Building an AI Search Strategy, Part 1: The Infrastructure

Michael 0:22 – 1:18

Hey, welcome back to The Search Signal. I'm Michael Transon. I am both the founder and CEO of a search agency called Victorious. And here at The Search Signal, our goal is really to help marketers understand and take action on what is happening in the ever-evolving space of search marketing. So we've been running search campaigns here for about 13 years, and we also work on hundreds of websites every single year for some of the biggest and also some of the fastest growing brands in the world. And we bring what we learn from those experiences into this podcast, along with a lot of our own first party data and also our first party research. And the goal is to help you with your own work in the world of search marketing. So this week we are kicking off something really exciting. We're going to be starting a four-part series on how to build an AI search strategy from scratch. But before we get into today's episode and jump into the first part of the series, let's talk about what happened this week in the world of search marketing news.

Michael 1:18 – 1:45

Okay, kicking things off, Microsoft, if you're not familiar, has a free user behavior analytics tool that they call Clarity. And this week they added a new card to its bot analytics dashboard, and they're calling it the AI scrape to referral ratio. So, what this essentially does is it compares how often an AI crawler comes and scrapes your site against how much referral traffic that same AI tool sends back.

Michael 1:45 – 3:38

And then it also gives a ranked list of these tools. So you can see which one of them send the most traffic, and then which one of them might be scraping the website very heavily, but then not sending much traffic back to you. So also this does link straight from that ratio into session recordings, and they filter it down to AI sources. So what that means is you can actually watch what users do from these AI referrals once they land on your website. And this is going to be like whether they scroll through the site, if they're clicking around or converting, or ultimately bouncing from the site. So the reason why I like this one is that most teams that I and that we talk to are a lot of them are making these like block or allow bot decisions, specifically on AI bots, based on like a general feeling of whether the AI traffic that they're getting justifies giving AI access to their content. And sometimes this is proprietary content. And when that happens, there's oftentimes legal teams there that are, you know, nervous about giving access to that information. So what's good about this is it gives us a per AI bot answer for our own website to make those types of decisions. So what I would do is you know, if you're in Clarity and you have access to that tool, definitely pull it up as you're getting ready for your next conversation, or if you're in the middle of making a decision about block botting, because this decision is about to get a lot more consequential. We're going to get into why later on in this episode. But again, you do need to have a Clarity account to be able to see this information. But if you're not on that platform, I would generally probably expect this type of report to start showing up in some other competing tools that you might have in your tech stack.

Michael 3:38 – 4:06

Soon. And moving on from Clarity is the first part of this week's news. Let's go to e-commerce specifically. Let's go to Shopify. So they published their Q2 commerce data this week. And that data revealed that AI referred sessions to Shopify's merchant stores grew this year about 200% year over year. That is a huge increase. Organic search also grew. They grew about 12%.

Michael 4:06 – 5:55

But that 12% growth was on top of a much larger base of traffic. And on top of that, it still sent much more traffic, way more traffic than every one of those other AI platforms that they tracked combined. So point being is both are growing, and organic is definitely still the bigger channel of the two. I think though, the conversion data in this report is more useful. And I think the more interesting part as it relates to how you might run your company if you are in e-commerce. So, what they found was when AI referred shoppers made it to a product page, they converted at about 80% better than shoppers coming from organic search. And in the spec heavy categories, and these would be like ones where people are comparing all the specifications of a product and maybe some compatibility issues and reading the reviews before they ultimately go through buying, they converted at roughly double the rate. Also, half of the AI referred sessions landed directly on a product detail page rather than a home page or a category page. So, point being in all of this is AI is sending fewer visits, but we can infer that they are arriving later in the buying process. And also these are people who have already been doing quite a bit of comparison inside of the AI conversation itself. And I would say the thing that we want to act on here, the report also showed that shoppers who came in through structured Shopify catalog data. And this is like classic structured data, machine readable titles, machine readable prices, descriptions, how much stock is available.

