Name: Q2 Yext Research Live: What Our Data Reveals About Citations and AI Search Recommendations
Uploaded: 2026-06-17T18:16:48.076Z
Duration: 58 min 32 s
Description: Q2 Yext Research Live: What Our Data Reveals About Citations and AI Search Recommendations

Transcript for "Q2 Yext Research Live: What Our Data Reveals About Citations and AI Search Recommendations": VP of data strategy, and today is about one shift. Search is splitting in two. Broad searches still find your location page, but the high end 10 ones, the specific service, the exact need, the searches that actually convert, AI skips the homepage and hunts for the page that answers the question. Specificity wins. Our team has the data to prove it, plus the answers to whether acting on Scout actually beats your competition. Two people are gonna break that down for you today. Anthony Rinaldi will go first. He's our VP of insights and analytics. He'll discuss whether better profiles win and if locations behind can still catch up. Arjesangwon, our director of data strategy, he's gonna speak, to which pages AI really sites and what. Questions in the chat anytime. Let's get into it. Anthony, kick us off. Cool. Thanks, Javi. Yeah. I'm gonna kick things off like Javi introduced with a session on what I'm calling do better profiles perform better and are poor profiles doomed? So this is research that I have been wanting to do for a very long time. I've been at Yext for seven years, and in my seven years, we've always given our clients recommendations broadly. Right? You need to be better online, and if you are better online, then you will perform better. And these recommendations historically have been based on, our client data, and they've been based on best practices. But once we got scout, and we've talked about this before, but once we've gotten scout, all of a sudden, the universe of data that we've had has expanded so much. Because now we have data not just about our clients, we have data about the world. We have data about who shows up alongside of our clients and who's doing things well. So for this research, I wanted to look at when we recommend things to do, do those things make a concrete difference in online performance? So from a data perspective, I'm really gonna try to set the stage in the first handful of slides here. Data, I looked at 500 brands across 13 industries with over 600,000 physical locations that have been scanned by Yext for six consecutive months and have been live with Google on Yext for over twenty months. And I'm gonna go over, exactly why each one of those qualifiers was used. So before we get into the research, what does it mean to be scanned by Yext? Well, if you're familiar with Scout, essentially what we do is we travel virtually to locations where we know businesses are and we search for the business's primary keywords. So for example, if we know that there is a coffee shop in New York, we go to that neighborhood, we search on Google, and we search across AI for coffee shops in different ways, and not only do we get back a ranked list on traditional search, we get back, AI answers. Right? We get back citations, which Arjun is gonna go in deeper in the second half of this, this session. But what I think is one of the most valuable things we get back is not who ranked first and who ranked second and who ranked third. It's what did the people do that ranked first, and what did the people do that ranked second and third. So we don't just record the rankings. We record everything about their public profile. Why, did I say we looked at six plus months? Well, the second half of this research was not how do you look today, but it's how did you look. So I wanted to make sure everybody in this sample size had a profile today and had a profile in the past so we can answer not just how did they look, but how much work did they put in. Because the title of the session was not just do better profiles perform better, it was are poorer profiles doomed from the start? So what does it mean to be good? Right? I like to think about targets or OKRs operationally. So historically, we had research, and I say we because I've been here long enough at Yext to have done this research myself, we have research about how many publishers you should be on. We have research about how much of your knowledge graph you should fill out. We have all this research based on historical best practices, but I think that is maybe missing the mark. Right? Because historically, we said 75% is a is a good number of long tail publishers to be on or 90% of your core information needs to be filled out. And what we learned is you set these targets of 75% or you set these targets of 80% or you set this target of responding to 85% of your reviews, and you start to congratulate yourself because you hit that target or you passed that target. And what we've learned is that with Scout, we cannot say we don't need to say 75% or 80% or or 85% carte blanche, we can say this location needs to have 500 photos because they're in New York City, and this location only needs to have 100 photos because they're in a more rural area. So when we think about setting targets and what it means to be good, I don't wanna say, are you good based on these static targets? I wanna say, are you good based on your competition at a local level? Because what we learn is that, kind of go through these scenarios here, there's a world where in your first scan you were bad, and not only didn't you hit your target, but your competition got better. In the second scenario here, you hit what we thought your target was, but six months later, your competition has gotten better, so let's not celebrate too early. The third one is the ideal one where you not only beat what you thought you needed to do, but you passed your competition. And then we think about scenarios where six months ago you started ahead. Right? So you kind of rest on your laurels, and there's a world where just because you were ahead of those targets we set does not mean you are good because all of a sudden your competition has added more photos or has added more reviews or has started to respond to more reviews and started to get more publishers. So when we think about what does it mean to be a good profile or a bad profile, I don't wanna talk about just static targets. I wanna talk about, in your neighborhood, do you have more photos than who's around you? In your neighborhood, do you have more reviews than who's around you? In your neighborhood, is your rating higher than those around you? So your target is localized, and your target moves as your competition improves. I keep talking about good. So with Scout, we get about 70 between 70 and a 100 operational metrics that have these benchmarks. Right? Your photo count and the best people around use photo count. I keep using reviews as an example, rating as an example, attributes, response rates, and then we have metrics like freshness. So are your newest reviews more recent than your peers' newest reviews? Are your newest photos more recent than your peers' newest photos? Is your most recent negative review more recent than your peers' most recent negative review? So instead, there's there's