The 7 Things Marketers Need to Know About AI Search

Hello, hello, hello. This is Sandy Carter with The Marketing Companion. Nope, you're in the right podcast. Mark Schaefer has passed over the reins of the podcast to me and it is in great hands. But don't worry, Mark will also not be going far. He'll be a guest on our show here and there as well. For those of you who don't know me, I am a marketing leader. I've had marketing leadership roles for the past 18 years. I was CMO at IBM WebSphere as well as Tivoli, and I ran sales and marketing at Lotus. I am currently also having CMOs report to me both at Amazon Web Services and Unstoppable, where I am today. I also know a lot about AI. I wrote a bestselling book called AI First, Human Always, talked a lot about marketing use cases with AI, which is what we'll explore in this podcast. And I'm just thrilled to be here with you today. So let's go ahead and get started. We'll keep it real and keep it going. Okay, we are gonna start out talking about something I think really matters. And in fact, SEMrush said that it's projecting AI search will surpass traditional search by 2028. That's right around the corner. I think it may even be sooner than that. And to prepare us to figure out how to work with the number one topic coming up in 2026 is Andy Crestodina. Did I say that right?

Basically, yeah, more or less. Crestodina. It's so long I stumble over myself. Close enough.

Well, before we dive in, Andy, let's talk about our sponsor. Our sponsor is SEMrush, and that stat I just mentioned, it came from their research. They have this really cool AI search tool. You can track your AI visibility score, which is a single benchmark of your share of voice. You can discover trending prompts. And here's a big one. You can benchmark against your competitors to find out the exact prompts that show you how to show up in AI. So if you're serious about GEO, this is where you can start. So check it out at semrush.com/ai-seo. Okay, Andy, I mentioned GEO in the intro. I hear so many people talking about it. GEO, SEO, tell us what those words are.

You know, it's hilarious to me that there isn't a settled upon name or acronym for this yet. I hear, I think GEO might win. Maybe we could start with that prediction. But this is the evolution of discoverability. And it's the art and science and the new marketing channel of getting our brands to appear, our content, our key pages, uh, in AI responses.

That's the game. And what does GEO stand for?

Search engine optimization versus generative engine optimization. AEO, answer engine optimization. Some people are changing SEO to suggest that it could mean search everywhere optimization. All the same thing. I hear weird minor distinctions made between them, but we're basically talking about the same thing. It, it's getting AI to, show us, getting AI to recommend us.

Yeah. You know, the other day I was reading a report and it said that most senior executives now go to ChatGPT or Claude or Perplexity, you name your tool, and they say, hey, I'm gonna search for a CRM tool. Give me the top 5. Here's a criteria I'm working out. And if you're not in that top 5, you're gonna be left out. And in fact, you just wrote about a stat that made me do a fast double take, which was ChatGPT traffic converts at a higher rate than Google traffic. So can you walk us through that? Like, how does this reframe how we should think about marketing, especially around search?

You just asked the perfect question, Sandy Carter. I love it. Okay, so, so, uh, uh, I'm the co-founder of a digital agency that builds websites and optimizes websites. So I have access, and we've been around for 25 years. So I have access to literally hundreds, something close to 500 Google Analytics accounts. I'm inside and outside of these accounts all day, every day. Very, it's typical for me, I'm used to seeing it, I expect to see it now. The so-called key event rate, as in the conversion rate from visitor into lead, form submitter, contact us, B2B lead gen, is higher for ChatGPT than from Google. Why would that be? And you already said it. It's the executive saying, I want to see the top 5, and then here are my criteria. I want one of these options, vendors, partners, software tools that check these boxes, that are good at these. I have these requirements. This is my context. So what happened there is maybe not yet obvious to everyone, but Google always just provided options, but AI is providing recommendations. That is wildly different. It changes the game. And so now visibility, even though everyone's talking about AI visibility, insufficient. If AI is not only making you visible, but also putting you at the top and showing that you're a great option for those specific criteria, for whatever that executive's searching for, then you may be visible but not recommended exactly. So this is— it's a fundamental shift. But the ways to adapt maybe aren't that difficult.

Interesting. So I still hear from a lot of marketers, look, Sandy, I can't give up SEO. I've got to keep doing SEO. So now are we at a state where marketeers have to think about SEO and GEO? We have to think about both of them. And when do you think it would go back to 1, or will it ever?

