMap

Just a quick mention of Grant Simmons. He's involved in entitymap.org, and I encourage you to visit the site, see what this proposed protocol is about, and try the tool where you can experiment with it. It's quite similar to Schema, but it's a different kind of thing. I think it's a supportive layer for getting your content and subject matter expertise out there on many platforms with a lot of third-party support, confirmation, and additional context. Those are the primary drivers of AI and search systems understanding you as an entity, and the stuff at entitymap.org can reinforce that even further.

Kyle Risley: AI Will Work Around Our Mess (To a Point)

Kyle Risley was from Shopify, and there was a lot of talk about AI agents coming and how they're going to do a lot more of the navigating people used to do. You need to make sure that from a technical, foundational standpoint, you're as sound as possible. But at the end of the day, he emphasized that you just need to nail the basics. AI will work around our mess.

He invited us to raise our hand if we thought people were going to make perfect websites. Of course they're not. Google deals with this already: there are websites with multiple H1s on the page, and Google can figure out the actual primary heading through other clues. Google learned to work around our mess, and AI is going to as well, but only to a certain extent. You do need to make sure you get those basics right.

The Four Bigger Takeaways

That's a little from each individual speaker. Here are the four larger takeaways.

So there you have it, all my takeaways from SMX Advanced Boston this year. I hope that was useful. If you have any questions, drop me a comment. I love hearing from you, and I'll see you in the next one.

But before we get into how to hook up the SparkToro MCP and my own use case, let's zoom out for a second.

Zoom out before you go nose to the glass

Let me catch you up on a conversation I was having with one of my large university clients. They were asking me about AI strategy, and they said Adobe has approached them about this content tool where you can tweak your content and optimize it for AI search platforms.

I think that's just fine to do, but it's a little bit nose to the glass when it comes to strategy. I prefer to zoom out and think bigger picture first.

And that's exactly the SparkToro workflow I'm going to show you today.

The three-MCP workflow

So here's what I did, stacking three MCPs together in Claude Code:

1. Screaming Frog MCP. I used the Screaming Frog MCP with Claude Code to crawl a specific segment of their site and isolate it to just the blogs, because I wanted to know what kind of coverage they had currently.
2. Otter AI MCP. Then I used the Otter AI MCP to add current priorities as context. Otter is an AI transcription tool, and I had this year's kickoff call saved in there. So I referenced that and brought it in as context just to be sure the output I was getting was going to be aligned with client priorities.
3. SparkToro MCP. With those two things done, I used the SparkToro MCP to find audience research keywords, AI prompt topics, and even social platform data, to tell me where on the web their audience was asking questions, doing discovery, et cetera.

Hooking up the SparkToro MCP

Before I show you that output, I want to show you how I found the SparkToro MCP documentation that tells you how to hook this up. I literally just Googled sparktoro.com MCP. It brings you to their official announcement, where you can connect your audience research directly to your favorite AI tool. There's even a use case in there, so I'd recommend reading it. But it links to the official documentation, which you can then open up.

And then here's your getting started steps. Grab that, and then just go into Claude and show it to Claude. Just copy everything on the page and tell Claude you want to hook into the SparkToro MCP. It's really that easy.

It's probably going to open up a tab if you haven't signed in, and you'll just go ahead and sign in. You'll authenticate, and then you are good to go.

What the coverage analysis revealed

I did those three things, a Screaming Frog crawl, I tapped into Otter for priorities, and then I used the SparkToro MCP. I just gave it my client's URL. I asked it to create an audience for this client, and then I told it I want to marry this audience research data with their current blog content coverage, and I want to spot the gaps. To create the page, I also used Claude's front-end design skill.

First of all, it spotted 42 blog posts and three program hubs that were covered. We've got five essential gaps, and then it gave me the total pages crawled, which is not too useful.

And look at the coverage plot. This is the really cool visual here. Here's where we're over-served, and you can hover over these and see we've got tons of faculty spotlights, and some soft skills and how to balance workload and things like that. Here's where we have our strengths, and I can hover over it and see where we're good there.

But then we've got our priority gaps, and this is really the useful stuff. Rankings, best business schools, that's pretty typical. People are looking for that kind of thing, and that showed up in SparkToro's AI prompt topics data. We know from query fan-outs, if you've seen them at all and studied what AI searches for based on what you prompt, it very often puts "best" in front of a query to put together an output. So that's not surprising.

But a lot of this stuff is really useful. We've got people asking about cost, tuition, scholarships, and it's a gap. We don't really have much content for that, so we're probably losing some prospects based on that.

Where the audience actually spends time

Then it tells us a little bit about the SparkToro data that went into this. You can see these are AI prompt topics. The top one with the highest affinity in SparkToro, which just means your audience is most interested in it, is best business schools rankings for this particular university client. We've got a ton of other great topics here, and it'll tell you if it's a gap, covered, or partially covered.

Now it's also got some SparkToro data on social platforms where this audience spends their time. They spend most of their time on YouTube, really critical. Then we've got Facebook and Instagram, got some Reddit, and then LinkedIn. Now there's one breakdown I want to zoom in on. Listen to how insightful this is, and this shows you what AI can do when it has the right context:

• YouTube isn't just number one. The specific channels this audience watches are MBA and grad admissions creators. They go to video to research programs, which is the strongest case for putting program content on YouTube.
• Facebook and Instagram score high, but for almost every audience. So there are no specifics there, and that's why it's going to recommend we go for paid amplification, not organic priorities.
• LinkedIn over-indexes for a career pivot crowd. That kind of tells you a little bit about what some of those short-form posts could look like.
• Reddit is where they ask questions, so engage in the threads. That one's a no-brainer, but the stuff before it is really good.

From a gap to a specific repurpose plan

Then I wanted it to give me some specifics. We have gaps, and you're giving me topic areas, but what's an example of a specific post we could create to cover that gap? So you'll see under cost, tuition, and financing, we've got "how much does an online MBA cost and how to pay for it." Then we've got comparisons beyond just online and full-time, so MBA versus MSBA: which master's fits your career?

Another section of this output I really liked was the way it advised me across platforms:

1. Start on YouTube. We're going to talk about a program, what an online MBA really costs. Start with the video, because that video could then become a blog post. You just take the transcript and you repurpose it. That's a really powerful move for AI search strategy.
2. Move to the blog. Now we're on the blog, and we're saying "how much does an online MBA cost and how to pay for it." That's our topic there.
3. Go short-form on LinkedIn. We go from long-form to short-form text, and we're going to have a series about the sticker price of an MBA isn't what you actually pay. I would probably break that up into about three to five different posts.

And that's a really nice repurpose workflow based on a gap.

Give this workflow a try

That is the SparkToro MCP, and it's super easy to hook up. I would suggest giving this sort of topic coverage workflow a try. It's going to reveal some really interesting gaps that you can fill with some non-commodity content, as Google calls it, and perform better in AI search.

Here are four areas to focus on over the next twelve to twenty-four months if you want real success with AI visibility: subject matter expertise, audience research, relevance networking, and YouTube. You can watch the full breakdown below, or read on.

1. Lean Hard Into Subject Matter Expertise

I'm starting here because I've been having so much success with it. When AI writing started becoming common, I naturally rejected it. It seemed insulting to the people who would actually read the content, and I had a feeling search engines wouldn't love it either.

So I leaned into expert-led content instead. The format is simple: schedule a fifteen to thirty minute expert interview once a month. That's it. From that small block of time, you get a ton of material to work with.

You can use it two ways. Add color to content that already exists to bolster your current pages, or build entirely new pages around it.

My one rule: go nearly verbatim with what you get, whether it's a live interview or a written questionnaire. I'll pass the answers through an AI for light cleanup, but I deliberately tell it not to change much. I don't want AI writing here. That's the whole point of the exercise.

This stuff performs incredibly well. My clients are getting pulled into AI Overviews, cited in ChatGPT, and ranking in the traditional listings too.

2. Take Audience Research Much Deeper

This includes keyword research, but keyword research is a smaller piece of a much bigger exercise. This is where SEOs really need to push into new territory.

One thing we've learned from the longer, conversational queries people use on AI platforms is that they're complicated. A single query often hits several audience needs at once, sometimes in combinations you wouldn't expect. So you have to max out the empathy and go deep on what your audience is actually asking, what worries them, and what myths and misconceptions they hold.