a totally separate study to be done about in what industry does what field matter most. What I wanted to talk about here is generally how often do you beat your peers in things. Right? So this location is better than its competition in half the things, and this location is better than its competition in two thirds of their things. So we've got good locations, and we've got poorer locations, from a profile perspective. And then we have to define what is good and bad. So if you think about the distribution within a brand, you have brands where they have this perfect bell shape. Some of their locations are beating the competition in a 100% of the things, and some of the locations are losing to the competition in 99% of the things. So this perfect bell curve where I can say it's pretty clear that the folks on the left are poor profiles and the folks on the right are great profiles. But with some brands, you have, every even the worst of the locations within a brand are beating their peers in 60% of the things, and then you have some brands where nobody, not not a single one of their locations is beating their peers in more than 50% of the things. So because we have this spread of in inter brand or inter intra within the brand good and bad profiles, we don't think of these as beating in 10% is bad and beating in 90% is great. We think of these as within brand deciles. So decile is like a quartile. It's just 10. So the worst 10%, the second to worst 10%, the best 10%, the second to best 10%, and we're gonna use this throughout so we can normalize that even if your brand is incredible, they're still poor performers. And even if your brand is terrible, they're still good performers. And finally, what does it mean to perform better? So we can look at these profiles in decile one that have fewer reviews and worse reviews and fewer photos and no posts and no landing page and no Google attributes, and they've never responded to to any reviews. And we get the other side within a brand of folks that are doing everything right. And we can look at what I like to talk about is Google engagement because, yes, we pull in rank, we pull in Google rank, we pull in AI rank, we pull in AI visibility and citations and sentiment. But for me, the reason that I that I limited this to 600,000 locations that we have at Yext is because we have historical, Google data. And I like to use this because as we talked about in the last research webinar is that, the rank's kind of fickle. Right? I'm not just going to congratulate decile one because they happen to rank well on a Tuesday afternoon and, and, you know, penalize decile ten because it wasn't their day. Instead, I wanna talk about did folks literally care more about this result. Right? When they saw it, did they click, or did they not even see it at all? So how do we normalize that? Because within big brands, some locations are gonna get a million clicks, and within small brands, no locations might get more than a 100 clicks. So I don't wanna say, you know, a great location at at brand two never gets more than 50 clicks, so it's a poor performer. And a great location at at or a terrible location at brand big brand never got fewer than a million clicks, you know, so it must be a great performer. Instead, not to bring up some PTSD from from everybody's college statistics classes, but the best way we can normalize this is to talk about standard deviations and z scores. So I don't mean to mention this too much more. I don't wanna get lost in in in the details, but the way that we can normalize location sorry, brand big to brand small and industry big to industry small, right, because financial institutions are gonna get way less traffic than a than a fast food chain, The way that we can normalize this is to talk within the brand, were the good profiles performing at a better rate than the bad profiles, and within a different brand, within a different brand, the same thing. So you see here, I'm not gonna celebrate that they got 350 clicks and and they only got 50 clicks, but I will say that that this bar, the 350 bar might have been three standard deviations away from the average location within that brand, and the 50 click bar might have been one standard deviation behind, its peers within small brand. So I don't wanna talk too much statistics again, but I did wanna say I I did not take the data at its rawest form and just say big is better than small. So what do the results look like? This, I couldn't have drawn up much better of a slide. So what we have here on the bottom is deciles, decile one through decile 10. So if your worst profile is on the left, then your best profile is on the right, what you end up seeing is that average performance happens right around d six. So average performance within a brand happens when you are within the sixtieth percentile of your peers. But if you are a great profile versus a terrible profile, so d 10 to d one, then you perform on Google about one almost one and a half standard deviations better, by just having a better profile. And I promise that I wouldn't say standard deviations and z scores too much. So what does that mean? What does that mean, in in layman's terms? That if your average location at a business gets 100 clicks, then your best profiles get about a 180 clicks, and your worst profiles get about 60 clicks. Normalized for industry, normalized for brand size, normalized for everything. And that is, pretty pretty, confirming research. Right? That's exactly what I would have wanted this to say, and and, thankfully, that that is kind of what the research bore out. Is that, repeatable? I said it was five over 500 brands in the beginning. I brought that down to 483 to make sure we weren't getting thrown off by too much sample size. So we look at all of the brands that had about at least 200 locations that we power, and that holds true that there's a positive relationship between the better profiles having more engagement than the worst profiles in 98.6% of those brands. And I apologize for these slides. I hope they looked better on my screen. I hope they look okay on yours. But, essentially, everything on the prior slide holds true in almost 99% of all of the cases we looked on a brand by brand basis. Metric by metric, every metric, is a bit different. So I'm gonna I'm not gonna drill into every one of these. I won't mention them, in a bit. But if you look at things like photo count, it it's an example I keep using because it's one of the handful that you can see as a user pretty obviously when you pull up a location in Google. But if you look at something like photo count, what you see on the bottom axis here is, how close are you on average to your peers. So the left hand side is you have less than 50% of the number of photos as everybody around you, and on the right hand side, you have double the amount of photos as anybody around you. What you end up seeing is that there is a bit of a messy middle where if you're somewhere between 10% of your competition and 10% of your competition, it's it's okay. Right? That's that's about par for the course. Everybody's within some kind of a a messy middle. But but what you see all the way on the left is