Michael 5:55 – 6:50

Those users converted about twice as well as the ones that were arriving through scraped or third-party feeds. So again, this is classic structured data work. Also, I would you know want to point out that Google's own guidance also says that AI overviews and AI mode run on the same core ranking system as regular search. So the same structured data is going to help us on both of the surfaces. So if you are running an e-commerce site, whether it is on Shopify or whether it is on a different CMS, this is probably the clearest data we have yet about prioritizing structured data specifically for e-commerce. And then the last piece of news this week: Google's research team they published a study this week testing 13 different language models. They included Gemini 3 Pro and also ChatGPT-5.

Michael 6:50 – 7:48

Among other models, on how reliably they can retrieve facts that they already have stored in their mental and model memory. So the Frontier models had stored pretty much all of the tested facts. It was above 95%. But they still failed to directly retrieve roughly a third of them. So point being, a model can know that our brand or company exists and have the right facts about you sitting right in the corpus inside of its memory, and it can still get us wrong. What that shows us is that the bottleneck oftentimes is going to actually be about retrieval of the information, not the information missing itself. And this is a different failure mode than the ones that we usually talk about in The Search Signal, which is usually about getting the fact in there in the first place. Also a good thing to point out is these recall issues and failures, they were not random.

Michael 7:48 – 8:43

So they reported that they were concentrated on what they defined as rare facts. And these were essentially facts that showed up less often in the training data. Popular facts, on the other hand, typically got retrieved pretty much just fine. So, what I would say is think about how that impacts or what that means for your company, because unless you're a very, very well known brand or a large company, your company's facts are probably considered rare facts to these models. So the more, the more often that our facts are getting repeated across the open web, the more reliably a model can retrieve them when somebody is asking a question related to it. So some of the corroboration work we're even going to talk about in today's episode isn't only about how a model learns who you are. It's also about what determines whether the model can find that answer when it ultimately needs it.

Michael 8:43 – 9:05

So all three of this week's news pieces come back to the same question, which is whether an AI system can reach and understand your business and whether it uses what you publish when someone asks a question. And this is exactly what today's episode is going to be about. So with that being said, let's dive right into it.

Michael 9:05 – 9:34

Okay, this is the first of four episodes on building a full funnel search strategy for AI search. And again, full funnel just for context, that means covering, you know, all the different stages of the buyer's journey that a buyer goes through. These are people who at the top don't know your category even exists yet. You work through it to the people who are, you know, comparing options and building a short list. And then you have people who are, ready to buy, who might just need, you know, one last reason to pick you.

Michael 9:34 – 9:59

These stages have always existed. These stages still exist and they will continue to exist. But what has changed is that AI now sits in the middle of each one of them. So when we talk about building a full funnel AI search strategy, today we're going to be talking about infrastructure. Once we tackle the infrastructure, then we can get into parts two and part three, which build on it.

Michael 9:59 – 10:57

And then by the end of this series, you should be able to sit down and really meaningfully dig into the work on building and executing an AI search strategy. As we're getting into this, there are really six steps involved. We need to start by doing an inventory of our entities. Then we need to run our keyword and our prompt research that tells us based off of our entities where we should be investing. Then we need to lay out our page infrastructure, confirm that AI systems can actually reach our content at all. We need to be able to declare what you are in a way that machines can read and understand without guessing. And then get that corroborated on sources on the open web that you do not own. That's going to be part one today. We're going to do an overview of that and talk about the infrastructure. Then we're going to dive into part two, which takes that infrastructure to the top of the funnel.

Michael 10:57 – 11:55

And then part three of the series will take it to the bottom of the funnel. And then part four is going to be how you measure the entire thing in process. And I will say this now, and I'm probably going to keep on saying this throughout the whole series. The point of this whole thing is going to ultimately come down to pipeline. It is not going to be focused on traffic. We need to have a paradigm shift on what AI search enables us to do and how it's contributing towards the growth of the business and where traffic and pipeline everything in between play. We've been evolving this at Victorious in our own strategy for a while. But one of the ways I know and I believe you can trust the process we're going to be walking through is we use this same exact approach to win this year's 2026 Best SEO campaign at the US Agency Awards. It was the same type of campaign built on the same type of foundations that we're going to be jumping into today. So with that being said, let's get started.