Here's where to mine for those gaps:

• Client conversations. They know their audience. Ask about it in your kickoff and beyond.
• Sales team feedback. They know the questions, the objections, and how to overcome them, which is gold for your copywriters.
• SparkToro. A treasure trove built for audience research: the sites your audience visits, podcasts they hear, channels they watch, keywords they search, and the AI platforms and prompts they use.
• Reddit. A fantastic place for audience insights. Add .json to any subreddit thread URL to see every comment, reply, and upvote. I wrote a whole post on the Reddit .json trick.
• Your own data. Social listening, internal call transcripts, and the forms that come in from interested customers.

Keyword research still lives under this umbrella, and so does prompt research. SparkToro can hand you real prompts to work from. You can also open Google Search Console and look for queries of, say, ten words or more.

You can even connect Search Console to Claude through an MCP and let it do the digging for you.

One more layer: query fan-outs, the related questions an AI engine spins off from your original prompt. The friendliest way to see them is Perplexity, which just reveals them to you. If you're more technical, you can watch ChatGPT's fan-outs in Chrome's inspector on the Network tab.

3. Treat Outreach as Relevance Networking

We all know link building outreach. I'm not saying abandon it. This is more of a mindset change, and that switch alone has gotten me better results.

I've started thinking about outreach as networking. If I'm on a service like Qwoted thinking like an outreach specialist, I'm just blasting journalists with my best responses. Fine. But the networking mindset makes it personal. I'm friendlier and more casual, and I invite follow-ups far more often, because I'm trying to connect with the person.

More often than not, the journalist responds. Sometimes we hop on a call, and a relationship starts to form. Whether or not I land that first placement, they're more likely to think of me for the next article. That's the networking paying off.

There's also a lesser-known SparkToro feature for finding genuinely relevant third-party sites that might link to or mention you. I covered it in a separate video. It leads to real back-and-forth over email, and I've had a lot of luck building ongoing relationships with publications relevant to my clients. We get cited, and we end up in their Rolodex for future opportunities.

4. Build Your Presence on YouTube

If you saw the recent Guardian article, YouTube just overtook Netflix for average daily viewing. So if you think your potential customers aren't on YouTube, you're wrong. They're absolutely there. It's a massive search engine, second only to Google.

It's time for SEOs to learn YouTube, and it is a different game. You'll have to prepare yourself through learning and experimentation, and you'll have to prepare your clients too, because getting them on video regularly is its own challenge.

Making it easy on clients is the most important and most overlooked part. The monthly SME interview is one path. For a local business, it can be as simple as turning on the camera at a job site and capturing thirty seconds of footage. I even heard a tip at SMX Advanced in Boston this year about using a pair of Meta glasses, which might be easier than a phone.

A few other things you'll need to learn to talk through with clients:

• Timelines. We're used to telling clients a website takes three to six or six to twelve months. YouTube is different and can take up to two years, so have case studies and the right language ready.
• Optimization. A different game, but honestly not too bad once you learn the baseline best practices and keep up with the trends.
• Audience growth. You have to engage and listen far more than you would with a website.
• Lead tracking and attribution. So you can account for the effort and show the value of your work.

My best advice: treat it the way we treat websites. The single best thing you can do, better than any course, is start your own channel. Post, experiment, and learn what works firsthand. You can do that right alongside your client work. Clients come in at different stages anyway. I've had brand-new channels and ones with 100,000 subscribers, and both were valuable because the game shifts at each level.

The Bottom Line

Do these four things and you'll thrive in the AI search era. None of them require a vector embeddings degree. They reward depth, empathy, and showing up consistently.

Do these four things and you'll do quite well in the coming years.

It gives you some visualizations on how the concepts within your content map to each other, which ones are covered well, and which ones are missing. And then, like it said, it's going to give you recommendations, which I love. It's going to tell you how you can update your content, connect concepts better, and cover them more thoroughly.

But the other thing it does, which is really cool and isn't immediately obvious, is it tells you if there's a concept you're covering that actually deserves its own page. I really like that.

Running one of my own posts

I ran one of my own blog posts through this tool: how to choose the right keyword for any page. So it's just an SEO basics kind of post. What you do is give the tool the URL, but you can also paste in copy, which is really nice if you're a copywriter and you want to test one of your drafts. I chose the URL option and went ahead and performed the analysis.

You get five free analyses. After that you have to start paying, and from what I could see, the lowest tier is about 29 bucks, which is right in the neighborhood of what our AI tools commonly run, so I'm kind of used to that. If you get a lot of value out of this tool, maybe that's worth it.

The goal of the tool is to improve your content in terms of AI's interpretation of it. I think that's fine, but I'd probably have to test it to see if it actually improved AI visibility. I love it just as a content tool, it gives you great suggestions on how you can improve your writing. So that's how I look at it: it's not necessarily AI-specific.

The coverage score and the quadrants

Let's start with the coverage score: 70%. So I'm doing most things right here, but there are some opportunities, and that's great. Then you've got four quadrants, with some quick wins and some low priority. It really comes down to about five recommendations that are going to take my content to the next level.

I need to make all of these more explicit. The flags were all things I actually mention in the post:

• A Google Search competitor review in my title tag
• A title tag optimization relating to organic search results
• My primary keyword in the title tag
• AI as a collaborator, with a mention of ChatGPT

So we're seeing a theme here. It's relating two topics, Google Search competitor review and organic search results, so I need to do a better job connecting those concepts.

Suggestions in plain terms

Something I love is how the tool does a great job of explaining everything in plain terms. The tutorial covers keyword selection for SEO, that's my blog post, and it does a good job, but three concepts need stronger integration to complete the picture. The most important gap is closing the loop between keyword selection for SEO and the title tag. Without it, learners finish the process without knowing where the primary keyword actually lands on the page.

Now, this is a cool call-out. It's basically saying I tell people how to choose the keyword for SEO, but then I don't tell them what to do with it. I actually wrote another blog post on that exact topic. So I'm going to take this suggestion and mold it to my scenario. This is just like any other AI output. I'm not simply going to let it tell me what to do, I'm going to hear what it has to say, and then I'm going to use my own discretion and make my own decisions.

So for me, what I'm thinking is, at the end of this how-to-choose-a-keyword post, I'll add a line like, "Hey, the next step is knowing what to do with the keyword." That's going to open up a natural internal link opportunity for someone to hop over to my next blog post. I should have included that, and I didn't.

The next note was that Gemini and ChatGPT are referenced, but they're not given enough individual identity to help learners replicate the AI-as-a-collaborator step confidently. I do mention using AI as a helper in finding the primary keyword for a page, but I don't go into enough detail on exactly how you do that with tools like Gemini in Chrome and ChatGPT. So for me, this sums up to two recommendations. I like them both, and I'd absolutely go ahead and do them.

You can go to each explicit recommendation, these are the five dots from the quadrant visual, click them, and get more information. It tells you exactly what it suggests you do, and it gives you an example of what it means: here's how that writing could look. So if you're having writer's block and need that, it's there.

Connecting concepts

Here's another really cool part: how it tells you to connect concepts. For one, it flagged Google Search competitor review, organic search results, and title tag together. Its comment was to sequence those three concepts in order: explain that the learner runs a search, narrows attention to the organic search results, then reads each result's title tag to identify competitor keywords. Keeping the sequence tight prevents learners from getting lost on the search results page.

The tool also has a reanalyzer. If I kept it active, updated my blog post, gave it the content again, and reanalyzed, hopefully it would A/B the difference for me. I'm not totally sure if it shows a side-by-side or if I just have to record the results in my mind and see if there are simply fewer suggestions after my update. But I'm pretty sure it gives you a comparison and shows you how well you've made the changes.

The coverage matrix

The next section is a coverage matrix, and I really like this: eight questions a large language model needs answered to fully retrieve and cite your content. This is something you might build into your own copywriting workflow, something you could hand to actual human writers. Check out the eight:

• What is it (the concept)?
• How does it work?
• What does it depend on?
• What does it affect and produce?
• Who interacts with it?
• What are the constraints?
• What are the alternatives?
• Can you give an example?

That's a pretty cool framework to put your writing up against and make sure it hits all of these points. There's another readout right after it that does this on a concept-by-concept basis, breaking your writing down on an even more granular level. I've got primary keyword and brainstorming explained quite well, so I get good marks there. Title tag, Gemini, ChatGPT, those AI tools, maybe they're not explained as well and need more detail.

