Latest / The Search Signal / How To Tell If AI Search Is Working

Transcript

How To Tell If AI Search Is Working · The Search Signal

• Michael Transon: Hey, I'm Michael Transon. I happen to be the founder and CEO of a search agency, but I'm also a bit of a professional skeptic in an industry that sells a lot of certainty. So this is The Search Signal, where we talk about what is changing in search and also what it means for you. So before we get into today's episode, let's run through this week's news. So the UK's competition regulator, which is called the CMA, just put out a set of legally binding requirements on Google under the UK digital markets regime. So the headline piece of this is what they are calling the FAIR ranking. What that means is Google has to rank organic results, including AI overviews, using what they call objective and non-discriminatory criteria. And give businesses advanced notice before any significant ranking changes happen. And they also have to set up a very real process for businesses to essentially raise their hands and give any concerns. They've they've given Google about six months to put this into place. So your company is not in the UK, I understand this might feel a bit ⁓ remote, but important that you know that the UK has. essentially been the place where Google tends to move first on type of transparency. So that AI overview impression reporting that started to show up recently last week in Search Console. it was it was this past month, that reporting also rolled out to the UK first. So point being, there is a track record ⁓ of the CMA essentially the lever gets Google to give a little bit more visibility into how their system and how this stuff really all works. Now, Google has said that they are going to work with the CMA and maybe they will. What I do not know though is whether that is going to turn into real actual insight and information into how AI overview rankings, for example, work or whether Google's gonna find a way to quote unquote technically comply without giving much away in the process. So ⁓ nobody can make that call yet. But if the last round is any guide to what's going to be happening here, it's definitely something that is going to be worth watching. Next, Similarweb tracked what happens to your site traffic after ChatGPT recommends your company. So what they did is they took US desktop data and they took it across a couple different industries, finance, travel, ⁓ and beauty, and they what people did in the week ChatGPT gave a recommendation. So companies that got recommended were about two and a half. times more likely to get a site visit in that week than companies that did not get that recommendation. And also more than half of the traffic that came through actually came through branded search and through a click directly from the AI. So what happened was people ⁓ the recommendation, they sat on it, thought about it, and then later on, They went and they searched the company themselves later. So point being is AI didn't even send the click, even though they gave the recommendation. So it's planting the seed in these situations. And then the click came later and it came through search. And those visitors, when they did show up, did two things that was really surprising. Number one is they spent about twice as long on the website. And then second, they went through close to twice as many pages as the other average regular user. So this for most people is an attribution problem. So if you are not tracking your citation share in ChatGPT or in other AI tools, you're probably ⁓ misreading your branded search is ⁓ actually from. So the signal of that. Right. The traffic, the branded search is very real. AI is recommending us, but it's just landing in a bit of a different attribution bucket than you would typically expect to see. Ahrefs pulled anonymized Search Console data from about 420,000 websites. And what found was the median click-through rate across most of the industries that they looked at for search sat somewhere between 1 and 2%. ⁓ So happening is most sites right now ⁓ are very thin ⁓ actual traffic. relative to how often they are actually showing up in search results. And then that doesn't even count what AI overviews are doing to that number. So what they also found was when an AI overview appears on a search or a query, clicks in the top organic search result, the number one position, dropped by an average of 58% compared to the same type of query or keyword that did not have an AI overview. So and if you don't know AI overviews are becoming more and more prevalent. They are now showing up in close to 60% of question-based queries happening in traditional search. So there is a very big category of keyword searches and in queries where this is happening. And on each one of those, where AI overviews are showing up, the organic click is losing a lot of visits. So what I'd think through. ⁓ for your company is whether the goal on how we rank and where we're showing up for searches has stayed the same in this new environment or if it's shifted. So ranking just below an AI overview on a very high value informational query, I would say is not as good of a prize as it used to be. So, what I would say is I'd probably want to work out. which informational keyword searches