Michael 11:55 – 12:24

Okay, so most of us when we learned search marketing, we learned when it was a deterministic field. And what I mean by that is you pick a keyword, you optimize for it, you track its position, and then if you search that keyword, you know, two different times, you typically got the same 10 results on Google both of those times. AI search does not behave in that way at all.

Michael 12:24 – 12:53

These systems are different in that they compose an answer every time somebody asks. And then there's variation built into how each person chooses to compose their question and how the LLM chooses to compose their answer. You can literally ask the same question twice in a row as one person, and you can get a different answer with different sources behind each of the responses. Point being, when a team carries keyword thinking.

Michael 12:53 – 13:45

straight over into AI search and into prompts, pick a prompt, optimize for it, check whether you showed up. The checking step, that's where this approach is going to fail. And it's really an easy mistake because a lot of the tools that exist out there today to track AI search, they're built to look exactly like a rank check. But whether you appeared for, you know, one exact phrasing is really going to tell you very little because that phrasing isn't the same one that everybody else typed when they were doing their search. Instead, instead of that, what we're optimizing for is being one of the things in the system that it associates with the topic behind the question, however, that question might get phrased. And just for context, Google has also made this shift before. They actually made it a long time ago.

Michael 13:45 – 14:13

Around 2013 they had their hummingbird update, and that started grouping keywords by intent and by topic instead of just matching exact strings. And you might have noticed this, you might be performing well on a short tail, high volume, high difficult head term or primary keyword. And then you also saw yourself starting to pull in traffic in position rankings for long tail variants that you had never really even targeted.

Michael 14:13 – 15:36

This is the same exact shift, right? But it's also just accelerating significantly with this new technology. So taking a step back, if a specific prompt isn't the unit you optimize for in this new era of search and an AI search, you know, not like what we used to do with keywords in traditional SEO, the question is what is the unit you optimize for? And the unit that these systems think in is something called an entity. A specific and identifiable thing or person or concept. Okay. Your brand is one. Each thing that you sell or offer to the market, that's another one. Every location that you operate in, that's another one. So when somebody asks a question, the system identifies the entities inside of it and then assembles the answer out of the entities and the relationships that it has indexed. So a website page on our own website in that type of world is simply a vehicle for declaring what your entities are and then also how they all connect. Which means that everything that we're probably already doing in search, keyword research, the pages that we're building, the links that we might be earning, any structured data we might be deploying should all be focusing on that one specific and underlying job.

Michael 15:36 – 16:04

And I would also say, you know, as a side note, in this structure, these these AI systems also get the job of associating entities and relationships wrong a lot. And we have to get ahead of them. They happen in two main ways. The first one is something called a name collision. And you might be running into this with your own business. We definitely run into it with Victorious. Victorious not only is an SEO agency, it's also a Nickelodeon sitcom. And it's also a just plain ordinary English word.

Michael 16:04 – 16:34

So if you ask ChatGPT, tell me about Victorious, it could land in a bunch of different directions and talk about three, at least three different things. That's the first way. The second way is probably a little less obnoxious, but might be something you're you're running into with your own business and something that we see commonly, more and more actually, which is that the AI systems, you know, know exactly who your company is, what you are, but it has the wrong picture of what you do or what you sell.

Michael 16:34 – 17:02

And this might be because, you know, there could be a random directory listing that's out on the internet from a couple of years ago that still says you offer something that you don't you don't offer anymore. You dropped, right? Or you in the last year or so decided to launch a new service line and the AI can't recognize that and never attached it to your name. So, you know, someone might be asking whether you handle or do the thing that you now handle and do. And the answer comes back as a no.

