Search isn't just about keywords anymore. Consumers are asking more complex, personalized questions, and AI platforms like ChatGPT, Gemini, and Claude are becoming the first stop for answers. So how do you make sure your brand surfaces — and shows up accurately — when there's no traditional search result to optimize for?
Christian breaks down:
How AI understands and interprets brand information
The difference between structured and unstructured data
Why citations are the new backlinks in the world of LLMs
How location, memory, and personalization shape results
What marketers need to do right now to stay visible in AI search
Plus, Christian shares insights from Yext's proprietary research on over 6 million AI citations — revealing how AI platforms really decide which brands to recommend.
If you're a marketing leader trying to future-proof your strategy for the AI era, this episode is your starting point.
Episode Links
Transcript
Rebecca [00:00:00]: Welcome to the Visibility Brief, brought to you by Yext. I'm Rebecca Colwell SVP of Marketing, and in every episode we unpack what's changing in search and brand visibility. Today we're diving into a topic that has sparked a lot of confusion, which is. How do AI answer engines like ChatGPT and Gemini and perplexity?How do they decide what sources to trust to answer questions? And you may have heard that these models are mostly citing Reddit or Wikipedia or random forms, but our data suggests otherwise. And today, to help us break it down, is Christian Ward, our chief data officer at Yext. Hi Christian, welcome to the show.
Christian: Hi, Rebecca.
Rebecca: Thanks so much for joining us today. let's just dive right into it. Can you explain to me how you and the Yext research team conducted this research?
Christian [00:01:00]: Absolutely. So over the course of 60 days, we gathered from not only our clients, but our prospects that utilize our Scout platform. All of the citations across three major models.
One was Gemini, the other is perplexity, and the last is ChatGPT. In that process, we were looking at roughly 1.6 million questions. Which amounted to roughly 7 million citations. So it's a very broad study. it is worldwide. We have more clients and prospects in different industries and different geographies, but we normalize that and broke it down and tried to understand how is it that we are seeing such a different behavior than everyone else in market.
And it really comes down to, if you speak to any AI about. A, business or a a, product of service, anything, the AI often, or all the time, unless you tell it not to, is going to use your location when it analyzes that information. So the major difference that we see with everyone quoting Reddit as you led into, is really if you're asking like a 00:02:00 top level brand question, it's gonna cite things like Wikipedia.
But if you actually ask it like a consumer, the human being in a location using ai. It always uses location and location then determines what is cited, what businesses are discussed, and what sources it uses. So that's why our research is so different.
Rebecca: That's really surprising result. before we dive into the, specifics, I'm also curious how consistent and reliable the citation patterns are.
So did you see variances between the, citations in ChatGPT versus Gemini versus perplexity?
Christian: Yes. so many of them utilized similar citations. the way that we did it was we categorized every citation and we ended up with roughly nine categories. Then we broke them down by how much control a business or brand can have over each category.
That was really important because in many cases. It's great. Even if Wikipedia or Reddit or let's say a, blog post by a traveler [00:03:00] about your hotel, if that person writes and it's cited, there's not much you can do to control that. You can engage, but you can't really control. so we broke them down into categories of what can you fully control predominantly your websites or your local websites, the directories, which you mostly can control through platforms like Yext or a citation management approach.
Then there's reviews in social. Which you can engage with, but again, you really can't control what everyone's saying. And then lastly, this other group which is like forums, like Reddit or government, posts. And so when we did that and we looked at that data, the reason we did that was did we find those patterns where different AI were behaving similarly?
And while there are some really interesting callouts by industry and geography. Inside of the typical industry or geography, you saw a lot of the same patterns where the AI was repeating, doing the same approach. We also saw that depending on things like Gemini. Gemini pulls, data from Google Business Profile, but it doesn't cite 00:04:00 itself.
So you'll see things where there are little nuances in the data, that it's critically important to understand that the way the AI cites things and the way it uses it is all captured in this research. And so we, spent a lot of time looking at this and cutting it different ways. And now we're even working with other industry participants.
To dive deeper into subcategories, things like law firms or, certain doctor types or healthcare, specific areas. So this will continue to evolve, but the main pattern that we found was that roughly 83 to 86% of everything cited is almost fully controllable by brands. That's very different than every other research out there.
And that was really the finding of their websites. And these directories, whether industry specific or more broad directories, those are the ones that really popped up.
Rebecca: Wow. that's really encouraging because that means that I, as a marketer, I have a lot more control than I thought I did. initially.
It's also curious [00:05:00] that, you were able to uncover the sources that AI sites, there isn't really a way of doing that on Google.
Christian: yes.
Rebecca: SEO feels a bit more of a. A black box, let's say.
Christian: yes. That's, it's such a great point. so classically, Google is what we call more deterministic, right?
It's rules based. many industries have popped up trying to figure out the rules, when the rules change, what you have to, what side of the ship you have to run to, to get the result you want. now it's a little different where with ai, yes, while it's a black box and there's a great deal of variability based on things like the model you choose.