Well, for simple prompts, AI can rely simply on its base training. It ingested the internet, short simple prompts, it just gives you the answer. But for that more complex prompt that you gave it, an example at the top where the, where the, the CEO looking for a CRM is putting in certain criteria, it searches, AI searches it starts doing searches. So SEO is still 100% alive. It's just that in many cases, AI is doing the querying on behalf of the user. So if you have, uh, if you're irrelevant and invisible in traditional search, you are also likely irrelevant and invisible for, uh, in the responses when these more detailed prompts are put in by your future prospect. So SEO is alive and well. It's just one level more abstracted because the AI is frequently doing the, and making up, you know, generating its own queries on behalf of the user.

So I know when we first started looking at this at my current company, one of the interesting things I found is when I went to these LLMs, they were pulling stats about my competitors from places like Wikipedia and Reddit. And I don't know, Andy, if you can believe it, but at the time we had no Wikipedia page. So the first thing I did to kind of hack the GEO logic was I put up a Wikipedia page. And then the second thing I saw was that there were lots of quotes and everything from Reddit. So I actually hired a person to focus on Reddit for me. We created our own subreddit and then we also started participating in other subreddits, right? Like AI or for us, you know, blockchain and cryptocurrency. And that seemed to have elevated us a lot. That seemed like two kind of quick hacks. And that's kind of what you mentioned. Maybe it's not quite so hard. What do you think about those two hacks that I talked to you about? And give us a couple of other hacks that people listening in could maybe implement right away.

Well, there are two ways to influence the likelihood that AI recommends our brands. One of them is by updating our own content because AI is training on our own websites. And that's like on-page SEO. I've been in SEO for 25 years. I love search and I'm very familiar with keyword research and semantic SEO and SEO copywriting. But what you just said is critically important because AI isn't only training on our web content. It's sourcing all kinds of stuff all over the web. The trick though, and you did it and I love it, is that you looked first to see where it's sourcing answers about you and your competitors. So I think it's also sort of strange that there's a lot of people out there that just say, oh, ChatGPT cites Reddit, I gotta be on Reddit. Wait a minute, do you have any evidence? Where's the evidence that it's citing Reddit for your category? If you have evidence that AI is looking at certain sources, for us, we're a B2B service provider, it's like Clutch, you know, for a software company might be G2. A lot of it just depends. So what I would recommend doing is Talking to the AI, asking it to make a buyer guide, asking it about you and your competitors, and then asking where it got that information or looking at the searches that it did. And that is what you need to do first before you know where else to do what we would have called offsite SEO, uh, not link building in the traditional sense, but to be present and have positive sentiment in those places, which by the way, our sponsor Semrush has the report. It does that. It has sentiment analysis. Yeah, Semrush AI tools show that. But, but it— let's be strategic about it, which is what you did. You first had discovered that there's reasons to believe these are the places, and then you took action to be active in those places.

Oh, that's great advice. Yeah. Yeah. So I guess my, you know, maybe telling everybody go to Reddit, go to Wikipedia, may not be the same by category as your point, right? Yep.

Every niche has different ones.

Yeah, it's quite, it's quite fascinating. The other interesting thing, you know, you talked about prompts, like doing some prompt reverse engineering, and it was so fascinating, Andy. So on my marketing team, I have a bunch of people who've been used to doing SEO. And so when we first started looking at some of these tools, the first thing they did is they'd go put in a keyword. They'd go put in, you know, hey, I wanna search for everything that matches this keyword or this keyword. And not a prompt. So I'm curious, you're, you talk a lot about prompt reverse engineering. What do marketers do to figure out what prompts people are actually using when there's really no search volume equivalent, right? So what do you do? Right.

This is also one of the biggest challenges and why SEOs are gonna be kind of frustrated in this era. It's hard. What I recommend doing is to first build for yourself a detailed persona. That includes the decision criteria that your target audience uses when making a decision in your category. And when you understand how they make decisions, what they care about, what they need, what they trust, you know, what they're hoping for, their hopes, dreams, fears, their concerns, that is the best input for insights into how your prospect is likely using AI. I need a CRM that connects to these systems. You know, I need something that has this level of security. We could keep going with your CRM example or anything at all. So it's, it's simply just, it's an act of empathy. There's a lot of gray area. No, there's no keyword difficulty. No, there's no search volume. Uh, but if you know your audience well enough and good marketers all should, then you know that these are the things that you should be surfacing and talking about and bragging about and showing. And you'll quickly conclude that, unlike traditional SEO, which was just about traffic and ranking and visibility, that now we have to inject into the AI training data proof points, evidence, comparisons, the details, the reasons to believe, testimonials, positive sentiment, reviews. And that is traditionally just conversion optimization. I, I'll, I'll try to summarize. SEO, traditional SEO method for accessibility, uh, conversion optimization methods are good for AI recommendations. So, as this is a new territory for SEOs, to make sure that our content is filled with proof, data, statistics, testimonials, evidence, uh, that AI trains on that, and it's going to reference that when deciding whether or not to recommend you and your CRM to that CEO when they go, uh, ask AI for, uh, you know, for a recommendation.

