Patrick Stox: GEO? AEO? LLMO? What's With All This AI SEO Stuff?| Ahrefs Evolve 2025

Hello, hello, hello. Welcome to Href's Evolve. Guess who gets to introduce AI to everyone here that is ready to evolve. That's me. I'm Patrick. I'm going to make this work somewhere. Okay, cool. Uh, I'm going to skip this, but I will say this to the day marks six years for me at Hrefs and I am honored to be at such a great company. So many great co-workers, such a great product and I get to help shape that product. It's awesome. What are we calling this? It's a geo SEO search everywhere optimization seemed to win in some polls where I won where that I ran. Uh, I was really hoping EIO would win, but you know, Old McDonald had a farm. E I E I O. The joke didn't land. [laughter] I'm calling it SEO for AI search and AI assistance. That's way too long, but that's technically what it is. Just call it marketing. It's much better. It is marketing now. Always has been. Google dominates search engines. Chad GPT dominates AI assistance sort of. You take into account AI overviews. We can't break this data out because Google doesn't give it to us. They suck. Uh but the reality is lots of people search on Google. Now we have AI mode on top of that. Also not broken out. Google really sucks at this stuff. You think they'd be better at data. They just don't want us to have that data because it looks really bad. The AI traffic is growing almost 10x over the last year, but that's pretty much all just chat GPT. Chat GPT is dominating that market. Google is also growing uh like 20% in the last year and way more searches than Chad GPT. Almost 400 times the amount for traffic. I don't know if you have seen this yet. Chad GPT versus Google. It's an awesome website we set up. This is data from HFS web analytics. Uh trying to remember how many were in this data set like 60 70,000 websites. Google sends like 41% of the traffic to these sites. You take all the AI assistants combined and it's like.3%. [snorts] 95% of the attention right now in our space is here. The traffic is still here. So SEO is the worst it's ever been and it's still your best marketing channel. I love this quote from Ryan Law, my coworker. It's very true. SEO is not going anywhere. But things are changing. We will see AI search become the default. AI mode will be the default in Google within a year or two. Then we're all just going to call it search again. It's not going to change. It changes a little, but like it's still search. We'll figure that out. There's a lot going on. We had the great decoupling. This is fresh data. That's what it looked like across 800,000 websites. This was a real phenomenon. It wasn't made up. [snorts] And it is the AI overviews that cause that. They're on about one out of five queries now. One out of five. They're dominating. They're basically getting rid of featured snippets. Like, this is the new system that's just going to replace And they're reducing our clicks by a lot. That 34 and a half% was data pulled back in when was that? March. Um, so fun fact, since then there's about four times as many AI overviews. We need to run update this number because I guarantee you that is much higher now. And AI mode will be worse. I ran a bunch of data models to spare y'all. Terrible math. Uh, lots of projections. Clicks are going to be about one quarter of what they were. That is the new reality. Not yet, but in a year or two, that will be the new reality. And so SEOs are like, "We're not going to report on clicks. We're going to report on impressions." That's dumb. No, you're still going to report on clicks. It's a better metric. If you want even better, tie it to money, revenue, SQLs, MQLs. This is what matters. I don't care how many people come to my website. I care how many people buy whatever I'm trying to sell. How many people sign up for HFS? If it's less people that come to the site and more signups, great. Perfect. Less resources used on my end, more revenue, awesome. Google and being keep telling us the traffic quality is better from AI search. Our data was like not really. I think the one metric that won was time on site, but every other metric was lower until we got to signups. For us, this traffic was half a percent of our traffic and that made for 12.1% of our signups. That is 23 times higher than traditional search. That is insane. This is highly qualified traffic. They're basically people are asking uh chat bots and like, oh blah blah blah, what's the best for this? And then they choose hrefs. So, a lot of this top of the funnel is gone, but that bottom of the funnel looks great. Really great. This is also brand new. I still need to publish this. Uh, we've been talking mobile first for a long time now, more than a decade. This traffic is not mobile first. It is desktop first. 