Michael 17:02 – 17:40

That second one does ultimately cost us a lot more because you know, companies expand and grow service lines all the time. And if AI doesn't know that you do it, it's not going to recommend you. So it's not enough for a system or an AI tool to know that you exist. It has to know what you sell right now. And that's the problem that the next six steps that we're going to walk through solve. So as we're thinking about building the infrastructure, the core of a full-funnel AI search strategy. This is where I would start, and this is how I would build it.

Michael 17:40 – 18:07

All right, we need to start by inventorying our entities. So, first things first, we need to start with our company and count out the separate entities that a AI model or system needs to learn about us. So the first obvious one is going to be the brand itself, which is the company. Then there is each product or each service or category that you sell into.

Michael 18:07 – 18:36

Then there are things like people whose names are going to maybe carry some of the authority of the brand. It could be your founder or your exec team or any experts on your team that might get quoted in content. For a lot of businesses, if geography influences your business, so if you're a local based company or you have franchise locations or you service certain regions, each location is also served as one entity too.

Michael 18:36 – 19:03

So those are some examples of entities that we need to index and document. And for most companies, that list can run small companies to maybe you know, a half a dozen to larger companies, you can have a few dozen or you can have many dozens. But for each one, we are tracking the same three things. We need to track the thing itself, the entity, and also its attributes. These are the facts about it.

Michael 19:03 – 20:37

For example, like where your company was founded, where its HQ is, who works there, and then for each of maybe your services or products, the problem it solves, who it's ultimately designed for, what the deliverables are, how it's priced, and then also the relationships. And that would be this person works at your company and this service belongs to your company. Relationships are exactly what's missing when when a model knows your brand, but cannot tie it to what you sell. Then after you've done this, it is very important that we tier the list because not every entity deserves the same work. And not every entity should be given resources to grow. It's the entities that drive real pipeline. Those are the ones that need to get investment. A legacy service that like barely produces any revenue for the business, or maybe you have an office in a market that you're not trying to actively sell into. Those are the types of entities that need to get its facts right, but I would say probably nothing more. So when you're done with tiering, you should have each of your entities written down and then a tier next to it. Maybe a primary and a secondary and then a third tier, right? But that top primary tier is the most important tier and the one that we will be using as this episode goes on as an example of how we build the infrastructure for a full funnel AI search strategy.

Michael 20:37 – 22:32

Okay, so now that you have the list of your entities and they have been tiered out, the next question is which of those entities deserve the investment? So I would say that's technically like a research question, but here's the problem with researching it. Nobody publishes prompt volume. We don't know how often certain entities get searched in LLMs. There's no tool that tells us how often somebody asks an LLM a particular question. And it's possible that no tool ever will, because you know, you and I could ask the same thing with one sentence of difference or one word of difference, and we could get different answers. And that's not even considering the AI personalization layer we have baked into each one of our own individual accounts. So the way that we can solve for this is actually building on the core component of any traditional SEO strategy, which is keyword research. Keyword search volume paired with competitiveness is the closest demand proxy we have to prompt volume. And it makes traditional keyword research the first research step and and not something that an AI search entity approach is replacing with something else. So we start with keyword research by mapping viable keywords that we would typically want to rank for that also have enough search volume to a specific entity on your list. And that mapping is effectively our decision layer. It's going to tell us which entities carry enough demand to justify an anchor page and a tactical approach, which we will get into in in step three. And it also surfaces the concepts each entity needs to be associated with.

Michael 22:32 – 23:24

This also gets you, in my opinion, the dual benefit of having an actual keyword strategy for traditional SEO, which in my opinion is a non-negotiable part of a full funnel search strategy because traditional SEO, as we learned about earlier in today's news segment, still a huge portion of traffic for a vast majority of websites. But then after, you know, for each keyword that we are investing in, we need to move keywords to prompts. So we need to be able to write out how somebody would phrase that keyword as a question, because that's typically what ultimately gets typed into these tools. And that same keyword usually, not always, but usually splits into more than one prompt. I'll use an example, right? So if we're Victorious, a keyword that we might want to target is SEO company, but SEO company could become.