and queries you are tracking right now matter most to the actual business? And what I would do ⁓ ⁓ on putting energy and into ⁓ not ranking number one for that position, but getting cited inside of the overview. Cause I don't want to be fighting for holding a position underneath it when it's only getting up to maybe one or two percent of clicks. Last piece of news for the week, SEMrush tracked their own content across about 1700 AI prompts. And they were looking at what earns AI citations that compounds over time. So what they landed on is that content that's built around what I would call buyer trigger situations did a lot better across ChatGPT and AI overviews and Google mode than the usual keyword cluster approach. So the framing that they were using really lines up with what Byron Sharp's marketing research calls category entry points, ⁓ is basically the moment someone first realizes that they've got a problem before they are in market and they're actively searching for a solution. So ⁓ does make a lot of sense when you think about it. So if someone asks an AI, like, how do I know if I need a new CRM, for example, that is a trigger moment, right? the AI is surfacing content that speaks to that moment, not a page optimized for like the best CRM software. So ⁓ ran one piece anchored to a trigger moment, and then they watched its share of voice in that topic cluster go from around 15% the week before publishing to about 26% the week after publishing. And Semrush mentions across the prompts that cited it rose roughly 30% in the two weeks after. Now, they were pretty careful about this process. And the cluster that they were looking at had already been gaining momentum for a couple of months before. So the piece that they published extended a trend more than it caused a jump on its own, if that makes sense. so I'd look at it how your own content is organized right now. So if most of the content on your site right now is built around buyers that are, you know, already in market or already doing keyword research, you're probably going to be missing what is the earlier and very important layer where a lot of the AI citation opportunity actually is. So that's the week's news. Now let's get into today's episode. So most marketing teams have some kind of dashboard that they are living in on the daily basis and looking at search. They're looking at things like organic traffic and keyword rankings. You also be looking at things like click-through rate and branded traffic. All of those metrics are still very important, ⁓ still much worth tracking. And ⁓ I think any of that stuff is going away, but There is a spot that most tracking platforms cannot see, which is essentially what's happening over in AI search. And the dashboards that we're used to looking at are not going to tell us whether somebody found our company through ChatGPT or through Perplexity you know, whether your company is even turning up in those answers in the first place. And the thing is, it won't even flag whether you're showing up, but it also won't flag if anything's missing. So I want to spend today's episode working through a couple of these things. First, I want to talk about why your normal analytics can't really see most of the traffic coming in from AI. What that looks like, you know, day-to-day ⁓ the stuff you're reporting just is not right now catching. I also want to talk about why it's worth setting up a way to measure this right now, even though ⁓ AI in general is really, for most people, a pretty small slice of what you're getting in terms of traffic. And I also want to talk about whether. AI platforms even have your company right at all, which turns out to be the foundation that everything else is sitting on top of. And then we'll talk about how you actually measure your visibility without, you know, kidding yourself about the numbers. So by the end of today, my goal is you're going have a decent sense of where your reporting is currently going dark. And then also what I think you should be doing about it. And everything, as always mentioned today, is going to be in the show notes. So let's get into it. Okay, so let's start with the one thing that's underneath everything else. So your GA4 is probably telling you something that isn't quite right. And it's not because your GA4 is broken or not working the way that it's supposed to. It's doing exactly what it has been actually built to do. But there is a gap between what it can see and what is going on when someone is finding you through AI search. So the way it works roughly is that when you click into a site from a Google result, your browser sends along a little tag called a referrer header. Okay. It basically says where you came from. So ⁓ it be Google or ChatGPT, or it could be, you know, LinkedIn, whatever it was that got you to the site. And GA4, Google Google Analytics 4, ⁓ leans on tag to sort your traffic by source. So the problem is, you know, between people that are rejecting cookies and browsers that strip that tag out, and then also some of the AI platforms just not sending it in the first place, ⁓ that ends up getting dumped into something called direct traffic, which has no label on it at all. Which means it looks very identical in a reporting system to someone who just decided to type