The visualization

Now here's where we come to the visualization, and this is where it maps out all the different concepts and how they relate. You've got some clickable options. By default it's on an observed-only view, which is fine, and you can see some of the things I was a little weaker on: the title tag, organic search results, and the AI tools, Gemini and ChatGPT. It's colored to show what I've covered well and what I've missed.

Then there's a gap view. If something's in red, it's totally missing, so the tool can come up with concepts you didn't cover at all and tell you to consider covering them. Some things are weakly integrated and come up in gray. So green is good, gray is not good, and red is totally missing.

You can also click Show Scope. Core concepts are circled in green. A supportive concept gets circled in yellow. And there's even a blue circle if it belongs elsewhere, and that hearkens back to the feature I mentioned at the start: if there's a concept that meanders away from the topic's center, it suggests you put it on another piece of content. There's also a dashed red for concepts that are excluded or missing. This visual is really helpful, and it pairs nicely with all the other information the tool gives.

And here's that indication of whether I should keep this as one page or split some concepts onto others. This one is telling me to keep it as one page, probably because 70% of it came up written just fine.

Keep your human brain working

The final table is the concept scope. This is another area where you might see concepts that meandered a little, small portions of your writing that could have drifted. Keyword selection for SEO, obviously that's a keeper, it's the whole center of the thing. Brainstorming, AI collaboration, keyword competition, everything's looking good. One it suggested I shorten: "a single clear sentence is sufficient," so it's saying maybe I went into too much detail there.

And then there's one it's telling me to move. I did have a mention of keyword strategy, and that's what the tool is picking up on. But here's what's interesting, and this is why you always have to keep your human brain working with AI. I went back to my blog post, and keyword strategy is only mentioned in the final little call-to-action row. So the tool is acting as if I covered keyword strategy within my writing, but I actually didn't.

But overall, this is a really great tool from contentgrapher.io. It's explicit, it's clear, it's actionable, and it's visual. There are a ton of great suggestions here, and to me, they very clearly would improve the writing. So try this one out and let me know what you think.

In this post, I'll explain what an entity map is, what it helps with, and walk you through how I built one for my own site with Claude Code. You can watch the full live demo below.

The Problem Entity Maps Solve

When an AI search engine answers a question about your brand, it pulls information from across a bunch of different pages. That creates three problems.

Disambiguation. When AI runs into the same concept across multiple pieces of content, it can get confused and output something that isn't quite accurate.

Attribution. Your content might be part of an AI answer, but you aren't clearly identified as the publisher. You're in there somewhere, just tucked away so deeply that people miss it.

Relationships. If someone asks whether your brand does a particular thing or can solve a specific problem, the AI can't always connect those dots. When it can't, it blows the answer.

An entity map aims to solve all three.

So What Is an Entity Map?

It's a couple of files that describe your brand as a knowledge graph: your brand, the concepts it relates to, its services, and the canonical (official, authoritative) information about the entity itself.

There are two of them. An HTML file you link on your site, and a JSON file you reference from the HTML file, from your robots.txt, and in the head section of your pages.

If you've written schema before, it'll feel familiar, since it's also JSON. The difference is that schema works page by page, while an entity map is much more all-encompassing. It's meant to give the canonical picture of your whole brand and everything it relates to. You can read the full proposed protocol at entitymap.org.

Why I Decided to Build One

After the conference, a colleague shared a case study from Waikay titled "We installed an entity map on our site. Here's what happened." The short version: they saw measurable AI visibility improvements, plus better clarity and accuracy in how their brand was described.

In that case study, the entity map was cited more often than the company's own About page across Google's and Perplexity's AI results, and on one topic their AI visibility score jumped 26 points in just 48 hours. It's one case study, not a guarantee. But it's a promising signal, and an entity map is free to create.

How I Built One With Claude Code

The protocol has real documentation behind it, so the most important step is teaching your AI partner before you build anything.

I started a study session with Claude Code. I fed it the key pages from entitymap.org one at a time, the spec, the "why" page, the predicates reference, and the implementation guide, and had it review each before moving on. Some content is hidden behind dropdowns, so I used a text-grabbing tool to capture the full pages.

Then I pointed Claude Code at my own website and asked it to model me and my site as a set of entities. It pulled my live pages, including my services, case studies, and blog, and proposed a knowledge graph: me as a person, my agency as an organization, my services, and the concepts I write about.

From there, Claude Code generated both files. I ran the JSON through the validator on entitymap.org, which works like the XML sitemap validators you may already use, and fixed anything it flagged. There's also a visualizer that renders your entity map as a graph, which is a great way to confirm everything is connected the way you intended.

One Tip for Agencies and Consultants

One of the fields in an entity map is a metric. Make that metric something like ROI or profit, not a leading indicator like traffic or rankings.

If AI ends up referencing your entity map, the bottom-line outcome is what you want potential customers to associate with your brand. When I first built mine, Claude Code set the metric to organic traffic growth, and I had it changed to ROI right away.

Getting It Live

Once the files were ready, Claude Code deployed them to my site. The HTML file is linked in my footer, the JSON file is referenced in my robots.txt, and a small snippet in the head of my pages points to it. You can see the live result on this very site. Just scroll to the footer and click "EntityMap."

When an AI platform gets a query that requires it to retrieve information in real time (a process called RAG, or retrieval-augmented generation), it can find your entity map and pull canonical information that's under your control.

Why This Is Worth Doing Now

Here's the forward-looking part. Over the next one to two years, agentic AI crawlers, systems that act on their own to go find information, are going to become much more common. Giving them a clean, canonical source to consume is a smart way to stay ahead while most brands haven't even heard of this yet.

The Bottom Line

Entity maps are free, brand-new, and very few brands are taking advantage of them yet. There's at least one solid case study behind the idea, and the downside is basically just your time.

If this protocol gets adopted, the brands that tried it early could end up as the case studies that prove whether it works. That's a fun place to be. Give it a shot.

Today I want to talk to you about two protocols that I think have a lot of promise with SEO: entity maps and the Open Knowledge Format from Google. If there are a couple of forward-thinking things you want to focus on to stay on the leading edge, a couple of ideas that have a ton of promise, it's these ones.

I think you should learn about them. You should build them and test them, and see if they have a positive impact. But before we get into those, I want to reflect on a couple of topics that were hot-button issues in the SEO community that turned out to not show any promise just yet.

The ideas that haven't shown promise just yet

The first one is the llms.txt file. Ahrefs analyzed 137,000 websites, and 97% of the llms.txt files in that experiment never got read.^[1]

Now, I don't have any issue with llms.txt files. I think they're pretty lightweight and easy to create, and there's no real downside to it. If you want to try one out, go ahead. But right now, there's really no proven benefit, at least not that I can see. I did see at least one counter study from Cyrus Shepard that showed an llms.txt file was getting hit by AI search platforms. It was getting visited and read, but as of now, no real measurable benefit. I don't see any case studies that are making this something we should all rush out and do.

The other one was markdown files, and to take that idea further, the thing being promoted was to create markdown versions of all your web pages. I saw this in a lot of different places. This has not panned out either. In fact, Google's John Mueller and Martin Splitt were on a podcast, and they said markdown duplicates of your pages give no SEO benefit, and crawlers do just fine with standard HTML.^[2]

Entity maps: what this site knows, how it connects

What I want to talk to you about today is two ideas I think have way more potential. I love the direction these ones are going in. The first one is entity maps. This is a proposed protocol from the folks at Waikay, led by Grant Simmons.^[3]

It's two files, and it charts the concepts on your site and how they relate. It goes beyond concepts, too: it can be people and services, ideas, topic areas, that kind of thing. It's going to map how they all relate to each other, and this is really good to help prevent hallucinations or inaccurate business descriptions by AI platforms.

It can ensure that when your content is cited in an AI output, you are referenced as the brand. A lot of brands were seeing that their content was being used in an output, but there was no mention of the brand anywhere, so it can help with that as well. And of course, it's very helpful in helping AI make relationships between certain things. So if one of your users has a prompt that intersects with two concepts on your website, it's going to help AI make that relationship and increase the likelihood that you're going to be part of that output.

It's not about taking your web pages and duplicating them in a file format. You've got to think about your website from a higher level than that.

Basically, think about it like two files: a JSON file and an HTML file, and they serve pretty much the same function. The HTML file is human-readable. You can see all the entities I worked out from my website and what each one represents. Like, Brian Gorman is a person, he's affiliated with the brand name, and then there's some evidence of that. And the service entities come in and show how all of that relates. By the way, I did an entire walkthrough of how to create an entity map for your site if you want to see the full process.