Michael 23:24 – 23:52

"Why should I hire an SEO company?" Would be top of the funnel, somebody that's probably still learning. And "What are good SEO companies?" And that would be further down funnel, and somebody who's probably focusing on building a short list. Those prompts also are representative and they are not an exclusive target or list that we want to be rank tracking.

Michael 23:52 – 24:29

But they do tell us two specific things. Number one, which questions your content on the site have to be able to answer. And then number two, which stage of the funnel each entity is going to ultimately attempt to earn its visibility in. So the output of this step is a keyword to entity map with a set of question phrasings under each entity that become our prompts. And then that document is the working document for the next three episodes as we are working through how to build a full funnel strategy for AI search.

Michael 24:29 – 24:52

Okay, at this point you know your entities and you also know the demand behind each of your specific entities. So now we need to turn that entity map into pages. So every entity you decided to invest in needs to get its own dedicated anchor page. And this is the one page on your website that sells the product or the service or is the location.

Michael 24:52 – 25:46

And is where somebody actually converts. Right? So for one of us, one example for one of us of ours at Victorious is B2B SEO services. We have an anchor page for that. The anchor page carries the attributes that somebody needs in order to choose. What does the service include? Who's it for? How is it delivered? What might it cost? And where can I get it? If your anchor page is missing one of those types of questions doesn't have content to address it. The model that's answering a buyer's question that might be about you is also potentially going to be missing it. So once we have that anchor page defined and we have addressed the questions, we then define something called our supporting pages. Supporting pages are the pages that explain everything around the thing that you sell.

Michael 25:46 – 27:34

The point being is the page that sells it doesn't have to get too deep into the weeds on any specific topic that is correlated to and important to the entity. So take your anchor page, for example, for one of your services or one of your products. That one job for that one page is to convert. What the service is and who it's for and what it costs. But a buyer and also an AI system, they need a bunch of surrounding concepts. And they need to be able to understand them before that page makes sense. Like it needs to know like what category is this service in? How is it different from one that seems similar and might be adjacent? Or when would I maybe need it? What does it cost? Is it just something that we offer one time or recurring. Each one of those questions has the potential to be its own dedicated page. And those are the supporting pages. And the supporting pages come off of that entity's attributes and the concepts that are attached to it. And generally, the rule we like to follow is one entity per page. A page that's trying to cover several entities at once ends up being very unclear about all of them because the systems are going to weigh prominence. And what I mean by prominence is which entity you have in your title, which entities in your H1, your first paragraph, and your URL. And then everything that is competing with that for that position is going to dilute it. So for example, for B2B SEO services, we need a page that defines what B2B SEO is, and then a page on how it differs from what a specific, maybe like a B2C company might need.

Michael 27:34 – 29:17

And so on and so on through the concepts that somebody has to understand before the service can make sense to them. And now the reason that an architecture like this with an anchor page and a supporting page works at all is something called cooccurrence. So when you have two things that appear together often enough in content that carries weight, a system learns that they are associated. And that's essentially what topic authority is underneath everything. It's also why publishing a bunch of scaled up AI-produced content about a subject doesn't get you there on its own because a brand and the concept have to keep appearing together on your site as well as off of your site. And then every page and every mention will help teach the system the same association again and again and again. Also as an aside, we talked a little bit about this earlier, but things like locations and people that are on your list also get the same treatment. Maybe on a little bit of a smaller scale, right? Your people might get like a page that has their attributes as a professional, maybe their role and credentials. I would also maybe recommend connecting it to the content that they have authored on your site or maybe somewhere on the open web. And then your locations, if you're a local business, each gets a specific anchor page for each one of those broader location areas with details that make that place specific and also articulate your company's expertise in that space specifically. So generally speaking, every entity that you decide to invest in needs somewhere on your site that states what it is.