your URL straight into the bar and go to your website. There was a company Lomely that looked at a few hundred thousand site visits, and they found something like seven in 10 AI visits are currently showing up with just no refer at all. And GA4 therefore files every one of those visits as direct. So basically, the bulk of your AI traffic is just sitting there. invisible in the reporting, bucketed into the direct. Now Google did end up patching a part of this. So back in May this year, 2026, they added a AI assistant channel to GA4. So traffic from a few of the bigger platforms that I'm talking about, like ChatGPT and Gemini and Claude, ⁓ now get broken out on its own instead of getting buried in the referral or direct. buckets. So if you want to, you can go in and look at that if you have not. But I would say that's more of a floor not the whole picture of what's going on because it only catches the visits that show up with a referrer. ⁓ Perplexity, for example, still lands in your referral traffic. And Google's own AI overviews and AI mode get counted as plain organic rather than ⁓ And the the big chunk. like all that traffic with no referrer is still dropping into direct because there's no patch for that part right now. And you can't recover a signal, you can't recover a tag to identify traffic that was never sent. So what you can do is set up your measurement so it accounts for the gap instead of pretending that it's not there at all. So that's like the traffic that you cannot see at all. But even in the stuff that you can see. The usual metrics are leaving out a lot of things you'd really want to know. So the top result on Google used to pull somewhere around a quarter or so to a third of all of the clicks that are happening. And then featured snippets came along. And this was like back in the mid 2010s. ⁓ And snippets ⁓ started into that ⁓ click rate. And once Google was answering the question that people were asking right there in the box at the top, it was just simply fewer people needed to click through to get it, right? But that was a very slow, would say pretty mild drift compared to what has happened in the last year or so. So AI overviews have really that a lot. The Ahrefs finding that we talked about at the top of the show in the news section, where the top result is losing more than half of its clicks once an overview shows up. ⁓ That's this same drift. It's just moving a whole lot faster. So your company can be, you know, sitting in that same number one spot it has always held. And it could just be getting a lot less clicks than it did two years ago, with nothing about the ranking having changed at all. And if for you, sessions are how you judge whether search is working, you're gonna watch that number slide. And you're also gonna see your rankings hold steady. And the reporting that you're using is not going to give you a reason as to why. And then there's the space between getting cited and getting a click. So when your company shows up in a ChatGPT or a Perplexity answer, ⁓ reading your name right when they're doing their research, even when they're figuring out who's like even in the running as a potential solution. And a lot of those mentions ⁓ simply never turn into a click. That person is, ⁓ know, reading the answer in your brand name and they are getting the information that they need. And then they end up moving on. And they never land on your website. So your company can be showing up like over and over and over in these answers. And you might not see anything for it in the referral data. Now that visibility does still do something for you. You're getting in front of someone right when they're deciding on who to put in their short list. It's just is not landing anywhere inside of your reporting or your GA4. So if all you are watching is session data, that whole layer is going to be completely invisible to you. And then on top of that, the platforms themselves, the LLMs, don't all behave the same way. So AI search is not like this one monolithic thing. You've got ChatGPT and Gemini and AI overviews, ⁓ got Copilot or AI mode. They are all building their answers very differently, and they are pulling from a lot of different places. So your company can be all for example, Gemini. So AI search is not like this one monolithic thing. You've got ChatGPT and Gemini and AI overviews, ⁓ you've got Copilot or AI mode. They are all building their answers very differently, and they are from a lot of different places. So your company can be all over, for example, Gemini. And just have a hard time performing well in another platform like Perplexity. And just have a hard time performing well in another platform like Perplexity. So being strong in something like AI overviews does not mean that you're gonna simply just carry over into ChatGPT or even into another Google product like AI mode. So when you roll AI and ⁓ search into just one number, all of that, the platform differences and know, where competitors are beating you, it just all ends up washing out. And now the fair pushback on that though is that AI search is still a very tiny sliver of a lot of websites organic traffic. But you know, I would say how small it is right now is absolutely not a reason to ignore it because it won't continue to be this small