The Open Knowledge Format: the knowledge itself

Now the Open Knowledge Format is the knowledge itself. It's an open standard from Google, announced on June 12th of this year, and it's a shared format for representing what an organization knows. This is basically where you package what your business knows. This is your knowledge base.^[4]

Try to keep track of the difference between an entity map and OKF, because that difference is important. The way I understand them, the two files can and should coexist, at least for now, because they serve two different purposes. Neither of these is a direct reflection of your website. It's not a duplicate of your web pages; it's a higher-level mapping than that.

We're dealing with markdown files here because it's easy for AI to ingest and interpret. These files are there so AI can access them and work through them as needed. It doesn't read all of them. It's kind of purpose-based, and it'll go to only the most relevant files. The files also have what you'd consider links or references to all the other knowledge files, so there's a kind of internal linking in an OKF setup, which is super interesting.

An analogy to keep the two straight

Think of a marketing agency that does local SEO. Here's how each one would speak:

• On the entity map: "I know about local SEO. Here's proof that I know about it, a little chunk from my website, and how it connects to me and other services, and customers and things like that."
• The OKF version: "Here's what I actually know about local SEO." That's very much evidence-based, and that's the knowledge itself.

And the OKF needs a way to be exposed to AI. One of the ways you can deliver your OKF and expose it to AI search systems is the llms.txt file. Now, I don't want to seem like a hypocrite for creating an llms.txt after questioning it earlier. I'm still not sure of the broader benefit, but this is a legitimate use for the file, and I'm going to test it. You might also be able to link it in your footer, and I'm going to test that too.

Tether them to your site, and keep them in sync

I am tethering these to changes on my website. Any time I make a change, like updating content or posting something new, I check in with that update and ask: is there anything relevant here that should end up as an update to my entity map and my OKF files? I think that's important, because as your site grows or shrinks, you might remove things too, and these files should be updated to represent that. Maybe not every update should change these files, but you should at least check in with it.

The other thing I've got running is keeping the two files in sync, because I want to avoid drift. I never want these to disagree and cause ambiguity, confusion, and thus hallucinations in AI outputs or understanding. That's another mechanism I've got going, and it's one of the big benefits of building these yourself: you're going to think of things like this.

The bottom line

That's what I wanted to talk to you about today: entity maps and OKF. I suggest you learn about them, build them yourself, experiment, and see if there's any positive impact. We need more people doing that to see if it has any legs, at least before it gets accepted as an official protocol, if it ever does.

I like these ideas. I think these have much more meat to them. It's very logical, it just makes a ton of sense, and I think there's a high probability that these do get adopted on a more official level in the future. So it's a chance for you to get ahead.

One quick note: I had to use screenshots for this walkthrough. I'd take you right into HubSpot, but I have to protect client data, and I'm not able to blur things while I'm recording video. You can find the tool in the far left menu. Go to Marketing, and AEO is right at the top.

Most tools stop at the data

Most tools kind of stop at the data portion. You might be able to track some prompts, get some citation data, see your brand visibility, maybe a share-of-voice metric. But the big question with those is: what are you supposed to do with them?

So look at what HubSpot does. First you've got your dashboard, then the prompts you can track, then some citation data. But the final tab is the big one: recommendations. And not only does it recommend things you can do to improve your AI visibility, you can give it the URLs for the content you create so you can track impact. I think it just shows how thoughtful HubSpot has been with this tool. I'm not sponsored, by the way.

Getting set up

When you get started, you're going to sign into your client's account. I think this costs somewhere in the neighborhood of 50 bucks if you want to track 25 prompts. Don't quote me on that, but it's around that price.

Then you're going to set a few things up. You can see the options: the engine, that's going to be ChatGPT, Gemini, Perplexity, and so on. The three I just mentioned are in there by default. You can set your products and services, and HubSpot will have some selections ready for you, plus the option to add some manually.

Then you're going to set your ideal customer profile. Same thing here: HubSpot intelligently makes suggestions, and you get the option to add something manually if you want. One other thing during setup that's really important: you're going to set up all the different variations of how your brand could be searched. That's going to be really helpful for the tool.

Adding your prompts

Next, you're going to add some prompts. Like I said, you can do up to 25, and you can do more if you pay more. You can manually add them or generate them with AI. The AI does a really good job. We've tried this on a number of clients at my agency, and the prompts are great, they're high relevance. I personally usually do a mix: about 80% with the AI, and then I'll add a few of my own for a good prompt set.

Just a note on prompts. My advice is to pick prompts you haven't won yet. Be aspirational here. We've had clients look at this and panic and say, "Wow, we're doing horribly with our brand visibility on AI search engines." But here's the thing: it all depends on the 25 prompts you pick.

The Prompts tab

When you go to the Prompts tab after the dashboard, you can see your 25 prompts. Click on one and you'll go to the right, where you'll see the prompt at the top, then a tab called Answers. That's where you'll see the three engines you selected.

Below that is a little thing that says how many times each prompt has been run. Each one was run on every engine six times, and I really like that, because we know that each time you prompt AI, you get a slightly different output. So it's important to run them multiple times and average it out. HubSpot does that for you. Scroll lower and you can see the actual prompt and the output, so you have a lot of visibility into how you're showing up.

The Citations tab

There's another tab after Prompts called Citations, and it has some cool data. It's fairly self-explanatory. I'll be honest, as much as I've used the tool, I don't go to this tab much. It's good to see the data, but I find the recommendations and measuring impact so much more useful, so I'm always in those tabs. This is the one I frequent the least. One thing at the top of the Citations tab is your top recommendations for the week, which I think is really cool, and that naturally brings us into the Recommendations tab.

The Recommendations tab is the real winner

This is the real winner for this AEO workflow within HubSpot. On the left side you can select a recommendation type: owned content, social recommendations, outreach recommendations. These are all really cool, and they differ a lot from client to client, which I think is really encouraging.

What would be a bummer is if you looked at a bunch of clients and started to see the pattern, where the recommendations got predictable. But honestly, from client to client, they've been varied. Some of them are throwaway. It's still a tool that's kind of automated and doing it on its own, so there's going to be some throwaway stuff, and you have to vet everything carefully. But there are some gold nuggets in here, and that's where the value comes in.

On the All Recommendations tab, you can see the table on the right: the title of the recommendation and what type it is. There's social amplification, new content for your website, outreach, all that.

When you click into an individual recommendation, it gives you the suggested content: a title, a little summary, and who you're targeting.

You can click See More and it'll tell you why it made the recommendation. I really love that. It shows you what's driving it: the prompt, and if you scroll down a little more, what was cited. In this case, it was a recommendation to do a LinkedIn post, because for that prompt, LinkedIn was cited very prominently.

Tracking impact

This is where you give it a URL that you can track. The tool will track it, and you can measure impact. That's super important with SEO. There's a lot of discussion in the AI search and AEO space that is purely theoretical. It sounds logical, it sounds good, but we have a lot of testing to do as a community, and it's this kind of feature that's going to help us really prove the value of these recommendations.

Was it a good one? Should we pursue more? Or did it fall flat? The more you do this, the more insight you gain over time into what works and what simply doesn't.

The full cycle

So the full circle within HubSpot is this: get your setup in place, see some great data, and the tool uses that data to put together a set of recommendations. It makes the data actionable. It's very specific on the recommendations, it's quite varied in terms of content types, and it includes outreach, which I think is fantastic, because that's the first-party and third-party signals you need to be building as a brand. And then the big one: you can track impact to see what works and make these recommendations prove themselves out in the wild.

That's the new HubSpot AEO tool. I encourage you to give it a try if your clients are willing to pay for it. I think it's great for rounding out your strategies and adding to those core SEO initiatives you might have going.

We're using the same example business from last time, Helen's Kitchen Cooking School, and the primary keyword we landed on was "cooking classes in Natick, Massachusetts." I took inspiration from that website and built a couple of mockups of the top of the homepage so we can talk through what to do with the keyword. The primary keyword is the most important one, so it gets prime real estate, which centers around the top of the page.

The Big Four

Here's what you do with your primary keyword for SEO. You do what I call the big four. You put it in the URL, the title tag, the H1 headline, and the first paragraph. When you use it in these four areas, they combine to send a really strong ranking signal. If someone searches for "cooking classes in Natick, MA" and you've used that keyword in all four places, it makes it really easy for Google to match your page with that search.