Michael 29:17 – 31:07

Okay, now that we have our strategy in place, we need to talk about the next very important part of our infrastructure for developing a full funnel AI search strategy, which is answering the question: can the AI get to your content? Because before a model can pick up your content and it has to be able to access it and to read it. And while not overly common, it's absolutely possible for a site to rank on page one on Google for a keyword or a term for an entity and not rank for correlated prompts in an AI tool or system. Ranking and being readable by an AI crawler are two different technical problems. Getting read is what we call inclusion, right? Your content made it into a pool that a system draws from when it builds an answer. And there are two things that can break inclusion. We've talked about these in prior episodes, and I'll briefly talk about each of them. The first one that's very common is related to JavaScript. Okay, so if your main content on your site gets assembled by code that runs after the page first loads, a crawler that doesn't run that code only ever sees what was initially there at the beginning. Right? So The point being, most AI crawlers don't run that code. They don't run it. The other is something called blocking. It's typically a rule in your robots.txt file, and that's a small text file that is telling bots what they're allowed to look at on your website. Or you might also have a setting in your firewall or your CDN. Sometimes hosting providers might have it as a default and it blocks crawlers from getting to the content itself.

Michael 31:07 – 31:35

So the two issues are JavaScript and then blocking crawlers. And so to check each of these, for JavaScript, the first thing, very easy, you can just go to your homepage, right-click, and then click the view page source option. And this is the source content, not the rendered page, and just search that page source content for a sentence that you can see on the rendered page of the website. If the sentence isn't in the source.

Michael 31:35 – 32:04

There's a good chance that AI crawlers also cannot see it. The second one, you can just pull up your robots.txt file. Oftentimes it's just your website dash robots.txt and look for disallow rules against things like GPT bot, claude bot, perplexity bot, and then google dash extended. And then if you can't find anything there, you can also go if you're on, like, for example, Cloudflare or another CDN, you can just go into your security settings and confirm.

Michael 32:04 – 32:33

That the AI toggles are accepting traffic. And that that check right there does have a bit of a deadline attached. So I do want to spend a minute on this. If you're not familiar, Cloudflare launched a one-click block AI bots option back in July of 2025. And then this July, they replaced it with three separate controls. They are called search, agent, and training. A search crawler indexes your pages so that they can show up in search results.

Michael 32:33 – 32:58

A training crawler takes your content and uses it to build a future model of itself. So if you block it, it keeps you out of what the next model might know about you. And then an agent crawler fetches your page live for a user, typically in mid-conversation. You might see this Chat GPT searching the web while you might be actively asking a question and waiting for an answer. So blocking that one.

Michael 32:58 – 34:22

actually makes you invisible while someone is actively asking and trying to get more information from your website. So per Cloudflare's own announcement on September 15th, the defaults are going to be changing. So newly onboarded domains are going to be automatically blocking training and agent crawlers on pages that display ads, with search still staying as an allowed bot. And then also there's another detail inside of that same announcement that I think a lot of people are going to miss, which is starting on the same day, crawlers that do more than one job are going to get blocked according to everything that they do. So Googlebot, for example, crawls for search and crawls for training. So if your Cloudflare settings block training, you could potentially be blocking Googlebot, your regular Google indexing right alongside of it. So if there are fixes related to this and building the infrastructure to make sure that AI can access your content. Those fixes are going to go to your developer or to your SEO team. These are things like server-side rendering or pre-rendering the content so that content exists before your JavaScript runs on your website, plus basic changes to your robots.txt or CDN config so that you can let in the crawlers that you want. So clearing all of that, taking a step back, clearing all of that gets you into the pool.

Michael 34:22 – 34:50

It does not get you picked. So let's talk about that. Let's talk about getting picked or selected when an AI bot or assistant is forming an answer. So getting picked out of the pool is what the industry calls retrieval. And the first thing that you control when it comes to retrieval is what you declare about yourself on your own website.