forever. the most recent number I've seen, which came from a company called Content Square earlier this year, had put AI referral traffic at something like a fifth of visits on websites. And to be honest, that is probably almost certainly low because ⁓ that only the sessions where the platform placed that referrer tag that we were just talking about. And since seven in ten or more just don't pass it. The real number has got to be a lot larger than that. And plus, on top of that, a lot of this stuff is just moving so fast that like a benchmark from even just a few months back is probably way behind where things are at right now. So you hear a fraction of a percent and you think, you know, why would I build anything around that channel, that AI search channel? And I totally get it. But measuring it, It's really not about what you need this upcoming quarter. It's about whether you have a baseline sitting there when AI search does get big. And you can't magically go back and make a baseline after the fact, right? Every month that you are not tracking it is just another month of trend data that you're not going to have later when this thing becomes, you know, the thing everybody is asking about and is using. And there's also a quality thing in the data that is also pointing to the same way. So that same company, Content Square found that. people coming from AI bounce from site less than normal website visitors. And the conversion rate on that traffic is going up. I think they said it went up by more than half year over year, which is a huge increase. And it also lines up with ⁓ other data Similarweb that we talked about at the top of the show, right? In the news section, where the visits that came through just simply engaged more. They were visiting more pages, they were spending more time. on the actual website. So when you've got a small channel that is converting better than your main one, that in and of itself is a reason to go figure it out and to figure it out sooner, not shelve it for later, which gets us to how you actually measure any of this. So we've got all of this data and stuff that we have been tracking on our websites for years. We know our organic traffic pretty well. But the question now is how do you stretch that to cover the AI side of things too? So you get a really good sense of what your AI visibility is and what it's doing for your company. And I'd say the first thing to look at is the one your normal traffic reports just really are not answering, which is whether your company even turns up in these AI answers at all. It's not about how much traffic these AI search systems are sending to you. It's just whether you show up when someone uses AI to look into what you do or into your industry. And that gets decided way before any clicks come to your website, before anything lands in your analytics, before ⁓ of these buyers are anywhere near actually And this is most impactful in the categories where people do a ton of homework before they buy. So you think B2B professional services or software, I think financial products, you know, that like kind of thing. In those situations, more and more of the buyer's research is happening just inside the AI systems before they ever hit a vendor or a product website. So for a lot of them. the AI answer is kind of like a research at this point of the process. So when your company is not in those answers, you are not in the room when they are deciding who is even worth considering. And that is way more valuable and way more of a business impact than just a couple of missed clicks that you might be missing out on. And checking whether you're in that room starts somewhere different than just like pulling a channel report from GA4. It starts with a question most teams never think to ask, which is whether AI even has your company right in the first place? So before you can even ask how often you show up, you have to ask whether AI even knows who you are and what you do. And to get why that's a real question, it helps to know, you know, where these models are getting their picture of your company. And it's to be frank, not from your website mostly. These systems are building it from stuff that they have trained on and the places that they trust. And there is in a lot of them one source that towers over all of the rest. In one look at 30 million citations, Wikipedia was just under half of everything that ChatGPT was citing. Your own pages are just a sliver of this, and something like less than a tenth of what these systems are ultimately pulling from to understand your business. So the version of your company that AI is carrying around is most likely assembled from other people's descriptions of you, not your own. And a lot of what it knows about you is not generated fresh every time it looks. Someone poking at ChatGPT recently noticed that when it tags your company, the little description it attaches comes back word for word, identical, run after run. And that is a tell that it pulls from what I would call like a stored knowledge graph, which is essentially a fixed set of facts about who you are. rather than just you know writing about it new every time someone asks a question about your company, which matters a lot because if the stored description of your company is out of date, or if