Awesome. I really love that. Um, so, you know, a lot of what you were talking about is really about nurturing leads. And speaking of nurturing leads, let me tell you about Brevo too. Brevo is a complete CRM suite We actually use it. We actually use it to really develop and build meaningful customer relationships at scale. So whether it's email or SMS or even WhatsApp, which I love too, gives you the toolkit to turn those AI-referred visitors into long-term customers. So if you look at it, I think it's really fascinating what they're doing. You can actually get started for free and you can use the code COMPANION for 50% off your first 3 months, which is Magnificent. Now, let's go back to some of the questions in the interview. We were talking about nurturing leads. You know, Andy, the other thing I noticed though is that it appeared to me that the different LLMs look at things differently. So when I was looking like at sources for ChatGPT versus Groq versus Perplexity versus Claude, it looked like they were looking at different things. So now as a marketeer, Am I now looking at not just one, but multiple sets of prompts that I have to look at?

Yeah, it's, it's gonna be hard for all of us to try to adapt. It's not like Google and Bing. I mean, they're not wildly different technologies. They are all, you know, giant probability matrices trained on the internet. They're all trained differently. But I don't think that we will— that you'll hear a lot of case studies in 2026 of marketers who are successful in doing GEO for one versus another versus another, because the methods— but, you know, we have to keep in mind, you know, search was always deterministic. It's an algorithm. It literally is ranking things. AI is probabilistic, and that is a completely different paradigm. So knowing that it's all a bit fuzzy anyway, I'm not sure that we'll come up with, uh, you know, completely different methods. But I've seen the data, and it's interesting. Uh, I read that, um, uh, Claude, compared to ChatGPT, Claude is just very unlikely to cite current sources. It just doesn't ever mention Associated Press or Reuters almost ever. It's like ChatGPT is like 100 times more likely to cite those. So there's, there's major differences in the models, but the marketing strategies that drive outcomes in each, I don't think will be so different that we need to get that granular.

Mm, that's helpful. I noticed for my company when we were looking at GEO, exactly the point you said. In fact, it was pulling, we were looking at support because support is one of our competitive advantages, and it was pulling reviews from 4 years ago. And we were getting frustrated because ChatGPT is looking at the new Trustpilot scores, everything new. Claude was looking at stuff from years ago, and we had to go back and address that, right? Answer stuff that we hadn't answered 5 years ago, 4 years ago, whatever. So anyway, I thought that was very fascinating. So you've given some really amazing tips. One of the things I read, though, that I really loved was your bookmark trick. You know, where you can actually see the grounded search queries that ChatGPT or Claude or others use. And that just blew my mind. Can you explain? I think you called it a query fanout, if I got that right, and why traditional SEO still matters in that AI-first world.

Yeah, query fanout is a term used in a Google patent. It's a real thing. It's a term of art, and it's, it's part of how search technology works. You enter the initial query and then the search engine interprets that and expands upon that and searches for multiple queries and then summarizes them. So that is a technology that, uh, we can adapt to by spreading out with a broader keyword strategy. When you enter a more detailed research prompt into any of the AIs, it then extrapolates multiple queries from that and then goes and searches for several. Sometimes 10 if you have it in thinking mode, it might be a lot. In ChatGPT, actually, the browser caches— it stores what those queries were. So there's a little hack, it's a bit of JavaScript. You can add the JavaScript to a bookmark, making a bookmarklet that exposes the underlying queries. Uh, it's fun and weird and cool to see how AI searched on your behalf, but I think that there will be better ways to do that. Uh, I've already been experimenting with Chrome extensions that, that have, uh, similar features. It's a little cluegy. It's, it's a, it's a straight— it is a hack, uh, but it will— it, it gives you strong clues into what key phrases are relevant in using traditional SEO methods, because that's literally what the AI did to go find, uh, which companies to consider when then coming back and making a recommendation. So yeah, it's, it's searching. You can— and, and if you want, you can use a bit of JavaScript to see specifically what it searched for.