The memory that you've created with the AI as well as, where you are. But if you do take those things and hold some of them constant. You get this great optimistic view versus what I think all the, our research we've seen prior to our research has been saying, which is, Hey, you need a Reddit strategy, good luck.
That's not, we found it's Reddit and forums like that. They were two to 3% 00:06:00 on average across all industries of actual citations. Once location comes into context, and I tell all, everyone watching, you really should go try this yourself. Just walk outside, open chat GPT and ask it about dentist. It's not gonna cite Reddit, it's gonna cite MapQuest and maybe, some of the industry directories.
It's gonna cite things that, yellow Pages, all these sites are right there, and all you have to do is hit the citations or the sources bubble and it'll pop over. It'll tell you all of these, so that's what we're doing instead of the high level brand. if I ask what is a dentist? Then Wikipedia is probably a pretty good source, but if I ask for a dentist and the AI knows where I am, changes everything and that we think is a far more likely engagement, meaning the AI is gonna know where you are and what you've been look looking for in the past.
And so that's what we really focused all this research on.
Rebecca: So interesting you threw out a couple of. Sites that I haven't heard about in years. Math test, yellow Pages. Okay. So what does this 00:07:00 mean for me as a marketer that's trying to get my brand to show up? Should I be making sure I'm listed on the sites?
Christian: Yeah, it's, I think it, sync those a couple things. First off the, I will say their websites. The things that popped up first, but that was typically for branded questions. So when something's branded, if I know your brand and I ask a question that a lot of times will show your website directly, and it should, that we, hope we're not rewriting all of digital marketing here.
We hope that your website is still the core asset to pushing that information out there. But what comes right after that in every study by every model, was actually the directories. And what I think is happening there is in our study, we broke down. The types of questions people ask. So we broke it down by branded questions and unbranded questions to see what the behavior was.
And then we actually broke those two categories down further into, objective questions and subjective questions. Now that sounds pretty granular, but the reason we did that is because if I say best 00:08:00 dentist. In my town that accepts my insurance. That sort of conversational interface, the word best, turns into a subjective query where there might be a lot of people for that insurance saying, this is a good dentist.
This isn't If I instead said, what time or are they accepting new patients? That's an objective fact at a given moment in time. And so you get very different citation patterns. So one of the fun parts about this research is that you can dive into each and see what is working and what's not. What again, though, in all categories, and this is in the research report at yext.com/research, but people can actually see and read the report where we break down all these categories.
And one of the most important things, again, to, point out is directories. One of the reasons we think this is happening, these older directories that, again, I confess, I don't normally use MapQuest anymore. Remember we used to print the, map and put on the car seat next to you. I don't use that as much anymore, but what I can say is the, reality is, these are great sites with structured data, and the more remember.
[00:09:00] Probabilistic models are what these are all built on. The more you see probabilities that this address with these hours, at this location, with these services, with this menu, all of that, the more times the LLM sees that you are increasing the probability to that LLM, that's the right answer. And so what we're seeing is, it's almost a resurgence in many of these platforms.
Again, I ask everybody, go try this. Look for a place to eat for lunch, and you're gonna see all these old directories. That have given the LLMA great deal of confidence on a probability basis that this is the right data.
Rebecca: Amazing. to wrap us up today, what are a couple of key things that marketers should be thinking about right now?
Christian: Yeah, look, first off, one of the great things about this is there's not two internets. so everything that the AI is being trained on and being grounded on is the same information that is also, powers, SEO, and Google, visibility. It doesn't mean they're the same, 00:10:00 it just means that the core process by which you are updating information should continue.
But just like classic search AI follows a very particular approach and, really it's around information quality. And what that means is the more knowledge you provide, so the depth of the knowledge, the more consistent it is in as many places as possible. The more frequently it is updated and there's a little hook there.
Do not update certain things and not everything at once. You need to update everything because if you're frequent but not consistent, you just shot yourself in the foot, you are always trying to send one consistent, strong signal with as much data as possible. The big difference I would tell marketers is I think they have to take a step back and really think about how do they link all the additional knowledge.
Using local websites and highly structured websites that are very attractive to large language models as they're parsing through information, you wanna think even deeper. So things like 00:11:00 offers, discounts, loyalty programs, menu items, allergens. The more knowledge you let an LLM find, the more likely that's gonna find its way into the ai.
And so that's a broad area to speak, but that's, it's really focusing on more data. Frequently updated, but perfectly consistent. Otherwise you break the probability. So those would be the main things I'd focus on.
Rebecca: Amazing. Oh, those are fantastic takeaways. And I really appreciate that this, feels like something actionable that I can do, that marketers can do. And it gets back to the basics. think of every question that someone might have about your brand and answer it, and answer it in a way where you are helping. Ai, I hoping LLMs answered it on your behalf. So that's it for this week's visibility brief. Christian, thank you so much for joining us.
Christian: Thank you for having me.
Rebecca: Oh my gosh, yes. If you like what you heard, subscribe and share with your teams and we're available anywhere you listen to podcasts and head to yext.com/research to get [00:12:00] the citation report that Christian was talking about. We'll be back in a few weeks with a few more insights to help you stay ahead.
Thanks for listening.