you, for example, repositioned or you know you moved up market or you changed what you sell or you launch a new product, every answer keeps describing the old you. And you cannot just go in and edit that field directly in AI to get a better answer. And the stakes on getting this right, what your business is and what it does in these knowledge graph or corpus systems just went up a lot. So back in May, there was a court in Munich that told Google it was effectively on the hook for what its AI overviews about a company. there were two publishers that found Google's AI was describing them ⁓ as running ⁓ and tying publishers' businesses to. dubious business practices it pulled that information from ⁓ in the actual sources. ⁓ AI essentially had, you know, basically blended them together with some other sketchy And the court treated AI overviews as Google's own words and not like a list of links. So for Google, the usual protections did not apply and it ordered Google to stop. So the knock-on effect of that is the part that we want to talk about here, which is, you know, once an engine can get sued for what it says about a company, it has every reason to go dormant and quiet about the ones it isn't sure about and will rely on confidently naming the ones whose identity is very clean and clear and also really easy to verify. So getting your business or your company's entity right, it's no longer like a nice to have. And we cannot rely on AI. And it needs to turn into the thing that decides whether you're even safe to mention as an LLM. So the move here is you need to go and check. So you can go and open ChatGPT or Gemini and you can open it up in a fresh window, do it in incognito mode where they don't already know you and just ask it straight, right? What does my company do what does it compete with? Then I would just grade what comes back, right? Meaning a stranger would come back understanding you correctly. You know, ⁓ old version that, something that we wouldn't like, meaning it's describing a version of you from, you know, an old position of the company or old services that you offer. But then wrong would be meaning it's. confusing you with someone else, ⁓ another company with the same name, or inventing things that it thinks you do or offer that you don't. Or it could also come back totally blank, meaning it doesn't know you exist. the one that I would say catches people off guard is that I would say wrong is actually worse than blank, right? Blank could cost you a potential buyer, but wrong costs you the buyer and hands them to a competitor in a confident voice and has a source attached to it, right? But there's no dashboard for any of this, right? The only way to know what AI says about your company is to go it the way that a stranger would. So let's talk about where to start. So, how do you actually AI recognition and visibility without fooling yourself? Right. The trap that you could run into is treating it like a big pile of metrics that you need to go out and just collect, right? But what holds it together in one piece is the entity, right? It's your company, it's your product, it's the thing that you want to be the answer for, right? ⁓ AI doesn't keywords. Okay. AI does not cite keywords, it cites entities. So you need to measure at that level and around the two or three things that actually drive your business, not every turn that you've ever chased. And I'd also split what you watch into two distinct separate layers. So the first one being leading signals, right? Which tell you whether AI knows you. And also ties you to your category. And then I would look at lagging signals, which tells you, you know, what that's doing to your traffic once people actually show up on the site. So most teams are running straight to the traffic because it makes sense. That's like what the dashboards that they are using show them, right? But the leading stuff is what moves first. And it sometimes can move months before. any of it lands in your analytics. So that is where I would start and I would recommend you start. And the recognition check we were just talking about is in the very front of it. So the leading signal I'd build in an AI search program is really around something called share a voice. It's not whether or not we came up for a certain search, but it's across the real questions a buyer in your category asks. And what share of the answers you versus your competitors that you actually care about? So it's the closest thing I would say AI search has to rank. ⁓ not a lot of people are tracking it. It's like I I was looking at stats, something like 14% of ⁓ are looking at share of voice, even though same study said there's like 40% ⁓ marketers saying AI search is a priority for them this year. So the measurement that is being used by a lot of these marketers is just simply behind the work, right? But you really have to be careful about how you measure it because this is ⁓ polling mechanism, not a counting mechanism, right? You run the same prompt twice and you're often gonna get a different set of brands back. right? We talked about this in the last episode, but there was a study that ran a few thousand of these, and they found that less than one in a thousand chance that two separate prompts gave the same ordered list of