That's pretty mind-blowing, actually, that you came up with that. And if I'm a marketeer and I really want to do that, how would I do that?

Oh, it's like a GitHub post that has this huge piece of JavaScript. I don't know. Probably, you know, there's a link on that article. It's at orbitmedia.com/blog. It's called— the article is called Prompt Reverse Engineering. And I link to it there. But I think I'm hesitant to recommend. If you want to play with it, it's fun. Okay. But I'm looking for, and there will likely soon be, good Chrome extensions that do the same thing. It's just referencing the bit of memory inside your browser and exposing that after. It only works for ChatGPT, by the way. You can't see what Google's doing. Gemini, you know, there's no exposing of underlying queries in Gemini and Google. But, but yeah, look around or find that article and you can, you can test it yourself. It's super interesting.

It is interesting. So you mentioned earlier, you know, we talked, you talked about like case studies and testimonials and, you know, brands need that. Obviously, when I was at Amazon, we were always looking for case studies. In fact, our mantra was, you know, we were customer obsessed. So instead of us talking about something, we always had a customer do that. But I think now they're even more important, right, in AI to get AI to recommend, not just mention. So kind of give me like a minimum viable proof stack that a brand must have today on their website.

Well, as a general rule, we should always try not to make unsupported marketing claims. So if you've got a page that says that you do X, Y, and Z, or you're number one, or you're great at this, or you're on time every time, that is not sufficient to either convince a, a persuasive visitor or train the AI. So ideally, when you make a claim, you have supportive evidence. Case studies are excellent for high consideration B2B offers where there's big long sales cycles and multiple decision makers, and you've got stories that go into the details about how you solved a problem and made an impact. Sometimes it's just Amazon, you mentioned, it might be a you know, reviews, test— you know, there's all kinds of— I, I have a list of like 13 different types of evidence you can add to a page. It's like impact data and, uh, you know, testimonials, reviews, years in business. Team credentials is a good one for professional services. But case studies are fantastic. They're useful in all kinds of ways. And yes, the AI can train on them. Just make sure that when you link to a testimonial from a service page, that even if the person doesn't click the link, they can see that impact. So put the data, the most compelling soundbite or data point from the testimonial onto the page itself because we can't really trust that the AI will crawl everything. So I would surface and make visible on these key pages the reasons to believe because we can't really assume that AI is going to crawl all, you know, 600 pages on our website and read everything.

Yeah, that's, that, that is interesting. Now I've also heard that, and you tell me if it's true, this might just be, you know, you can't believe everything you read on the internet, that if you do like the 5 reasons that customers love your feature and you have 5 pieces of customer testimonial or proof or evidence, you'll do better than if you just have like random references throughout your site. Is that true? Does, does AI like numbers like 4, 10 of, 11, whatever? Does it like numbers better like that?

Yeah, I think that there's a benefit to having content that is easily crawled and scanned and parsed out, understood. So having short and sharp value propositions, for example, or good formatting like lists, it's gonna extract that meaning. It's, it's, it's, it's language-based, so it's going to extract that meaning more easily with good formatting, which is basically what you said. But I think even more important than that is the context and the relevance. So if the, if the user, you know, we want Brevia to be recommended, the CEO's asking for recommendations. If the CEO said, I need one that does this and this and this, then your proof points should be relevant to those, the decision criteria of your future prospect of the person prompting the AI. So yes, let's make everything scannable, let's make everything concise, let's make it nicely formatted, let's add schema markup tags and, you know, all the, all the things. But I think it— these are language models, so the, the relevance of the proof point to the decision criteria of your prospect probably is the most important consideration.

Fascinating. Okay. So if I'm listening, Andy, to you today, and I'm getting a little lost because I am a marketeer who's really done zero AI optimization work, give me one thing I should do this week to start training AI so that AI recommends my brand.

My top tip right now is to go to AI and ask it to make for you a buyer guide for your category with your brand and 3 or 5 other brands listed out as rows, and a column for pros and strengths, and a column for cons and weaknesses. Because you can actually look directly into the training data and see why it doesn't love you. If you know, once you do that And then maybe follow up with a prompt for like, where did you learn this stuff? You'll know what to write about to remedy those things and to correct any misinformation inside AI's training data, which proof points to add to bolster your claims, which places to be active. You gave great examples. So I think that's the beginning. Pretend to be your audience and go ask AI to make a buyer guide for your category within a table format with a column for pros and a column for cons. That is way more fun than SEO ever was because you can see directly into the training data. And I think that, that will, for clever marketers, that will lead to tons of ideas.