65% desktop for AI versus 20.9% for traditional search. So start looking at your desktop website again if you stop doing that. I know I kind of did. It's mistake. Should look at both. The traffic is also different. Again, it's more qualified. People are lower in the funnel. So we saw a lot more on like product pages, homepage, but even ourformational content actually performed better as a percent of traffic in AI search. Stuff like free tools did not perform well uh because people are just using the chatbot, the AI assistant to like build their own tools to run this stuff. So things are changing a bit, but still looks good. International pages are not doing well in these systems. There are a few reasons for that. One, it's a lot smaller corpus. There are less pages, so English is more likely to be cited. But that doesn't mean don't make these. If you don't make them, you're not going to be cited at all. Still do that, please. It works. And original data, especially for these markets, uh, is working really well. My colleague Eric will be publishing some stuff on these on the HRS blog pretty soon about this. It looks pretty awesome. All these systems are citing different websites. They have different partnerships. They're licensing data from different people, different magazines, blah blah blah. They're making different decisions. We've been spoiled. We basically just optimized for Google and we're like, "Well, that's good enough for Bing." It may not be the case with all these new systems. They're making different decisions and they have different biases. Again, they're biasing more towards like things that they have access to, things that uh they think are better sources. So, it's the wild west. We're back to you got 20 different search engines, blah blah blah, and they're all different. It's hard. [snorts] So far, the brands are still winning in Google. They've had a bias towards them since, I don't know, 20ou probably pre20. There was like a VP that once said uh they were talking about cleaning up spam and he said uh brands are not the problem, they're the solution. These other systems haven't really figured that out yet. There's a lot of spam in those systems. There's a lot of nonsense results that aren't good. Let's talk a little bit about how these AI assistants actually work. It's a next word predictor. Over and over again, it predicts the next word. It looks at the context, what's been said, and just decides over and over what is the next word. And that sounds good to people, but that also makes it hallucinate. [sighs] There's multiple probabilities involved in this. You have the probabilities in the training data. So, this is all the data that they got from Common Crawl, Wikipedia, whatever newspapers they licensed, etc. uh all that is one set of probabilities that it's using to predict the next word over and over. Then you have the rag model. So this is retrieval augmented generation. All that means is they go and they pull a bunch of related pages and then they look at the probabilities on those what's being set on those pages. And then they have what's called temperature which just adds more randomness. So it's randomness on top of randomness on top of randomness. But if you know what's going into these, if you know that this is the probability that the next word is this, this is the probability it's going to say this in these different systems, you can influence that. This is modern SEO now. And you have topic coverage. So we've look, we've seen query fanouts is basically doing related searches. What are related things people want to know? You've got chunking, which no one seems to understand. I just want to point this out. Okay, everyone thinks like the chunk is this. I'm gonna go from here to here, right? No. What topics are being covered can actually be in different parts of the content. So, this is from AI content helper and we added this highlight function, but this basically works the same way that we do it in our own search engine. Yep. So, these chunks can come from different parts of the content. It's whatever content is related to that particular topic, that is your chunk. It's not I need to write an H2 and then that is my chunk underneath. That doesn't work that way. Why that matters is because people are writing stuff that no one wants to read. It is annoying. 50. Imagine like a page with 50 FAQs. H2 H2 H2 all these different things. Paragraph two paragraphs under. I've tried to read some of that content. Unfortunately, some of it is working right now. It is not future proof at all. It annoys me as a user. I'm sure if y'all read this, it will annoy you, too. Don't do that. You also have personalization in all these systems. They're taking into account your own search history, your preferences, what they know about you. Their bots are not rendering JavaScript. So everyone's like, "Oh, well, you need your content in uh like to be rendered for the bots and blah blah blah." Sort of. It sort of does work that way. Uh but they're augmented by different search engines. I threw in yep there. Is this true? the probabilities in the training data will not be there if they do not have the content, but the probabilities from rag will still be there. So even if your content isn't actually rendered, so I'm a technical SEO. I'm going a little nerdy here. Uh even if your content is not actually rendered for bots, guess what? Google has that content. Bing has that content. They rendered that. So you may actually still show it may still site your page. And some people haven't figured out like why that is. It's because of this. You have multiple of these probabilities. There's LLM.ext. This is nonsense. Do not do this. Do not believe any tools that have this. No one uses it. There's actually a competing standard. I think it was OpenAI introduced. Um, there's no backing for this. No one is going to use this. Schema doesn't work the way people think. I still want you to do schema. I don't think it's really going to impact many things. is going to be more like I love it from the search side because we can use it and say like oh this is the pricing and blah blah blah. It's very easy to extract. We'll try and get that data anyway. Uh it's not going to really help you like rank better, be citing more, understand your content, all that's nonsense. Don't believe that. And all of these tools, all of these systems, it's all black hat. We're