the folks that have half as many photos as their peers are getting crushed, and the folks that have twice as many photos as their peers are crushing. The pattern is similar but holds probably even more true with something like review count where every new review is important no matter where you are in this curve. If you have half as many reviews as your peers, you're getting crushed in engagement, and if you have, if you're crushing the competition and the number of reviews, then you're you're getting the most possible actions on Google. But if we see something like Google attributes, this is something where you can fill in the the the official attributes that you can send and sync to Google, there seems to be a ceiling. Right? It goes up and up and up and up and up, and everybody loves it when you have more attributes than your peers. But when you start to get to two times as many attributes or or more than that, it probably starts to feel like noise. Right? There is a ceiling. There's only so many attributes that you can fill out, and if you have 10 times the number of attributes of somebody else, that probably speaks less to them having poor profiles and more to you just adding fluff for the sake of fluff. How about AI? So, essentially, we have I I mentioned this before. Not only do we have traditional rank in all of our Google metrics, but within Scout, we can look at AI visibility, and AI sentiment. And we saw that in those same, profile performance buckets, I have this in quart quarters quartiles here, and it goes up into the right. And if we go to deciles I started with quartiles because it was very clean, but deciles are generally up into the right. We have this normalized to brand at this point to make sure that we had a good sample size in each one of these deciles. But what we see, again, is generally when we get to these big numbers that AI is favoring these better profiles across the board. So the first half of this research was do better profiles perform better, and the second half was an interesting one, was are poorer profiles doomed? So what do I mean by that? So we looked at the scan from last month, so April of, April or May 2026, then we looked at the scan from six months ago. And when we looked at the scans of six months ago, we looked at the worst quartile of every brand. So remember my slide from from a little bit ago, even the good brands have poor locations relatively. So if we just look at each brand's worst 25% of locations and we look a year before each of those from a Google perspective. So I didn't wanna say based on the October data and based on the April data, you went up on Google or you went down on Google because I know there's so many industries that we work with that are incredibly cyclical. Right? You might have gone up because of tax season, or you might have gone down because you're no longer in the holiday season. So what I wanted to look at, and this is why we had to limit the data to those that we had at least two years for, what I wanted to look at was on our initial scan six months ago, how was your Google performance twelve months before that? And then our new scan, right, the scan from April, how was your Google performance twelve months before that? And using this, we kinda get rid of some of that seasonality, so we can say the bad profiles six months ago, how were they trending in Google performance? And if they got better, how did they trend in Google performance? So there's two ways of looking at this. One is the x axis here is what happened between six months ago and today. So declined, flat, minor, moderate, big. So declined are folks that between six months ago and today ended up winning fewer metrics. That doesn't mean that they deleted some of their photos. That probably means that their competition added photos. That doesn't mean that they deleted some reviews. That probably means that their competition added reviews at a faster rate than they did. So if we think about this, the left hand side, the declined buckets are the ones that were bad six months ago operationally, and now they look even worse. And then we have flat, minor, moderate, and big being that group of folks who put effort in. Right? They were a poor looking profile six months ago, but they caught up in 5% of the metrics, or they caught up in 10% of the metrics, or they caught up in 15% of the metrics. So that we see on the right hand side, big, being 10 to 15 percentage points, are folks that six months ago were better than, their competition in 10% of the things, and now they're better than their competition in 15% of the things. Sorry. 25% of the things. And on the x on the y axis, we have the growth. Right? So luckily in our dataset, we have even the poor performers that declined growing at about a a percent on average, but the ones over here are closer to 5%. So the ones that were not great looking profiles six months ago, if they put the work in, they were able to grow much faster than those that didn't. And statistically, again, I don't wanna take you back to college stats, but there's a handful of ways of looking at this. We did first winsorized mean. So essentially saying, what if we took the average in this in this large dataset, which doesn't always feel perfectly, responsible because we have outliers? So winsorized means, essentially, we're not just taking the average that gets skewed by the outliers in each direction, but we're gonna take the top outliers and try to bring them back down to reality. So this is a method that some researchers do, but if we want to do something that's a little easier to comprehend, we can just take median. And if we look at the median, this chart feels a lot easier to swallow. So instead of saying, hey. The the the average or the mean is skewed because of x, y, and z, so we're gonna play with that. What we can just say is, you know what? There's there's plenty of noise here. Let's just take the midpoint or the median. And when we do that, this shows that the folks that were poor performers sorry, poor profiles that declined in their metric win rate actually fell 3% in growth rate on Google. And the ones that were poor profiles six months ago and they put in a ton of work grew. Right? Black and white. They fell versus they grew. They fell by 3% versus they grew by a percent and a half. So this, again, is is reassuring for when we work with clients and we say, hey. These are the things that you should be working on. Instead of telling them to do work for the sake of doing work, we're gonna tell them to do work because that work can turn around a business. Right? It can start to send stronger signals. It can start to, tell the world that they are open for business and and one that they should be visiting. All of this data has been aggregated a million different ways, but we have the raw data. And the raw data allows us to play by industry, by business size, by anything. And what I was happy to see for the study on the last two slides is that that positive relationship between the poor profiles that didn't put in work and the poor profiles that put in a ton of work holds true to different extents in every industry. So we get, something like health