Michael 34:50 – 36:41

And that's mostly the on-page entity content, which we have discussed and also schema work. Schema, we've talked about this before. It's code that you add to a page that spells out what is on it instead of leaving a machine to infer from just the visible text. I would say this is similar to tagging a photo with the people that are in it versus hoping that a software might recognize the face. So At a high level, two very basic ones that I would definitely recommend for most businesses. And number one is organization schema. This is the block that describes the company itself and also attributes that might disambiguate. Like your industry or your founding year. And then also something called same as property. It's very important. This is a line of schema that lists the other pages that describe the same thing. What I mean by that is a model like ChatGPT or Google AI mode learns that your website and your wiki data entry and your LinkedIn page are all the same entity instead of it having to guess or work that out through context. So that covers the markup, but two more things about how you write the copy itself on the website because The words that you are writing on the page are going to be doing work for the entity, whether you intend them to do that or not. So, first things first, I would say use the canonical name of your company on first mention. For example, Victorious, an SEO and AEO agency. That is legible to a model in a way that saying something like "we" or "the team", instead of saying "Victorious," never are. And a paragraph that's going, you know.

Michael 36:41 – 37:36

Five or six sentences with pronouns only, stops being in the eyes of a bot about the entity. And there's a really easy way for you when you're developing your content to check this to make sure that you are being explicit about associating your entity's name with the topic. You can go into Google's natural language API demo. This is on Google Cloud's site. You don't have to set anything up and paste your content into it and look at the entities that it detects and how prominent it scores them. So if the entities that it finds are not the ones that you meant the page to be about, then the page is not about them yet. And you can schema wrap everything around it and it will not fix it. So we need to rewrite that before we ultimately do a republish of that piece of content. And then also, just for a note about the schema itself, just to go back to it.

Michael 37:36 – 38:05

You know, a machine ultimately can parse a page without any schema on it at all. But what schema does is makes the content easy and faster to read and less error prone. And schema that's filled in with real and accurate facts are going to give the model a reason to use your content when it decides what to answer with. So an entity that has thin attributes is not as likely to be cited authoritatively.

Michael 38:05 – 38:15

as another brand that has it, no matter how clean the markup is. So it's the facts, not the markup. Is what the model uses and is more important than the markup itself.

Michael 38:15 – 40:06

Okay, so I said this earlier, but you can literally do every single thing that I just described, and you can still not get named when somebody asks a question about your category. And our own Q2 quarterly search report at Victorious looked at this exact thing and whether a verified presence in things like Google's Knowledge Graph, which is its database of entities and how they all connect together, tracked with how often a brand gets named in AI answers. And we found really no stable relationship. But what did correlate, and correlated very strongly with getting named was how much the rest of the web talks about the brand and the entity. So it's the same co-occurrence that we talked about in the architecture step, except the content is on somebody else's website. So getting your company named in other websites is, in my opinion, the single most important and efficacious variable we currently know about to move from just being accurately understood to being actually named in an answer. So before I give you, you know, the order that I would typically work in for this, it also helps to understand why the off-site record, why third-party mentions and being mentioned in the open web matters this much because it does change theoretically where we might want to spend our effort. So a model answers in AI models answer in one of two ways, right? Sometimes it answers out of what it has absorbed and usually during training. And they don't typically fetch anything. Your pages on your website aren't actively being read. And it's working from what I would say like is a compressed version of what the web said about you as of whenever that training data was last captured. That mode rewards being described consistently and accurately across a lot of the web over a long period of time.

Michael 40:06 – 40:55

And also remember the Google study we just talked about from the news. The facts repeated most often across the record are the ones that the model retrieves most reliably. But oftentimes, in other times, it's running a live search and it's doing it mid-conversation. And it builds the answer on top of the documents and information that it pulls back from the open web. And that's where citations come from. And it rewards having a page in the pool with a section that it can take away directly. So you need both. You need to have easily understood content about your entities on your website, but you also need an off-site record of third party mentions feeding the AI models when it triggers a web search so that I can pull that information and use it to enrich the answer.