back to 15 years ago in search. Spam works. I'm not your mom. I'm not going to tell you not to do this, but I will warn you there are risks. We're seeing a lot of things with parasite sites, hacked websites, expired domains, Reddit or spam is spammy out there and it is working. There's even data poisoning. Like if I say uh the owner of Hrefs is Patrick Stocks and I put that out on the web and that starts coming up instead of Demetro who's the actual owner. Well, this has legal issues like people are lying about their competitors. They're lying about like data sizes and what the product is capable of and all kinds of things. Uh there are going to be some very interesting lawsuits related to this. These systems are cleaning some of this data like but they're not doing it well enough. They will they will have to do it more because it is spammy. It is nonsense. You probably have a year, two years, 3 years where you will likely be able to spam and get away with it. But that doesn't mean you should. We got lucky as SEOs. Google was like, "Here's Penguin." It hurt for a few years. They did a few rollouts. Sites got decimated. Businesses got decimated. Businesses went under from this. And then in 2016, Google's like, "Yeah, we just forgive all those bad links you built. We were really, really, really lucky. AI assistants, they might hold a grudge for the next year, 5 years, 10 years, 50 years if they're still around then. They might penalize you for spamming So, I'm not saying don't do it. I'm not telling you to do it. There are risks and that is something every business will have to determine for themselves if it is worth the risk. There are some legit things you can do, but all of these can also be abused. Uh, one of the things, you know, even Google writing multiple articles, we used to be like, "No, no, no." Well, at one point we were like, "Yeah, write a page for every single variation of every single keyword." And then we're like, "No, roll that into one." And like it was the one page to show it. And now, like, um, for a while you could potentially have two, but it might be like you rank fourth and fifth. They were always together in the search results. uh what about a year and a half two years ago almost um Google changed this so now you could rank like first and 34th so that's already been changed but multiple articles can actually influence that training data multiple could be cited for one uh for the same thing there are other websites so we're back to like micro sites being a thing that was a thing for a while in Google um they're back for AI assistance uh you can even buy news media You know, that's that's a popular tactic right Influencer marketing, affiliates, paid outreach, content syndication, digital PR, any of these are good good on their own, but they can also be abused at scale. This is a lot of different teams involved. So, there's a lot of people saying that it's still just SEO. Well, kind of. But like, were you working with all these different teams already? Probably should have been bigger brands, more likely they were at least aware of the teams or have teams to do this. But it's more than SEO. And it's more than just those teams. Now you've got branding involved, product, social media. Everything you do matters. These are your top AI search tactics. And this has nothing to do with SEO. Be a better business. Build a better product or service. Solve any issues or complaints that people have. It's be a real business. Do good. If people are talking about you in good ways, you're likely have fans. you likely will show up in these systems. If they're saying bad things, the system's going to say bad things. So, this is how things have changed. SEO is basically optimize your website. Maybe some off-site links, blah blah blah. Now, you have to optimize how the internet talks about you. If you're a small business, you can probably influence that decently well. If you're a large company, if everyone starts saying like, "Oh, uh, AT&T has terrible network coverage." I don't care how much reputation management the SEOs are being paid for. It is not enough. So fix whatever they're complaining about. Be better. So it is more than just SEO now. Everything matters. All these different teams, all these different actions, the entire internet, everything that's being said matters. So just be better. If you have one takeaway, that's it. You can repurpose your content. So, repeating messaging is a very common strategy. Now, uh if you want it to say something, say that thing. Say it repeatedly. Do it on different things. So, blogs, presentations, social media, you can repurpose your content in a number of different ways. This can also get very spammy. We used to do this even like plague sites and blah blah blah, fake videos, all kinds of stuff. It can be spammy but done right it also helps. You also want to diversify your traffic. So social media YouTube is the second largest search engine. Let me repeat that. Chat GPT is not the second largest search engine. YouTube is YouTube. Tik Tok Instagram shorts. There's a bunch of these short video platforms. Now if they're not using these for sources yet they will. So what is said there? All the comments that are said there, what is said in communities, it all matters. Like people are going and spamming Reddit. Well, I'm a Reddit mod/tech SEO. I delete 80% of the posts on our subreddit. 