care really respecting a profile's turnaround story. You know? The the ones that didn't have hours and didn't have pictures and didn't have a cover photo and didn't have their services or whether or not they're accepting new patients, if you didn't have any of that and your competition did six months ago, within six months, there's a there's an opportunity for a quick turnaround story. And I keep I keep using, some of these metrics up top of my tongue, as examples like profile completion and, profile completion and freshness and visibility and all those things. These were, again, intuitively, the top 10 things that pushed the needle in these turnaround stories. If the text is is over the chart for everybody else, like it is for me, the first one is accuracy, the second one is visibility. Right? Are you in as many places as you should be? That's our visibility metric. Is your data synced and locked in all of these places and consistent? That's our accuracy metric. So, again, I think great research is research that that isn't too surprising because if it was incredibly surprising, I I would think that we're, you know, we might have missed something along the way, but there are surprises in here. Like, rating change in the past year on your Google reviews is almost as important as just your rating in general. Right? And I think we probably understand that, generally, I want to see more five star reviews recently because it means that the business has get been getting better. But if they've changed and gotten better, that's that's actually sending a signal too. So in conclusion, I think, from a data perspective, it's important to make sure you know your competition because setting those setting those original benchmarks might not be enough because those benchmarks are consistently changing. Beating your competition, understanding that it means not just hitting these arbitrary, targets but making sure you beat the exact people that are around you and don't stop improving because, yeah, there's an opportunity for a turnaround story whenever, whenever one presents itself. And then the last slide that I had in here was just, I'm a Knicks fan in New York and we're, you know, a day away from the parade and a week removed from winning the championship. So as I look around at my movie shelf, I saw do the right thing, and I was like, that's that's kind of the that's the summary for this entire presentation is don't just do work for the sake of doing work, do the right things, and understand your competition can make sure that we point you in the right direction. Hey, Anthony. So for that. That was super insightful. I got three questions just kinda round it out for the crowd here before we move on to Arjun. The the first one is the dataset is big. Right? 500 brands, 13 industries, 600 plus thousand locations. We scan six, consecutive months and live on Google for over twenty months. Right? Why those filters and what makes this a a a credible base, to draw conclusions from? Why those filters? I I think I generally went over it, but to say it again out loud, we needed we needed, yes, we put picked data from people we we don't work with, so we had these great targets. But in in order to get the Google data, which is the what I would say the truest outcome that we can we can measure, we needed them to be our clients because we don't yes. You we can see how a profile looks, but we can't exactly see how a profile performs if we don't work with them. So that brought the brands down a bit. Their competition is is millions of businesses. That's how we get to those targets. But the businesses themselves with the Google data have to work with us for us to get that. And we whittled that down even further because we wanna make sure that we worked with them long enough, so those two years to be able to get that year over year to cancel out the sick the the seasonality. And we had to make sure they've been on our scout product for at least six months because I wanted to know if they've actually been doing the recommendations that we've been giving. So you a lot of the research, I think Arjun's gonna talk about millions, and I would love for the study to say the word millions in it somewhere, but we needed to make sure that we had enough data, over a long enough period, and and that's why we had to cut some of that. Awesome. Thank you so much. The next one is, you know, you basically bold it down, or or you boil all 73 metrics down to a single win rate, right, or that optimization score. Why the one number? Like, why is the one number useful? The one number is a bat signal. Right? I don't think I don't think the one number is the thing that you should of course, that's not the thing that we're going to tell you on a Monday morning what to change. We're not gonna say, hey. You're losing in this the seventieth most important thing, but if you change that, it'll fix your overall win rate. But it's kind of a canary in a coal mine. I think it's pretty easy to say somebody who's losing in half the metrics is doing poor than somebody who's winning in in 90% of the metrics. So for this research, it felt more responsible because because we were trying to do it at such a high level across so many different industries, right, to just say black and white, these look good and these look bad. Because the truth is, I I mentioned from a from a geography perspective that certain certain areas a different metric matters differently. We use this example all the time. We learned, and it's a fun one, but we learned, for example, in financial services that having a headshot makes a huge difference in Florida but doesn't make a difference at all in New York. So those are the types of localized recommendations that for whatever reason, when you're trying to work with somebody in Florida, you wanna see what they look like and whether or not I wanna play golf with them. And in New York, you just wanna see if they have the right services. So there's things like that. Same with a fun anecdote I like to talk about a lot is is hours and hospitality. I think we did some research that said, generally, I would recommend that you fill out every field you could possibly fill out, but sometimes we see in hospitality that having hours actually sends a signal that you might not accept bookings all the time or might send a signal that if I say open now, that you're not. Even if you fill out your hours being twenty four hours, it sends an interesting signal. So, generally, for this research, I just wanted to say they look good and they look bad. If I was going to do it, more specifically, you know what? A business that's only winning 5% of their metrics, those could be incredibly important 5%. So there's probably follow-up research to be done on an industry by industry basis. But, yeah, that that's my long winded answer. No. Thanks for that. That's amazing. And and it, it really, at the end of the day, boils down the signals. Right? It's all signals of trust. The last one that I have is the set it and forget it trap. Right? Why does keeping a case with your peers over time matter as much as the initial build when you first went live with that listing? That's the the original