You won't always land all four, and that's fine. Google will get the point if you use the primary keyword in at least two of these four areas. Let's walk through each one.

1. The URL

The URL is basically the address of the web page, the part you see at the top of Chrome. (Worth mentioning in case you're a beginner.)

Helen doesn't use the keyword in her URL, and in my mockup I don't either. Without introducing an exception too early: you have to use your discretion in every SEO situation. This is a homepage, and her business is called Helen's Kitchen, so the homepage URL is going to be the brand. We can't really sneak a keyword in there, and I wouldn't try. I'd let the homepage URL be the brand.

If it were a deeper page, though, it would work really well. For any page other than the homepage, you can usually work the keyword in, like helenskitchen.com/cooking-classes-natick-mass. So the URL is the first area to consider, but if you're in a scenario where it doesn't fit, don't force it.

2. The Title Tag

The title tag lives in the code of a web page, so you can't see it just by opening the page in Chrome. I want to keep this simple and skip the code, so here's an easy way to see it: in any browser, hover over the tab and wait, and the title tag pops up. For the real site it reads "Helen's Kitchen Cooking School," with no keyword.

In my mockup, I'd make the title tag "Cooking Classes in Natick, MA | Helen's Kitchen" (that vertical bar is called a pipe). That gets our keyword into the first of the big four, and I recommend front-loading it so the primary keyword is the first part of the title tag. If you're not sure how to update it, you don't necessarily need to touch code, platforms like WordPress let you edit it in the back end. (I'll cover that in another video.)

One more thing about title tags: when you search Google, the title tag is exactly what shows up as that clickable blue headline. So if someone searches "cooking classes Natick, MA" and sees those words right there in your title, they think, "perfect match." It's not all about Google, we want people to see that match too. That's why the title tag is such an important place for your primary keyword, and such a loud ranking signal.

3. The H1 Headline

The H1 is your primary heading on the page. Helen's current heading has no keyword in it. In version one of my mockup, I set the primary keyword as the H1, which sends a really powerful signal.

Here's where discretion comes in again. Maybe the client doesn't love that, her current H1 is something like "Welcome to Helen's Kitchen," and she may really want to welcome people warmly. So we might compromise. In a second mockup, I put the primary keyword in an eyebrow (the small line above the H1) and then use warmer, friendlier copywriting as the actual H1. That's good technique: you get the keyword signal in, while working with the client to accommodate what they want.

So go for the keyword in the H1 if you can, but if it reads unnaturally, the eyebrow-plus-H1 setup is a great option.

4. The First Paragraph

The fourth area is the first real paragraph of the page. You might ask, "isn't that the text right under the H1?" Arguably, yes. But I want to teach you to build pages tastefully. The goal is to use the primary keyword, true, but we don't want the page to turn spammy or sound unnatural. We're not stuffing keywords everywhere and ruining the experience.

So rather than force it into the very first line, I'd often place it a little lower in the opening content, where it fits naturally. The keyword still lands in the first paragraph of real copy, it just reads like something a human would actually write.

Be Flexible

That's the big four, and that's how you use a primary keyword for SEO:

• URL - use it if you can, which mostly applies to deeper pages; on a homepage the URL often just has to be the brand, and that's okay.
• Title tag - definitely try to get it in, and front-load it.
• H1 - try for it, or use the eyebrow-plus-H1 compromise if it reads unnaturally.
• First paragraph - work it in tastefully.

If you can get all four, it's extremely powerful. If you have to compromise on one or partially compromise on another, that can be okay too. It's fine to be flexible, Google will get the point if you use the primary keyword in at least two of these four areas.

What's Next

So now we know how to choose a primary keyword and how to use it. In the next video, we'll talk about secondary keywords: how to choose them and how to use them. If you have any questions, or you're in a scenario where you're not sure how to get the primary keyword into one or more of the big four, leave me a comment and we'll figure it out together.

If you've been following along, we've been using Helen's Kitchen Cooking School in Natick, MA. It's a real business that we're using as our example, and our primary keyword was "cooking classes in Natick, MA." For this secondary keyword exercise, we can just drop the location and think about the "cooking classes" part.

What Are Secondary Keywords?

If "cooking classes" is our primary keyword, what are the variations we could build off of it? Here are some: cooking school, culinary school, cooking training, cooking lessons, cooking experience, classes for cooking. There are all kinds of variations that could exist, and it's a good idea to use them.

Remember the big four from last time: you use the primary keyword in the URL, the title tag, the H1, and the first paragraph. Once you've done that, you've pretty well established the primary keyword. Then it's time to start using variations, so that Google can match your page to all the different ways people might search for what you offer.

Where to Use Secondary Keywords

I suggest using them in the second half of the title tag, in subheadings, and in body paragraphs. There are other areas where you can use keywords too, like images, but I want to keep today's lesson simple. We'll cover images and other fields in the future.

How to Come Up With Them

This workflow looks a lot like the one for finding a primary keyword:

• Brainstorm variations on your own. Look at your primary keyword and think of the different ways people might search for it.
• Work with an AI. I love using Gemini in Chrome because it's so convenient, but Claude or ChatGPT work too. Give it the full context: here's the page I'm working on, here's some info about it, here's my primary keyword, and now I need some secondary variations. You'll get new ideas and confirm some of your own.
• Use Google's Keyword Planner. It's free to sign up for (if you have trouble, search "how to sign up for Google Keyword Planner free").

In Keyword Planner, go to Tools, then Keyword Planner, and this time choose Discover new keywords. Type in your primary keyword, "cooking class" or "cooking classes," it doesn't really matter. Notice I dropped the locality and didn't include Natick, MA. That's because of the tool's own guidance: try not to be too specific or too general. "Cooking class" is enough to help the tool understand what I want variations of. I'll also drop in Helen's URL, then click Get results.

Now I can see my primary keyword plus a list of keyword ideas. "Cooking school" is a fantastic variation. "Kitchen cooking," not so much, at least in my opinion, and this is where human discretion comes in. Don't just accept everything a tool gives you.

"Culinary school" is a pretty good one, but I wasn't sure at first. I thought it might surface more established schools or colleges, so I verified it: I searched "culinary school Natick, MA" in Google to make sure it's relevant to Helen's business. Sure enough, she shows up number one in the map pack and number one in the organic results, so it's a no-brainer, I'll use it. That's basically what I did, going down the line and vetting each keyword. You can also check the competition column and make sure everything is medium or low. As we decided in a previous lesson, we steer clear of high-competition keywords unless you're a well-established brand.

Two Ways to Use Them: Full or Partial

So now we have all these great keyword ideas. What do we do with them? Like I said, use them in the title tag, subheadings, or body paragraphs. But here's a cool technique.

One way is to use the full keyword, obviously, like "cooking school." Another is to use just part of the keyword, so you don't sound too repetitive. Say your title tag already reads "Cooking Classes in Natick, MA," and then you come in with "cooking training." It doesn't read all that well. So I'd drop the "cooking" and just use "training." We've got plenty of great words for that: training, lessons, experience. Since it's already obvious the page is about cooking classes, the keyword is almost implied, so you don't have to repeat the whole thing. If you did, the page would turn out spammy and wouldn't read well.

I want to teach you to be a tasteful SEO who creates content that reads well, so know that this technique is available. Here's it applied in the title tag: "Cooking Classes in Natick, MA - In-Person Lessons and Training." The primary keyword leads, and "lessons" and "training" are partial secondary keywords worked in naturally. That's a great title tag.

Seeing It on the Page

Let's see things in action on the page. We have the title tag above, the primary keyword in the eyebrow (from last lesson), and some great audience copy in the primary heading. So where do the secondary keywords come in?

• First paragraph: the primary keyword, which is part of our big four.
• Subheading: "More than a cooking school," using the variation a little deeper on the page.
• Body paragraph: "Helen's began as a small culinary school and grew into..." so we naturally fit in another variation.
• Echo of the title tag: "in-person lessons and real training," using the partial secondary keywords in an organic way.
• Another subheading: "cooking experience" and "classes for cooking."

You get the idea. We're getting these in well, and it all still reads naturally.

Final Thoughts

That's your lesson on secondary keywords, the variations on your primary keyword. I've given you three ways to come up with ideas (brainstorm, AI, Keyword Planner), three places to use them (title tag, subheadings, body paragraphs), and two techniques for using them: the full keyword, or just the partial keyword when you want to be a little more tasteful.