recommendations, right? Citations on the same question also swing a lot, sometimes 40 to 60% month to month, right? So a single check is really not going to tell you much, if anything. So what you want to do is your share of voice across a lot of runs. Reported as a range. You cannot have a tidy little sweet number that you pretend is precise because it's just simply not. That's not how the systems work, right? ⁓ are two things that I would say would make this very practical. Okay. Number one, like I'm saying, runs matter a lot more than prompts. A small set of questions run a bunch of times tells you a lot more than a huge list of questions that are just running once, because a lot of the wobble that happens in these responses happen in the reruns, not in how many questions or prompts you are tracking. And then number two, the platforms aren't equally wobbly. Okay. So like ChatGPT tends to reword your question into new sub questions every time that it answers you. So it swings a lot more and needs a lot more runs to really settle down and get to the center. On the other side, Perplexity runs a lot closer to like a straight search. So it's a lot steadier and a lot fewer wobbles. So, like for a real company and you know, not an agency or something, the floor I'd start with is something like, you know, 30 questions. And I would tie those to your core entities. I would run them several times each. And I would also run them across the platforms that from your perspective, your buyers are most likely to use, right? And then I'd pick definition of share a voice and I would just stick it. And don't ever blend the platforms into a single number unless you have to, because these platforms like they barely cite the same sources. If you don't want to do this manually, you could point to a tool like Scrunch and use that and let it run the sampling for you. And then underneath the share a voice, there's also I would say an even earlier. signal worth watching, which is something called co-occurrence. AI works out who belongs in a category by how often your name turns up next to that category and its players out in the open web. this is in things like comparisons and roundups and things that are not on your own website. So when that climbs, your AI visibility tends to follow a few months behind it. So if you're keeping an eye on where your brand shows up alongside the category, you've got the earliest read on way before where it is heading, before it reaches a share of voice number, and well, well, well before it ends up turning into any version of traffic to your website. So for Google's own AI systems, there is free shortcut. And I would definitely recommend getting on this while you are at it. So Search Console recently added a report this month in June that shows up whether your pages are turning up in AI overviews and in AI mode. It is impressions ⁓ only. it does only tell you that you showed up. It does not say whether anyone clicked. And it's also rolling out slowly. So if you don't have it yet, I'm not surprised, but hopefully you will have it soon. Bing also has a similar one for Copilot. they only cover Google and Bing. So for you know, ChatGPT or for Perplexity or Claude, you have rely on. doing your own manual sweeps or using a tool like Scrunch or or Peec. So that is the leading layer, which is recognition ⁓ and of and So the lagging layer, which is after the leading layer, is what actually shows up ⁓ once AI actually starts moving toward you. And this is absolutely, in my opinion, worth tracking. But we need to treat it as a confirmation of the work, not as a headline for the work. So the first thing I would look at is branded search. So when somebody runs, you know, into you in an AI answer and doesn't click, they're often going to search your brand name instead, which is that Similarweb that we talked about ⁓ from top of the show, where more than half of the traffic ⁓ after recommendation comes through branded search. So Your branded search in Search Console drifting up, and there's no like strong big brand campaign that you were pushing across marketing. That is what I would say is a fingerprint of AI visibility that you're probably not otherwise catching. I would hold it a little bit ⁓ but I would also watch it move across some of your other leading signals. So The second one I would look at is your AI referral conversion rate in GA4. Okay. So since May, GA4 has been breaking out some of this on its own in that native AI assistant channel. So you might already be seeing slice of this, so I would build your own segment alongside it for the platforms that it misses. I would say, especially Perplexity, right? there's also a blog post in the show notes that we will publish that walks through the setup for this. it's going to undercount because of those missing referrer tags we were talking about earlier. So I would say call it a floor, not a ceiling, when you are reporting on it. what holds up is the comparison, right? How often AI referred visitors convert against your organic baseline. And that is going to be a true thing that you can look at really no matter how big or how small the volume is you're looking at. And then the third thing I would do is just ask, right? This is an interesting one, but you know, you can put just like an