structure. Sure. I've been here for seven years. I think when I first started here, we had we got that question a ton just about listings management in general. Right. Right? Not even talking about updating your photos and updating your hours and updating your whatever. We got we got asked about that a ton when we said, hey. We wanna get you live on the long tail network, and people would say, thanks for doing that. Goodbye. And. I think we learned just from a listings management perspective that it wasn't set it and forget it because publishers go out of sync and, information gets stale or somebody can check-in on Facebook and change their hours with a recommendation, and you needed to make sure that you're up, up to snuff on everything all the time. This conversation's a bit different because it's not just do I need to make sure my listings are are, you know, synced and locked all the time? Of course you do. But the the work is not set it and forget it means that, what was good yesterday might not be good tomorrow. And if you're not checking it, then you wouldn't know. Right? I keep using that example of, you you added a 100 photos and everybody went out for happy hour because that was their goal for the quarter, and then you come back from happy hour to find out that your competition now has 300 photos. So it's not what it's not only not set it and forget it for the work that you're doing, but it's definitely not set it and forget it for the things that you're tracking because you wouldn't even know if you're falling behind if you're not constantly, making sure that those signals are checked. That's right. And every location is different. So what works in one doesn't work in the other necessarily. So, Anthony, thanks for your time. Super insightful again. Let's go ahead and kick it off to Arjun to talk all things citations. Sweet. Hello, everyone. I'm just gonna share my screen real quick. Okay. So I'm gonna be talking about AI citations today. Some of you might be familiar with the research that we've been doing. We've been doing this basically at a quarterly level for the last three quarters since we started getting this data. And I'm gonna walk you through some updates about trends we've seen, signals that have held steady or not, what's changed, what stays the same, and then add in a little bit a little tidbit at the end about what types of pages and what types of sources are being cited more and more often these days. So a quick overview, I know Anthony talked about it briefly, but all of this data is powered by Scout. So this is all public information that we're scraping millions and millions of times over. So we basically continuously scan millions of search results and AI answers, across all the major AI models. So that includes Perplexity, Gemini, OpenAI, and more recently, we've been scanning a lot more anthropic. And then we parse down these responses to, like, individual citations. So, basically, for every single question or every single keyword that we search for, we see what the brands that are, surfaced are and what type of source of information that they are source sourcing that information from. So for the study for q one of twenty twenty six, we had about a 155,000,000 citations across 1,600 brands and, as I just mentioned, the four major AI models that we scanned. Since q three, we've almost tripled the number of scans that we have in the database. This is not just for the 1,600 brands. We wanna make sure that we're being consistent across our analysis to make sure that we're always looking at the same control group. We started out with 770 brands in q three of twenty twenty five, and for the for that exact same cohort, what we've seen is that we've tripled the number of citations. Now a lot of that is because we're also scanning more, but when you actually start to look at the per scan basis, so when we look at for each search or for each scan in these AI platforms, for all of these original 770 brands, we've seen a 28% growth quarter over quarter in how many times they are cited per scan. So, again, it's not just that we're scanning more. It's not that just we're looking for things more often. It's that they're showing up more often when we do look for them as well. A quick breakdown of the different platforms. So we have 53,000,000 citations from Perplexity, 48 from Gemini, 35 from OpenAI. And like I said, we just started scanning Anthropic, last quarter, and we have about 19,000,000 results from that. But what matters is not just how many times, you know, we're scanning and how often brands are being cited. What also matters is the sources of information that they're citing from. So the four main ways that we bucket them, and I'm gonna go into controllability factors in just a bit, is websites. So this is your brand own web pages and domains. Listings or any directories that are public. Reviews and social, so this is like Google reviews or like social posts, and then news and others. So this is like PR buzz that, you know, a brand could be generating or even forums like Reddit, which we classify as, you know, the other category. And so the way we classify these four is what we call zones of control is, we have websites, which are, in our opinions, you know, completely controllable in your brand's domain. Then we have listings, which are generally what we classify as moderate control. But if you have managed listings, like, for example, with Yext or any other vendor, generally listings are fully within your control as well. So for the purpose of this study, listings is something that we consider fully within your control. Reviews in social is something that we consider you to have limited control over. Reason being, yes, you can control how many, you know, you know, review responses you get sometimes and, like, generally the sentiment about your brand, but one or two bad reviews here or there, you know, are not gonna be in your control, and there's always gonna be things that, are outside of your purview. And then finally, there are things that are never gonna be in your control. That's PR news forums, your Reddits, etcetera. And what we've seen is that four out of five, so about 80% of the 155,000,000 citations in q four came from what we saw with the controllable or influenceable, citation sources. The other 20% came from sources that you have less control over. If any of you remember or have seen the research that we did last quarter, I think that number was somewhere in the mid to high eighties. And part of the reason that's dropped is if you go looking back, to this slide, you see that anthropic is at 69% because anthropic has actually a much higher tendency to source from reviews and social than it does from listings and websites. So whereas OpenAI tends to cite much more from listings, and Gemini and perplexity tend to source a lot more of their information from your web pages, Anthropic looks a lot more at reviews, social, and other, sources of information. And that's brought the average down in our study quarter over quarter because we've 15 x the number of anthropic scans and citations that we tracked in the last quarter. Then if you look at the top third party domain, so