If you have any questions, leave me a comment and I'd be happy to help. I'll see you in the next lesson.

It's a fundamental, critical skill for SEO, and we're going to keep it simple. This is one of those topics where it can start to get complicated and you can tie yourself up in knots overthinking it. The best approach is to keep it super simple, so here are a few steps to help you choose the right keyword for your web page.

For this example, we're using Helen's Kitchen Cooking School. This isn't a client of mine, I just randomly selected it.

Step 1: Brainstorm on Your Own

Start by reviewing the content of the page you're working on. If you're building the page, you'll be quite familiar with it. But if it's a client, say you're a new SEO who just started at an agency, this might be the first time you're really taking in what the website is all about. So take a read.

Obviously, we're looking at a cooking school. In the main navigation, I can see "classes" as one of the central items, so it's probably cooking classes at a cooking school. I'm starting to get an idea of what this page boils down to. What's the main topic? That's really what I'm after.

Then I look at this sentence, which kind of says it all: "We offer hands-on cooking and baking classes to enthusiastic home cooks at our kitchen in Natick, MA." To me, the keyword, which should be a short phrase that my target audience would search, is jumping out: cooking classes in Natick, MA. I think that's the perfect keyword for this page, which is the homepage of the cooking school.

I can keep reading and I'll see more specific things like handmade pasta and sashimi, which hint at the different classes and foods you can work with. I might even learn a little about Helen herself. But it all boils down to cooking classes in Natick, MA. That's my first step: my own brainstorming, and I'm happy with that keyword. It makes sense and it's super relevant to the page.

Step 2: Tap Into AI

For step two, tap into AI. We've got this amazing tool now that we didn't have before. I wish I had it early in my SEO career, and we should definitely use it.

I like to use Gemini in Chrome for this, because it can see the web page I'm working on. All I have to do is open it in the upper right of my Chrome browser and ask: "What do you think is the primary keyword for this page?" I'm using AI as a collaborator, a partner in the brainstorming exercise. It gives me a second opinion, and hopefully it confirms what I landed on.

Here's what it said: based on the page content and the headings, the primary keyword is "cooking school" or the slightly more localized "cooking classes in Natick, MA." I don't like the first choice, "cooking school." It's way too general. In general, you want to be as specific as possible, especially for a smaller organization, and this looks like a small, private operation, not a Sur la Table.

More specific tends to be less competitive, and the less competitive terms are normally what you want to go after, unless you want to fight it out with the giants. The second choice, "cooking classes in Natick, MA," is exactly what I landed on with my own independent review. So this is really confirming. We're already on our way to choosing the right primary keyword.

Step 3: Check Google

The third thing you can do is take one or two ideas from your brainstorming and search them in Google. What you're looking for are pages that rank on page one and are similar to your own organization.

I already searched "cooking classes Natick MA." What's kind of funny is that Helen's cooking school already ranks number one, which tells me Google agrees with the keyword we chose. It thinks it's a great match.

I'll also look at the other results. I try to find ones similar to my business. Yelp and TripAdvisor aren't similar to Helen's cooking school, so they're not too useful here. Sur la Table is a big organization, but you can see they chose the exact same keyword we were targeting. I tend to focus on the organic results rather than the sponsored ones, but you can still get good ideas from sponsored listings.

Look at the title tags from the competitors to see whether they chose the keyword you thought of or something slightly different. That can give you more ideas and help you finalize your primary keyword. For example, one sponsored result from Cozymeal used the word "in-person." That's interesting, and it might be a way to make our title tag even more specific: "in-person cooking classes in Natick, MA." The more specific you get, the less competitive it tends to be, which gives you a better chance of ranking well and actually getting the click.

What If You Don't Have a Finished Page?

You might not have a finished web page to look at. Here are a couple of other scenarios:

You have a copywriting draft. Say you're a designer and you haven't built the page yet, but you have the copy. Read through it yourself and do your brainstorming, then drop it into an AI tool like ChatGPT or Claude and ask what it thinks the primary keyword is.

You don't even have a draft. Brainstorm on your own first: what page am I about to create, what's it all about, what does the topic boil down to? Then describe it to an AI tool in detail, the same way you'd think about it. For example: "I'm soon going to build a page about cooking classes led by a woman named Helen, who has experience from Europe and cities all over America, and she's based in Natick, MA." The AI can pick out the key details, boil them down, and give you an idea. It's a nice parachute for when you don't have a page to look at.

Validate It With Free Search Volume Data

Let me point you to another tool: Google Ads. Search for "Google Ads signup," go through the free signup process, and you'll land in the interface. On the left, go to Tools, then Planning, then Keyword Planner, and choose "Get search volume and forecasts." Then enter your keyword ideas.

When I was doing this exercise, I came across two ideas. One was "cooking classes Natick MA." The other was "cooking classes Boston MA," a suggestion I actually got from AI, which pointed out that Boston is a bigger area with potentially wider interest, so it might help attract more customers. Interesting idea.

But watch this. The average monthly searches for "cooking classes Boston MA" are way higher than the Natick variant, so a lot more people are searching for it. AI was on the right track there. As you'd guess, though, look at the competition column: the competition for "cooking classes Boston MA" is high, while Natick is medium. Helen's Kitchen seems like a smaller organization, basically one individual, so I'd go after the lower-competition term. Once again, that overconfirms my choice of "cooking classes in Natick, MA" as the primary keyword for the homepage.

Don't Forget the Title Tag

When you look at Helen's result in Google, that clickable headline is called a title tag, and it's really critical for SEO. When you have your keyword in the title tag, it's a very strong signal to Google that your page is a good match for anyone searching that keyword.

So even though Helen already ranks number one, putting "cooking classes in Natick, MA" in the title tag, followed by something like "Helen's Kitchen," would help lock her in place above Sur la Table and the other larger entities. (I might do a separate video on how to find the title tag in the code and update it.)

Final Thoughts

Those are the three steps: brainstorm on your own, use AI as a partner, and do a Google search to see what competitors are choosing for their primary keyword. Add in a free tool like Google Ads Keyword Planner for real data, and you can make a great choice on the primary keyword for any page you're working on.

Sometimes it's tricky to choose, because it isn't always this straightforward. If you're in one of those scenarios, leave me a comment and we can talk through it together.

What Are Skills in Gemini in Chrome?

If you open Gemini in Chrome and click the plus sign, you'll see an option to browse skills. That's the whole point here: you can create skills for Gemini in Chrome, and it's incredibly useful if you're used to skills in Claude.

These are not as detailed or powerful. Claude skills can do a lot. They can deploy multiple agents that report back up to a manager or orchestration agent, they can crawl the web, and there's so much you can do with them. The skills in Gemini in Chrome seem more limited. Basically, it's saving prompts that you already have. But that's exactly what makes them handy. There are a couple of workflows I typically do with Gemini in Chrome that were perfect for these skills. I can save them, call them up quickly, and run them really easily.

How to Create a Skill

Creating one is simple. You go to the skills section, click your skills, click add, name your skill, and then drop your instructions or your prompt in. That's it. I've already done this for a couple of mine.

The first is a content comparison skill. It compares the content of multiple pages and shares the most effective elements, like outline sections and multimedia. I put right in the instructions that I want to learn why the content might rank well and get strong engagement, because typically I run this on top ranking pages to try to figure out why they're ranking. It's at least another perspective.

The second is a design comparison skill. This one also runs on multiple pages, and it looks at things like UX and CRO and tells me what the most effective elements of those pages are.

Demo 1: Comparing Design Across Pages

Let's run the design one first. I've got three websites here: wise.com, Miro, and Airbnb. These just showed up on one of those sites that aggregates the best-designed websites.

I come down and click the plus, and I make sure that Miro and Airbnb are also selected so Gemini in Chrome can see them. With them selected, I can choose my skill from the menu, or I can summon it with a forward slash, and there it is. I call it up, run it, and in seconds I have the insights I'm looking for. It starts with Wise and works its way through Miro and Airbnb.

Now, this is AI output, so some of it is going to be throwaway, but some of it is really good. It talks about the immediacy of the calculator that Wise puts in front of users, which is super cool and kind of just screams engagement. It calls out the social proof, and there are all kinds of other insights from the other pages as well.

Demo 2: Comparing Content Across Pages

Next, let's run the content comparison workflow on a couple of Yellowstone National Park guides. This skill is going to tell me what it is about the content that helps it rank well and get good engagement. I'm assuming it gets good engagement because it ranks well, since we know engagement is part of ranking well. I can't confirm that through their internal data, but it's a reasonable assumption, and that's what I want the AI to look for.