Michael 40:55 – 41:23

Okay, so when we're talking about third party mentions and how I would go about doing this, and we will get more to this into some of our other episodes here in this series. I would first and foremost start with third party sources that we have in our first party control. Because as far as a, you know, a model is concerned, your LinkedIn company page or your Crunch Base entity or your Google Business Profile, or also, you know, all the listings on all the review sites that you might have access to.

Michael 41:23 – 41:52

You can log into each one of them and use them as a third party mention. But if they do not accurately and fully describe your business and the associated entities in the way that you plan in the strategy and the way that you built your anchor pages, then we have to get into these accounts and make those updates. And I would do that first and foremost. Then if you do not already have one, I would strongly recommend building a wiki data entry for your company.

Michael 41:52 – 42:20

And this gets confused with Wiki Wikipedia a lot. And I want to be clear, this is a different thing. So Wikipedia is prose written, and it's meant for people. Wikidata is different. It stores facts as flat binary statements that a machine can read without having to interpret. For example, what year you were founded, what city you're located in and headquartered in, what industry you operate in, who's your CEO.

Michael 42:20 – 42:46

It also gives you a stable Wikidata ID for your business that you can use in your schema and every other source that you can point at. And you can also create the entry yourself rather than having to wait to be written about by somebody else. Then you can also do your own people who are entities in their own right. Consistent names, a main profile, a linking back to the profile of the company.

Michael 42:46 – 44:13

So that when someone on your team might get quoted or get published somewhere on the open web, these systems can tell it's the same person and that they're also attached to your company. And also, all this is also how pages and websites demonstrate something we've talked about for years in SEO, which is EEAT, right? Google's shorthand for expertise, experience, authoritativeness, and trust. Declaring your entities and then covering all of their attributes and then getting them corroborated and then attaching them to real people is what we've been doing for a long time. And we get benefits from both traditional search and in AI search when we are executing on this. And then also I want to be very clear about this. You can do all of these changes. You can get your third party mentions going on the ones that you own. We can also talk, and we'll talk about in future episodes about getting third party mentions from websites that you don't have control on and you don't own. You can get your wiki data set up perfectly and you can do all of that and none of it can fix an answer instantly. And sometimes maybe not even in the short term. So don't forget, most of these models, they run on training data that's captured at some point in the past. So the entity work that you might be focusing on today, when you're building your full funnel AI search strategy and you're executing on it, you know, it prepares you for the next training cycle. And I want to be honest, that could be a couple weeks away. It could also be a couple months away, but we're taking a long term approach and it is one hundred percent worth it.

Michael 44:13 – 45:05

Okay, so today's episode is really the first of four episodes to build a full funnel AI search strategy. And I said using the same tactics that we use to help us win best SEO campaign this year at the US agency awards, and also to do it in a way that is specific enough to help you take action now. This first episode has been really about building the infrastructure of the strategy itself. How to build a search strategy that is meant to approach entities and improve broad performance across many different prompts and keywords, but also while making sure that the site is able to be crawled and understood by AI bots and search bots alike. There is a ton of ground that we did not cover today, including technical health, internal linking strategy, other foundational parts of SEO that are honestly just going to be way too down in the weeds.

Michael 45:05 – 45:35

My goal instead here has been to help you get a practical framework in this first episode for creating a plan because the next episode we're going to be talking about how to use that plan to grow the company's footprint in the top of the funnel AI search activities. And you will not want to miss next week's episode because we'll be talking all about how the top of the funnel search activity, which traditionally has been a significant entry point for traffic and for visits to your website, is changing.

Michael 45:35 – 45:47

And it's changing due to the zero click environment. And I'll be talking about why there's a strong argument to be made that traffic should no longer be the primary goal of your top of the funnel search strategy.

Michael 45:47 – 45:59

Thanks for listening this week to The Search Signal. If this was useful, please feel free to subscribe to The Search Signal so you get the rest of this series wherever you are listening or watching. And thank you for being here. We will see you for the next episode next week.