again, this is excluding all brand owned domains or websites. We wanted to see what are the top 10 platforms that are actually dominating in terms of citation sources. And it's a lot of what I like to call the forgotten publishers. Right? Some of you may or may not have seen our research about, the rest of network where we talk a lot about the importance of small publishers. You know, in the past, before classic search sort of converged at Google, you had a fragmented landscape where there were a lot of different little publishers at play and people would find different sort different information from these different sources. And now even though there is some level of, of a monopoly in terms of the AI platforms and search platforms that most people tend to go to, a lot of times the AI platforms are still using these smaller publishers to validate that information. So even if not a lot of people are going directly to MapQuest, you can see right at the very top that six out of the 155,000,000 citations for q four came from MapQuest because that's where these AI platforms are going to validate this information and make sure that the data that brands are presenting are true, not just on Google, not just on Facebook, but across the smaller publishers as well. And you see the same holds true for TripAdvisor, which is, you know, again, more hospitality centric, or Yelp, which is more food and beverage centric, Healthgrades, Uber Eats, booking.com. And if you look on the very right hand side, we've also we're also tracking how these change quarter over quarter. So while some of them have held steady, like MapQuest has consistently be been at the top, some, like Tripadvisor, rising up the rankings, and some of them are new entrants. So, like, Yelp, Wikipedia, and Grubhub are things that we have seen a lot more citations from in the last couple of quarters. So far, a lot of the websites and the pages we talk about, the brand and local pages, we talk about them as being the largest controllable source, but we haven't really dug deep into which page gets cited specifically. And a lot of that is very dependent on how specific the question that you ask actually is. And so I'm gonna give you a quick example. This is an anonymized mailing and shipping brand, a US enterprise brand, and you can see that they have a location store page. And within it, if you scroll down, you can scroll down to three different dedicated service pages for shredding, notary, or passport photos. Things that they might not be known for primarily, such as mailing or shipping, for example, but things or services that their brand or company still provides. And so the analogy I like to give here is think about the tip of the iceberg. If the top the the visible tip of the iceberg that you see above the ocean when you're on a cruise, think of that as like your brand and local pages. You know, this is the thing that most people will go this is when you go into a Google listing, most often you click onto a website. You'll you'll be taken there. A lot of times, the real citation sources are hidden underwater underneath, and these are what we're gonna classify for today's study as secondary categories. So primary category is something that is pretty normalized. A lot of you will be familiar with Google prime primary categories, which are, you know, pretty much standard. Secondary categories are something that we've taken some time to tinker around with and, like, hone hone in on with a smaller sample size compared to our overall study because it's a bit complicated to actually, nail them down. So for this study, we had 32,000 AI citations that we tracked for almost a dozen businesses across almost 2,000 locations. And what we did was we used the brand and local pages as the control group for comparison, and then we looked at their dedicated service and product pages that we also are are gonna reference as citation pages. So again, two kinds of queries for one brand. The primary category is what the brand is actually known for, so think about in the examples I'm gonna give you, it's gonna be an auto parts company, a pizza company, and a mailing company, and a secondary category is gonna be a narrower service that that company also provides, but it's not necessarily what it's mainly known for. So think about terms like shredding or windshield wipers or wings with the three examples I just gave before. So in the national mailing brand example, what we saw is that when you search for the keyword or primary category mailing, 61% of the time, so more often than not, it's actually citing the location or the storefront page. And that's pretty much expect expected. Right? But what's astounding is that when you start to look at the secondary keywords, the more narrower the high intent keywords like shredding, you see that 84% of the time, it's actually it's actually citing the dedicated service page. When you do that for notary services, it's 86% of the time. When you look at passport photos, it's 89% of the time. So, again, it's a massive shift when you go from primary to, like, then these narrower, keywords or categories. The even though they're still citing your pages which are within your control, the types of pages tend to change a lot. And the same flip shows up in other categories that we're looking at too. So in the national auto parts chain example, auto parts, 83% of the time, is tracking the location page. But when you look at narrower services like windshield wipers or car batteries or brakes, and some are gonna be cited more often than not, 64, 49, and 41% of the time for those three respective keywords, it's still citing the service page, not the actual location page, but the specific dedicated service page. And in the national pizza chain example, 47% of the time, it would track the location page, so a little less than it was for the other ones, and we expect some level of variability across different industries. But when you look for wings, 77% of the time, it looks not at the overall storefront page, but at the dedicated service page. And when you look at this across, the dozen or so businesses that we tracked, we're starting to see these trends hold true across all industries. So we see that for primary category queries, what the brand is known for, 64% of the citations land on the location page. And for secondary narrower services, 70% of the time, it lands on the dedicated service page. And while this research is pretty new, something that we've been working on the last couple weeks, we hope to have a lot more examples and a much more detailed analysis coming out in the form of a research paper too. I know Logan has shared a link to it in this webcast, so please feel free to go in there and keep an eye out for anything that we're we're putting out soon. But, yeah, that was pretty much the the crux of what we've been looking at from an AI citation standpoint. And, Javier, if you wanna ask me some questions. Yeah. Absolutely. I'd love to get the, reactions to MapQuest. Yep. It's still very relevant and very funny. Arjun, we as consumers have choices. Right? So, you know, I mean, like, for example, I like to use ChatIQT for all things personal. I think those