80%. And that's getting worse. If you want to legitimately control what is being said there, empower your employees. A lot of companies have uh ridiculous rules that are like don't go and be involved in the community because you can't say this, you can't represent the company, blah blah blah. That all has to go. We have to get rid of all that nonsense. You want your people, your people are already a paid workforce. They are your strongest allies, your strongest advocates. Empower your employees to publish content, to be inside these communities, to be active, to help you control that narrative. Because I don't care how much the SEO team is being paid, how many resources they have, it is not enough. It is so many people involved in this. Now, if you actually want to be able to control what is being said, you can use HS brand radar to actually see how you're mentioned. Um, what's fun is we actually have all the data of all the web pages that we crawl too and the history of that. So, you'll see some some very cool things coming there. my clicker. Okay, there we go. Uh, also check your competitor gaps. See what that is being mentioned about them, what they're being talked about for that you're not being mentioned for. In any of this stuff, I I tell people, our content creators, like write like you would with for featured snippets, but you're doing that over and over and over again. Be very specific in your messaging. Be very specific in what you say. I know we're going to have some presentations that talk more about that. Maybe bluff framework, a bunch of different things, but like be specific. Don't be vague. Don't use a bunch of marketing jargon. Actually say what you do in your product pages, please. I hate going to somewhere and I'm like, "Okay, I went to like 10 pages. I still don't know what this thing does. It's ridiculous." And cover the topics well. So, look at those uh uh the related searches. Look at the related topics. Write for those. Make sure you have some kind of content either on that page or another page that covers those things because if someone's asking about this, you want to be mentioned. You want to be there. We've also found statistics work well. So, you saw earlier the 34 and a half% less clicks. This got picked up in all kinds of news articles. It's Ryan Law's study. Uh the 23 times higher conversion rate for organic also got picked up. And this these were our two top traffic pages from AI search. These two statistics drove most of our search traffic from AI search to hrefs. It works really well. You also want to redirect these hallucinated URLs. This is mostly chat GPT that hallucinates. When we looked 3.6% 6% of our traffic from AI search were to URLs that don't exist that never existed on hrefs.com. We actually just rolled out on Friday uh some more features within patches and site audit. If you don't haven't used patches, it's basically automated tech fixes. So we rolled out like fixing canonicals, redirects and uh automated internal linking just on Friday. My first project right after evolve will be looking and see like can I just redirect these hallucinated URLs? Like sure I could ask our devs to do it but I want to show off patches and be like anyone can do this. You don't need a dev team. Like just click and redirect them. Done. A creative solution I also saw was just asking AI to remember certain URLs. probably not likely to last very long if that gets abused, but I thought that was that was neat. Ryan ran um a study for content marketers. 87% are using AI to help create content. I'm going to ask y'all one question. This is it. How many of y'all are using AI to help create content? That is maybe half the room. Guess what? companies that use it publish more content. Do you want to publish less content? I hope not. But most that are using it are at least doing it right. They're editing. They're reviewing the content. I wouldn't put out anything that's just AI. Uh, okay. Maybe some programmatic stuff, data driven. Uh, but mostly I wouldn't just put that out there without any kind of human review, especially likeformational content, that kind of thing. Um, but at the same time, like I wouldn't put out anything written just by a human today either. Sorry. I know that sucks, especially if you're a writer. Lots of people are already using this. lots of the content that is ranking has AI mixed in. We use our AI content detector to to actually see this. But a person and AI is going to create better content in less time. I don't think a human can create as good a content as the combination. I don't think AI content should be published just by itself without any kind of review. Add your own expertise. add your own insights. And this goes for everything, not just content, workflow automation, every business process. I don't think it needs to just be humans anymore. It shouldn't be. We're going to do a worse job and take more time to do it. That is a losing strategy. So, if you have one takeaway, use AI in pretty much everything. Now, these there's a lot of businesses that are getting this wrong, though. They're like, "Oh, we're just going to replace humans with AI." No, not going to happen. Uh, it's the combination of the two that's going to be successful. We're going to see a trend probably for the next 2, three years where companies are going to lay off people. Everyone's like, "We're just going to replace everyone with AI, and then they're going to hire them all back." But the companies that succeed and will dominate them in the meantime are using both. You need the people. You need that creativity. You need that expertise. You need those insights and you need AI because it's going to it's going to make it better faster. That is how everything is going to work And that is what I have for you. Thank you all for your time. [applause]