optional field on your lead forms, how you hear about us drop-down, and you can put some AI tools ⁓ as to select on that. ⁓ It what the other two might right? We've seen people type ChatGPT or type Claude into that box on our own forms, even when HubSpot has the session filled as direct or even as So it's really just like the one signal that comes straight from the buyer instead of ⁓ a signal that getting inferred like the rest of them that we see. So that's pretty much the shape of it. The leading signals are going to be telling you where you're heading, and then the lagging ones will confirm it once it arrives. And it is most convincing when these two things work together. So if you are going to do one thing this week, I would say ⁓ it the front of the leading layer, which is to run the recognition check and a share of voice sweep your or three. core entities of your business. That's what your business is and what it does, right? That is your baseline. Most companies, like nine in ten, are going to find out that they are starting low. They might be starting near zero. And you cannot show that you have moved anywhere until you know where you currently are standing. So, like we talked about at the beginning of today's episode, most people listening to this podcast are working off of analytics systems and dashboards that are really only telling one part of the story. And the part that is missing is not like an edge case. It is the stuff that is happening inside the AI platforms before anyone decides who's even worth a visit. So the signals that we went through today are not going to show you all of that. conversation and all of the data and analytics in one place either, but they're going to show you a lot more than what your standard setup does on its own. And really the whole thing comes down to, you know, which question you're asking. So if the question is how much AI traffic is sending you, your normal reports are going to be handing you a And I I'll just I think it's just the wrong one. And it's not the whole number anyway, right? If the question is instead whether your company even turns up in these answers, ⁓ that takes signals and a different place to start. And now, if you follow what we've talked about today, you've got both. So the baseline that you put in this week is the one you're going to wish you had, you know, a year and a half from now, when you are looking at AI search becoming a bigger and bigger part of your overall website traffic. So here's what I would take ⁓ with you as you leave today. So, number one, your normal analytics have a ⁓ blind spot right now, today, for AI traffic. And inside of those systems, there is no clean fix for it. So between cookies that are getting rejected and the way that platforms handle their outbound links, most AI visits are landing in direct traffic and they have no label. Google's native AI assistant does catch part of it now, but the bulk of that traffic is arriving without a referrer and cannot be recovered. So you are ⁓ around the gap instead of pretending that it is not there. The second thing you need to do. Before you measure how often you are showing up, we need to first check whether AI even has you right. So we need to ask these search engines and LLMs who you are in graded as either correct, old, wrong, or blank. Wrong is the one to fix first because a wrong answer does not just lose you a buyer. It's going to hand them to a competitor. who has a good, strong, confident voice and has a source attached to the recommendation. That is what I would call a knowledge graph problem. And it is the floor and foundation that everything else is going to be sitting on top of. The third thing that I would focus on, the signal I'd build, reporting around your share of voice, which is your share of answers against the competitors that you care about, measured at the entity level. And we need to treat this, like I said earlier, like a polling. okay, not like a rank. I would recommend you run a focused set of questions and you do it several times for each of them in sequential days and keep the platforms separate. And I would report a range of share of voice. And then I would pair those leading signals with the lagging ones. And if you don't recall, those are things like branded search, your GA4 segmentation, and the form field that we talked about. So these are a few things that are pointing the same way before you can ultimately call it a trend or a growth or decline. And the fourth, the reason to start right now is your baseline, which you cannot go back and create later. But I would say be clear and honest with yourself and with your leadership about what this is and is not, right? Seeing the gap, knowing whether you show up and how often. That's what these signals are for. Closing it, actually getting into those answers is the building work. And that's the job we got into in the last episode. This one is about being able to see it very clearly. And once you can see it, you can ultimately go and fix it. If you got something out of this, please go ahead and follow the show wherever you are listening. So the next one finds you. And as always, all of the data that I mentioned is sitting in the show notes. That's it for me. This is The Search Signal. We will see you guys next week.