of us have been using it for years now. You know, you have been using ChatIQT for the longest because it was the first one available and all of that. But what about for brands? Right? So should a brand tune in its strategy per model, or is there a common foundation that works across all of them? What should a brand do? I'd say, like, lead with the common foundations. Right? So, like, if you go back to what, like, the the graphs that we're showing, websites and listings are still the backbone for every model. That's still, like, out of the 80% code controllability, like, most of that is still from your website and listings. And so chasing per model quirks is it might be a losing game because model behavior can even shift quarter to quarter within the same model, and we have the data to prove it. But if you get those controllable sources right, I feel like that's the number one step you can take to actually starting, help your brand start to show up everywhere. Thank you for that. Alright. The next one. Practic so basically, when it comes to, to what you just said, does this mean that every branch should spin up a dedicated page for every service and every product that they offer? No. I would say prioritize for sure. So building dedicated pages for the high intent services where there's real demand definitely makes sense. And definitely, like, where you're not only where there's real demand, but where you're not the default name for that category or keyword since that's exactly the gap that the data is showing that AI rewards. But spinning up pages for everything might actually end up, like, diluting quality and adds maintenance with, like, no payoff. So if you start to start with a handful of secondary services that customers are actually searching for, I feel like that's the best way to to start out and prioritize. Absolutely. And the last one, if you're a marketer watching this, like, we have many on right now, what's the first thing you'd make, you you you would do on Monday morning? I'd say to the best of your ability, audit which of your pages are currently getting cited for these secondary service queries. So a lot of this is, like, it it it it requires, like, domain knowledge or knowledge of your company that maybe even someone like me or Anthony don't have. Like, you know what your primary and secondary services are. You know what is in demand. You know what's not. So if you start to look at those queries, those secondary narrow high intent service queries, and you start to resolve to a generic location page instead of and you start to see that, a lot like, some of the citations are resolving to, like, a generic location page for those instead of the dedicated one, that's probably, like, the highest leverage fix. Like, that's where you know there's some room for improvement. And then if you know, going back to my initial answer, if you can make sure that your listings and your reviews, you're on top of all that, going back to Anthony's point, your profile is complete and you're doing the best to, you know, make sure that your fill rates are good and your profile is accurate and current, then those controllable third party sources can sort of do the heavy lifting for you. Awesome. Well, thank you for that, Arjun. We do have a few questions here, so folks keep keep, keep them coming. I'll start from the first one here that we had earlier. I think you just pretty much answered most of this, but how can we structure our information slash data on our websites to better rank in AI search? Yeah. I mean, I think that goes back a lot to, like, what we were talking about just now. Right? Like, we're also starting to just scratch the surface of this this this kind of data, but it's not just, like, having pages built for, like, an individual location or for an individual store, but it's also, like, for dedicated service page. Like, for specific activities or services or products that your that your brand can, provide, being able to link out to that from your location pages, the data is starting to show that that matters as much as, like, the location page itself. And I wanna answer one real question. Okay. I think, Javier, you covered the OPI me jump. let me jump in here just to add some context. to the Rachel question. Yep. I am probably anthropomorphizing AI a bit too much, but I when Arjun and I were looking at this stuff, we keep talking about whether or not AI is lazy. So I I think about, the answer to this question the same way I would have answered it ten years ago with Google, I would in with regards to schema. Right? So for example, historically, if you had your hours listed in a long paragraph on your website, could Google find it? Yes. Is it gonna take it some time? Yes. Would it rather not have to look for it? Yes. So, historically, I would have said schema is the best thing to do because your, your traditional search engines are lazy. That's why, you know, when you search for something, it says a billion results in a quarter of a second. So. from an AI perspective, I think of this exactly the same. We want to, we wanna make sure that the lazier they are, the easier they can still find this information. So the answer to these questions today is is similar to the answer I would have given ten years ago, which is make sure this information lives in an easy to digest way. But, also, maybe I was surprised to find out they're not that lazy or maybe if it's in the chat is a better word to use. Because if they were truly lazy, a lot of these landing pages that Arjun talked about that talk about the main business, offering also mentions their secondary service. So if, if this AI was truly, I keep saying it, lazy, they could have went to the main landing page of this mailing company and seen that in the FAQ, they mentioned shredding, But they didn't do that, and they actually checked their work, and they skipped over that main page and went to the secondary page because the secondary page is all about share shredding. So I would love to say that, you know, scheme is the only thing that matters because it makes sure that these things can read it, can read it quickly. But what we learn is that it's like some combination of easy to read, like, schema and, authority. Right? And the best way we've talked about in the past having authority is making sure that you have dedicated landing pages for these things because the dedication tells these search engines that it's not just the thing you do, it is the thing you do. Right? Because having an FAQ is answering the question, is it possible? But having it in its own page is almost sending a signal that it's not only possible, it's something that we're really good at. So, Anthony, that that was basically the structured data, the necessary things on the location page. Now Travis is asking basically the same thing but for your listings. So what are some of the realistic continual maintenance tasks that brands can apply to their listings to ensure they have a chance at a higher ranking. Sure. I a bit a bit morbid, but I like to talk about during COVID how, one out of every three businesses closed in the country. Right? I think there was some crazy stat that about 60% of