And do I use the tool of my choice? Do I pick Gemini or Grok or ChatGPT, or should I pick one that would get me the best kind of overall view?

Well, if you're emulating the experience of your target audience, then you should focus on their preferences, not yours. I love what I love, but I know that, you know, Google and ChatGPT are by far the most popular. So I would do it in Google's AI mode first because Google is still 370 times more popular than ChatGPT. And I would also do it in ChatGPT. Another thing you can do, like, it almost is— it might even suggest that it does it for you, but after you get this little buyer guide, have a conversation with it about how to increase the likelihood that you get recommended by AI. You know, what's— what are the gaps? What's missing from my pages? You know, how could I, uh, you know, why didn't you recommend me for that? Why do you think that I'm not great at this? But, uh, as for platforms, I would just simply focus on the ones that are the most popular because they're likely the ones that your target audience is going to be starting with.

I love that. So using AI to help you with AI, essentially.

It works. Yes, you can. Absolutely.

That's pretty wild. Okay. So I'm gonna ask you one other question. I, I love stories. I like storytelling. And you work, Andy, you know, you've got this great agency. You may not be able to tell us the customer, but could you give us like a great story about someone who started out looking at GEO and then they got this result? Hmm.

Well, here's one that's great.

Like your own preference, right?

Like, yeah. Will Reynolds is the founder of SEER Interactive. SEER is an SEO, PPC, AEO, GEO agency. And when he saw this coming, he just knew that he wanted to be relevant. And so he just shifted his content strategy to start talking about AI all the time. And he said it took 2 years of speaking on stages and podcasts and writing and recording and just And now when you talk to AI about, you know, agencies like his, he shifted the conversation, he modified his topics, he adjusted his content strategy, and eventually trained the AI to believe that they are highly competent in, you know, for this topic, this keyword, this channel. So Will's a fun example. He's also a super fun presenter to, to see if you know Will Reynolds. He's in Philly. Yeah, he is.

He's good. Yeah.

Yeah.

So, uh, that's, that was 2 years you said. Do you think it takes 2 years on, I mean, like how many, how long before I as a brand, as a marketeer start seeing results?

I'm thinking about this a lot and there isn't a definite answer, but if you, if you give Google's AI mode a detailed prompt and it goes and searches, then the speed of the changes are as fast as the speed of the changes in search results. So Gemini has, is very quick. All search engines have been fast for years and there's not this long knowledge cutoff that you have with ChatGPT and an even longer knowledge cutoff that you mentioned with Claude. You said it was citing things years old. Yeah, that's the knowledge cutoff problem.

Yes, 4 years. Yeah, 4 years old. Yeah.

Yeah, so I would say, uh, you know, AI is going, uh, Google will be crawling your site within days. Uh, the difference in search results might be in a week or two. The difference in AI mode recommendations might be in a couple of months. ChatGPT, maybe twice that. I don't know. We're not sure. It's not really even knowable. Hard to tell. I've never seen a research piece on it.

But these—

we're all playing a long game here now. So it's like go to the gym, eat your vegetables, and train AI to recommend your brand. There isn't one action that you'll be able to point back and say that was what I did that worked. But these are just things that we all need to be doing starting now and forever after.

Well, Andy, thank you so much. I mean, this is why I love this podcast, because you've just shared such incredible knowledge that would have cost me thousands and thousands of dollars of consulting to get. I love the buyer journey guide that you recommended everybody go do. I'm gonna go do it. I actually, actually, after we hang up from this podcast, I'm gonna go try that out because I think that's a great way to see what AI is recommending about your brand and what it's finding that people don't like. And it may be a big eye-opener for you. So thank you so much for your time today. I really appreciate it. And to our audience, don't forget to try out Semrush. A lot of these tactics and tools and things that Andy's talking to you about, you can see it in that Semrush AI. SEO tool. We actually use Semrush in my company and we also use Brevo. Don't forget that code Companion and you can get that off too. Well, thank you guys very much for listening today. Andy, you blew it out of the park. Thank you so much. I really appreciate it. Give us feedback, send information. We want to hear what you had to say about it and what else you need. Thank you so much.