I open Gemini, make sure both pages are selected (my National Park Lodges page and the other one), choose the content comparison skill, and run it. Gemini in Chrome makes it so easy, since it's all right there in the same pop-up.

It looks at structural elements and content outlines first. One page focuses heavily on actionable tips, while the other employs an exhaustive, geographic-first approach. Then it gets into multimedia: the Lodges page uses polished, professional imagery, authentic storytelling, and even the power of captions, which is cool. Finally, it covers the trust-building and authority signals on the pages, which are really impactful for readers.

A Third Use Case: Comparing AI Search Studies

There's one more use case I want to share, though I won't demo it here. I use it a lot, especially with the number of AI search studies coming out these days. I'll open up multiple studies and run a skill that compares and contrasts them. It tells me where there's consensus, where there's disagreement, and it draws out insights that can only be gathered from reading all of the articles together.

Final Thoughts

All three of these use cases are fantastic if you're an SEO or a marketer. If you hadn't seen that Gemini in Chrome has this Skills feature, I hope this was useful, and I hope you'll play around with it. And if you have another use case, let me know in the comments. I'm sure other people would like to see it as well.

Lesson 1: Thin Content Does Not Perform (Google Panda Update)

The first one of these tectonic shifts for me was Google's Panda update, which really focused on content quality. What you're looking at on the right side here is an actual client page that I worked on, and this is all the content that's on the page. It's barely 100 words.

My big takeaway from this time period is thin content does not perform. It certainly did not perform well after Panda. I know a lot of clients are averse to having really wordy pages because they feel like their site visitors don't want a reading assignment. But if the content's really good and it's what people are looking for, they really do want to consume quite a bit of it. Sometimes they need that before they convert.

If you want to be a strong marketer, a powerful partner, or a collaborative client, it's time to embrace robustly informative pages. This could be a home page, a service page, or a blog post. Any one of those really does warrant a strong outline and deeply informative sections. It's going to be good for your customers, and it's going to be great for your SEO.

Lesson 2: Easy Links Tend Not to Help (Google Penguin Update)

The next update I wanted to talk about was the Penguin update. This one focused on backlinks, particularly low quality backlinks. This screenshot right here is just a snippet from some of the backlink removal work I did at the beginning of my career. Basically, what it represents is a large network of sites owned by a single individual who used to sell link placements to help sites rank well. Back in the day, you could rank number one if you bought enough links.

When this update came, it was pretty much overnight that sites participating in this were just knocked off the first page completely. There was a panic to remove them all. Unfortunately, individuals like this made a lot of money not only selling the links in the first place, but charging to remove them when everyone panicked and tried to get them scraped from their backlink profile.

The takeaway here was that easy links tend not to help. Essentially, any link worth having was going to be more difficult to get, if not really hard. And that's kind of the way it should be. Anything worth having shouldn't really be easy to get.

There are still links that are fairly easy to get, and some of them are pay-to-play. If you're a local site, you might purchase a link on a chamber of commerce site or something like that. But this was a completely different thing. This was just spam, and I'm glad Google made this change.

The takeaway for marketers and clients is that we really want to get authoritative, relevant, high-quality backlinks into our profile.

Lesson 3: Keyword Stuffing Is Dead (Google Hummingbird Update)

The next update that was significant for me was the Hummingbird update. This was Google's way of getting more sophisticated in applying context to a user's query. But it kind of extended for me to a philosophical change where we were once focused on keywords and using exact match keywords in the copy of our content, using them a lot—even keyword stuffing, if that's a term you're familiar with.

After this update happened, it really helped a lot of marketers mature into the next phase of optimizing content, which is to make sure that content is meeting the user's search intent. Personally, I not only like to meet their intent—I like to anticipate it.

If I'm creating a page outline, I'll do audience research into things like People Also Ask questions, and I will try to make sure that the subsections of my page anticipate the next thing the user would be wondering about. What's the next most likely thing they might want to read? How can I draw them into the next section of the page to keep them engaged, keep them on the page, and develop a good relationship between that user and the brand?

This is also a period of time where I started utilizing more natural language processing, like keywords and entities, and incorporating those kinds of things into my content. It wasn't just relevant for the user, but it could be relevant for search engines as well.

Big takeaway here: keyword stuffing is dead. That didn't make keywords unimportant. They're still important, and you can still win with exact match keywords. But the emphasis has been greatly reduced. In fact, when I produce content now, I just do audience research and write first. Then I use keywords kind of like a final treatment just to make sure that we've got those signals somewhere on the page.

Lesson 4: Always Lead with Expertise (EEAT Updates)

The next tectonic shift had to do with a couple of Google updates like the Helpful Content update and the Medic update. I'm not sure if those were the official names—I can't remember anymore. But this was all about EEAT: Experience, Expertise, Authoritativeness, and Trustworthiness. Getting those kinds of signals into your content and onto your website, along with things like proper author bios, credentials, and professional titles.

What I started to do in this era was source my content from the experts first. Make sure I'm collaborating with the client. Make sure I'm getting in front of the experts and having them source the content and the writing that's going to go on the website.

The results were far better, and it got me further away from that sort of "set it and forget it" marketing where some intern from an agency is writing content about plumbing in a certain area when they don't have that expertise and they haven't lived in that area.

If you want to write great content, you need to lead with expertise.

Lesson 5: Amplify Your Signal Across Platforms (LLM & AI Search Era)

The final lesson is one that a lot of marketers have neglected for a while because we didn't really have to think too much about it, but it's becoming more and more essential. If you're someone who can crack this code and figure it out—because it is high effort—you are going to outperform the competition.

Your footprint is going to be much larger if you've got a website, you've got mentions on the web, plus a presence in whatever platforms are relevant to your business.

That's why I've incorporated a workflow that allows me to interview an expert for just 30 minutes every month and then utilize and repurpose that information across a number of platforms. Really powerful. It helps a lot with your presence in LLM search, like ChatGPT web search. Honestly, it's just downright fun. Once you start doing it, the old way of just focusing on the website kind of feels stale.

How It Works

The process is pretty straightforward. A user can ask ChatGPT to help them find a great housewarming gift, providing details like "handmade, ceramic, under $100." ChatGPT surfaces a number of products, and when you click into one of these, a mini session opens up on the right side. You're still within the ChatGPT session and window.

After you select the option—and by the way, you'll have a dropdown with a bunch of product options here based on what's set up in the store—you get a "Buy" button. You click this and go to checkout. This is, again, happening all within ChatGPT. Then you can click, in this instance, "Pay Etsy," and the checkout and purchase is completed. It says it right there: "Etsy confirmed your order."

Early Days, But Promising

This is a new experience. I've heard mixed reviews about it, but we're just in the first iteration, and it's very likely to improve. So it's good for e-commerce sites to be aware of this.

For now, this is already working for Etsy shops, which is great—you don't really have to do anything. But it is coming soon to Shopify, and according to OpenAI, other platforms. I'm not sure what those platforms are going to be, but for Etsy and Shopify, you probably don't have to do anything.

How to Get Your Store Connected

If you're an e-commerce site or merchant and you aren't using one of those platforms, you'll probably want to think about this. OpenAI released information on chatgpt.com/merchants. When you scroll down the page, you'll see a form. This is for e-commerce websites that aren't on Etsy, Shopify, or one of the soon-to-be-released platforms.

Go ahead and fill out this form, and OpenAI is going to respond to you. They're going to work with folks on a rolling basis to get your site hooked into this shopping and checkout experience.

The Technical Details

One of the key elements of this is going to be the product feed spec. Now, if you've set up Google Merchant Center before, or you've had it set up for you before, you're probably going to be okay because this looks really similar.

For the product feed, they give you a number of file format options—CSV is one of them, so that's pretty familiar and convenient. What you'll put in that product feed are a bunch of attributes and details, just like Merchant Center: product ID number, product title, product description—that sort of thing.

You can work with your developer or your agency to get this set up. My recommendation would be to just fill out this form and get that ball rolling, because we're not sure of the response time from OpenAI, and you definitely want to get hooked into this.

Why This Matters

From what I've seen, referral traffic from large language models like ChatGPT isn't a large percent of website traffic, but it does convert really well. That's because users can work with the large language model to do a lot of research, build trust, and get much closer to a purchase decision in a much shorter amount of time. It's really convenient.