them stayed closed permanently. Again, a bit morbid, but when we talk about something like that, this catastrophic thing that happens, I you get in the mind of a small business owner, and on one of the worst days of their life, they had to close their doors. They had to, you know, finish their bills. They had to let go of their people, and they also had to call Google. Right? And that's not that's not on the list. That's not something that somebody would think about in that in that situation. So we have all of these listings in the world that end up being what sometimes we call zombie listings. Right? A listing that nobody ever called to Google and said that I'm permanently closed, but they just kinda fade into the darkness. So because of this, because there's so many listings out there that just fade into the darkness, I think what we've seen is that a lot of the metrics that I talk about in the scout findings for you versus your peer have to do with sending consistent signals. Right? So updating almost anything starts to send a signal that you didn't just close and not tell us. So a lot of the things on that bar chart that I had earlier were, age of your most recent photo, age of your most recent owner uploaded photo, age of your most recent review, age of your most recent, review response. So, it's not just you can't there was that counterintuitive dip on one of those metrics slides because you can't just keep adding more attributes. I don't think it's smart to say, the 700 thing that your hotel offers is maybe not gonna add as much to your Google rank as the first thing that you you've mentioned that they offer. But, making sure you keep sending a signal, updating your hours, making sure you you know, if you're closed for Juneteenth, this Friday, make sure that you have told them and updated your hours because these freshness signals start to send not just the the message that you exist, but you. still exist. So continual maintenance, I would just make sure that things are things are fresh. It's not just adding things for the sake of adding things. It's adding things if you have new things to add. No. Absolutely. And and and and I think that's what starts to build that trust, especially within the AIs. Following that, Allison has a good one here from Jiffy Lube. So for reviews, right, longer, more detailed with relevant keywords, photos, authentic versus stock, attributes more relevant than others as it relates to ranking, comparing against competitors, Does quality of these get weighted heavier, or is it true that that quantity and volume rules? So, you know, that also goes into the to the notion of of topics. Right? Review topics. Right? It's no longer about how many reviews, how often are you responding to negative or positive, but what are people saying? Are there different topics within your 2,000 reviews that location has? We, Arjun and I are and, and Javier are in the data data insights team. We also have a data re data science team, and the data science team did an incredible study internally, call it two years ago now, where we think about that secondary keyword conversation that Arjun was just having. We were trying to look at everywhere in the world that you could tell the world that you everywhere on a profile that you could tell the world that you'd do a thing. Right? You could tell them that you'd do a thing by putting it in your name. You could tell them you'd do a thing by putting it in your description or by putting it in your services or your attributes, and then we were wracking our brain. Right? You have the landing page and you have the services or the citation pages that Arjun talked about, but then we have review content. Right? And the data science team started to do research on if you really wanna show up for x, are people talking about it in your reviews, and are you responding to it with those words too? It's the example might be, if you really want the world to know that you do shredding, turns out it actually does make a difference if you have a lot of review topics about shredding and your owner responses to those reviews also mentioned the word shredding. So we started to do the relationship between, Google search term impressions and these specific search terms and and Google Search Console queries and these search terms based on things like reviews. So it yes. We have review in the Scout data with you and your competitors. We do have number of review topics. So I don't think from a competition perspective we go so deep into the content itself, but we do go into the breadth of the content. But within our internal research, we have seen a positive relationship between what is said. And that could be what is said by them or that could be what is said by you. Think about that as just two more places where you can tell the world these specific things. Excellent. Alright. We got three minutes left on those. So, Rachel, Durand again, should individual location pages include relevant related services on them to increase those listings to populate AI citations? I'm I'm gonna keep Arjun, please, add color commentary. I'm just on a heater right now. I I would say, yeah, the it wasn't a 100% for the service pages. Right? It was it was call it 60 to 40 or call it 70 to 30 or 80 to 20. So having shredding in your FAQ on one page wasn't never getting cited. Way more often, it was going to the shredding page, but it wasn't never finding shredding from a different page. So I would say, yeah, having a dedicated place to say that you're an expert in this service or product or thing that you wanna offer is is gonna do a lot of the heavy lifting, but I wouldn't strip it from other places. In fact, I would probably have it in your FAQs or your footer of any other place because it it wasn't a 100. It was just 80 or something like that. Yeah. And, like, anecdotally, we see sometimes the citations come from, like, a random FAQ PDF that you that some brand push pushed online, like, three years ago. So, like, even sometimes, like, chat ChatChippy tier or Gemini will go to the most obscure places to find that information, and it's not it's not the end of the world if you have that information somewhere else. You know? Alright. Thank you. And then the last one here, are services specific pages examined, for example, breaks typically specific to a location, breaks at my brand Main Street location, or per whole brand, breaks at said brand name. So in the examples we were looking at, it's tied to the location. So it's not just breaks at brand. It's breaks at this location at this brand so that you know that you can go to that location specifically for breaks. And, Jesse, one thing to add there, eight, you know, we we've we've done these studies with AI as well asking all the models the same question, within you know, and and every single time, it takes location into consideration before it provides the answer. So I think those were all the questions. Thank you all for joining us today in q two. We look forward to seeing you as well, and our session, coming up in q three, date TBD as well. But, thanks everybody for joining us today, and have a great rest of the week and a better weekend. Thank you. Thanks.