E-commerce sites might see that here too. It might not account for a huge portion of their sales, but the conversion rates might be quite high. So I wouldn't miss out. Get yourself signed up and acquainted with this shopping experience. I think it's going to improve, and it's something you don't want to miss out on.

What This Looks Like in Action

Here we have a search for "how to get oil stains out of clothes Reddit." You can see in a couple of examples, it's got the Gemini symbol, which is Google's AI or large language model, and it is basically summarizing the thread with AI.

For now, we're only seeing this test on Reddit threads.

Google Has Been Rewriting Meta Descriptions for a Long Time

It's important to note that Google has been rewriting meta descriptions for quite some time. And currently, it still does. Remember, this AI thing is still just a test.

In a Search Engine Land article, it states Google rewrites somewhere between 60% and 70% of all meta descriptions, depending on what study you look at.

The Big Takeaway

The big takeaway here with what Google currently does and what it's testing with AI is that it's just not a worthwhile activity to manually write meta descriptions anymore.

In fact, it's not even super efficient to write them with AI. Google is mostly rewriting them, and it's looking to incorporate AI into rewriting them as well.

Data-Backed Support

I've got a little support for this. Mark Williams Cook is basing his thoughts on this on data and testing. You can see right here: we don't recommend writing meta descriptions anymore.

So in both cases, he feels like it's not a worthwhile activity. There's plenty of other things you can be focusing on, and that's our agency stance as well.

Watch Out for Outdated SEO Audits

So watch out for this sort of thing, and make sure the initiatives in your SEO strategy make sense.

This is a great solution for that. What you do is just take the URL of the thread you're looking at, add a little extension to the URL, open it up, and it's JSON. You have the original post, all the comments, all the replies—everything is right there for you to access.

The Simple Trick

Here's how to do this:

First, access the thread that you're interested in. In this example, I'm looking up how to learn guitar to see what people are saying about it. One thing I definitely like to do is choose a thread that has a lot of answers. I want a lot of material to work with in this exercise.

Now, here's where the trick comes in. When you visit the thread, it's going to produce a URL. Just grab that URL and add .json to the end of it.

That's the trick. And what that is going to produce is a mess of information—but this is all the information from the thread. It's the original post and all the comments and replies.

Understanding the JSON Structure

Now let's dig into the code just a little bit. When I first look at this, I like to get my bearings on what I'm looking at.

The selftext key has a value of the original post. For example: "I just got my first guitar a few days ago, but I don't know where to start." If you go back to the Reddit thread, you'll see that is the original post.

If we go a little bit lower, you'll see things like body and body_HTML. The body key is going to be one of the comments below the original post. You'll see this repeated in a key called body_HTML—it's just a repeat of the same comment.

What's really useful is the key called ups. Ups is Reddit's upvotes, and there's a specific reason I like this so much.

Three Uses of Reddit JSON

Grabbing information from a Reddit thread, there are three uses I want to talk about:

1. Discovering Questions and Pain Points

I love getting into these threads and seeing what questions people are asking. The original post is often a question, so that's gold right there. But you'll see people talking in the comments and asking follow-up questions. They share things that they struggle with. That kind of stuff I love for learning about an audience and coming up with potential content ideas.

2. Finding the Most Upvoted Comments

This is where those upvotes come in. You can just hit Command+A (on a Mac) and select everything and copy it. Then you can hop over to your favorite AI tool like ChatGPT or Claude. You can drop this information in. I use Claude and I like to turn on extended thinking and use the Opus model because it's better for complex tasks, and there can be a lot of JSON here to sift through.

What I'll do is I will ask it to find the comments that have the most upvotes. Why am I doing that? Well, I want to see what those comments are because people are clearly finding them useful. A lot of times what you'll see in the most upvoted comments is advice that people really value, and I think this is a great thing for brands to be aware of.

Often you can use that to guide some of your content initiatives. Maybe we're thinking of the next social post that we're going to do, or we think of a blog post we wrote on this topic but we didn't include a piece of advice. The insights we get from looking at these comments could become more effective social posts or a great update we do in a blog post refresh.

3. Identifying the Most Engaged Comments

There's one more thing I like to do with this information, and it's a little bit more complex. You might want to explore different solutions for it, but you can get away with using an LLM for this. Once again, I use Claude with extended thinking on and the most powerful model.

What I'm going to ask it to do is find the comment that has the most replies. There's no key and value for this in the JSON—I wish there was. So the tool really just has to count this manually, and it doesn't do it perfectly every time. But I would suggest that you at least try to do this and always bounce back to the thread itself to verify the accuracy of what the LLM has returned to you.

I'll ask it to find comments that have the most replies. Why am I doing this? Because that can be a signal of engagement. If I find a comment that gets a lot of replies or a lot of engagement, again, I want to dive into that a little deeper and investigate: what is it about this comment that's getting people to reply so much and engage so much? There might be an insight there that we can apply to our own social media posts or content.

Summary

That's how you turn Reddit threads into JSON. That's how you get all the expanded comments and replies without having to manually click them and then copy paste. And that's three ways you can use this information to inform content and learn about your audience.

I hope you found that useful!

Getting Started

First, sign into Google Analytics and go to Reports. We're going to look at Traffic Acquisition.

When you're in here, the default view is going to show you all kinds of traffic sources like organic search and referral. We would like to see traffic from what I'll call organic AI. There are actually a couple of ways you can set this up, but this is the way that we found works the best. It gives you the most accurate data.

Creating Your Custom Report

Go over to the right and click Custom Report. Then we're going to go down and Add a Filter.

We're going to select the dimension and search for session source. Go ahead and select that.

Now we get to the match type. The match type I like to use is matches partial regex.

Setting Up the Regex Filter

Don't be scared of this. There are some characters we're going to use here that make it look complicated, but it's actually not all that complicated.

Here's where we start listing AI traffic sources like ChatGPT. The next one I want to list is Perplexity. But before I do that, I'm going to hold shift and press this vertical pipe character |. It might be a little hard to see, but it should be over by your delete key.

Now I can type in perplexity. I'm going to put in another vertical pipe. These pipes work like commas in regex.

The next AI traffic source is going to be Claude. Then I'll put another pipe. I'm going to finish with Gemini and Copilot from Microsoft.

Now I've essentially got a list of AI traffic sources and I'm done. Go ahead and Apply this.

Verifying Your Data

I'm going to scroll down and just confirm that it's bringing in the right thing. This is very commonly what you're going to see: Unassigned traffic and Referral traffic. That's really the big one we want to see, because traffic from these AI sources should basically count as referral traffic. It's a referral from ChatGPT if they find you there and then they visit your site.

Adding Session Source Medium

Before we save this, there's something else we're going to do. Come over here to this little plus sign and we're going to add another column to this report.

Select session source medium. You'll see that this is going to add a little more information to our table that helps us see exactly where our traffic is coming from.

Saving Your Report

Now we can go up to Save and we're going to save this as a new report. You can call this whatever you want. I'm just going to name this Organic AI Traffic. Go ahead and save that.

Let's click Back to go back to our previous view.

Now that we've saved this, we don't have to keep building this every time we sign into Google Analytics. That's great. But what would be really great is if we could click over here and see it within this acquisition menu.

Adding the Report to Your Navigation

We've got the Traffic Acquisition report here, but there's no AI traffic view. It would be very convenient if we could have that, and we can.

All you have to do is go down here to Library. If you notice, this report is under the heading Life Cycle.

I go over here in my library and I just find that group right here: Life Cycle. Go ahead and click the three dot dropdown and hit Edit.

We're going to search for what we called the report, which was Organic AI Traffic. You can see it shows up right here.

What we'll do is grab this and just drag it over here. I like to put it under Traffic Acquisition. Let me scroll down so you can see that better. We're going to have our Life Cycle report, then our Traffic Acquisition, and then our Organic AI Traffic right underneath that.

Go ahead and Save it. This one I'm going to save changes to current collection. I don't want this to be a new collection.

Viewing Your AI Traffic Report

Let's go ahead and hit Back. Then let's go under Life Cycle > Acquisition. You can see right here: Organic AI Traffic.

Let me go ahead and click that. You can see here that my filtering is already on. If I scroll down, I can not only see what type of traffic it is, but I can also see exactly where it came from because I set up those two columns earlier.

Conclusion

This is a great thing to have in Google Analytics 4 so you can follow the story about the impact AI traffic is having on your overall traffic.