Google AI Mode vs Traditional Search - #SEOFOMO TL;DR July 1st, 2025

The latest in SEO with today's edition of SEO FOMO Too Long Didn't Read. Let's go through the top news, the hottest news from the last edition of SEO FOMO. And to go through them with me, I have today the wonderful Roxana Stingiu, who's the head of SEO and search at Alamy. Hello, Roxana. How are you?

Hey, Aleyda. Hello, everybody.

Thank you very much for joining me. Before starting though, I want to Say thanks to our today's sponsor, who's SimilarWeb. Thank you very much for your sponsorship. Let's start with the first news, which is pretty much the comparison that I created of key traditional versus AI search differences, a visual comparison to highlight those where there's an actual difference. So we have search behavior from short keyword-based one-off queries with high navigational intent to AI search, long conversational-based multi-turn queries. And I created infographics like this to show it more clearly. In this particular case, it's very straightforward to understand though. However, there are other cases like the query handling, right? Like a single query match to the query fanout, multiple queries matches that we know that AI mode and also other AI search platforms use. So that also changes optimization target, which is another thing that I wanted to highlight. And finally, the results presentations that goes beyond what we have seen, the 10 blue links and the implications to the expectations in traffic and engagement from users. And at the end, I created this overall infographic here to go through all of these aspects that you can download to have with you, to share with your clients and so on. I believe that we have seen, especially I see this a lot in social, like these very extreme stances of nothing is changing. It's just the old usual SEO, but apply in a little bit different way. And then the other extreme takes about SEO is dead. This is the new thing, completely new thing. I am more blended in the middle, let's say, of these 2 extreme takes.

I'm in the middle as well with that. I can't, I can't decide either way because there's good pros and cons in both sides. So it's really hard to take one side and go with it. But seeing traditional search versus AI mode side by side, yes, it looks very different. And I can see why a lot of people say this is not the same thing. But if you kind of go back to basics, like when we started doing SEO back in the day, and you think about it, the concepts are somewhat similar. It's the method that changed. So to give you an example, the query fanout, which is like a new terminology, it comes with AI. LLM-specific, whatever. If you think about 10 years ago when Google put out Hummingbird, it was taking your search query and then finding synonymous queries and then doing like a bunch of them to find you different results that were a bit more relevant. And the query fanout is not just synonymous, it's also just closely related as well. So it's like that next step in the evolution of search, but the concept is pretty much the same. You take a search and you expand it. Which Hummingbird tried to do so badly at that point. Although 10 years ago, that was the best improvement in search we've seen in a long time. It just shows how much technology evolves while keeping certain search concepts the same, but bringing methods that can do it so much better, like the query fanout. And I think it's kind of the same with the passage ranking. Google's done that before. But now it just does it to a different level.

100%. I love that you bring these thoughts regarding concepts versus practicalities, because conceptually, like most of this, we're already applied or using SEOs, but in different ways and different ranges. And because of those differences, the impact that we will see, and we are already seeing a little bit of the outcome of the type of engagement, the type of traffic that AI search and LLMs refer is very different than the one that we are used to in traditional search results. That difference in interface and those differences in how the links are placed or where they are placed change the impact completely and the engagement of the user. And as a consequence, then we need to rethink of how we approach things too, because what we were doing is not generating the impact that we wanted. And this is what we also see right now. Due to Google AI overviews, all of the impressions going on, the traffic going down, it's like you cannot expect just to continue pushing and doing more of the same and expecting it to go up again. So how do you rethink what you're doing to diversify? The next resource that I want to go through is the Semrush AI mode comparison versus traditional search and other LLMs. On one hand, now we have, I think that little by little, better understanding how all of the pieces fit together. And then we have this data. This should allow us to make better decisions. This research that was done by Eugene Levin, we see that a few of the key findings are that AI mode prefers sidebar linking. 92% of AI mode responses feature sidebars showing 7% of unique domains. These links had a 51% domain overlap. With 32% overlap with the top 10 results. These results are grounded from Google's index. I'm not surprised to see this very high overlap. Organic links still matter. 7% of AIMO queries where additional links appear below the response, uh, there was 89% domain overlap and 80% URLs overlap with Google's top 10 search results. And then again, Reddit dominates from a citation standpoint. Which unfortunately we know the risks attached to a platform like this one. The query intent affects the response length. Commercial queries trigger responses twice as longer than informational ones. Here, something interesting regarding intent is that there was also another piece in the newsletter last week regarding the new type of intent we can see in AI mode, which is the one that is sometimes called task-oriented or generative that realistically is the most popular one at the end of the day and far less navigational as we see in traditional search results. And then that AI mode looks more similar to ChatGPT than AI Overviews. And this again, due to the format, the interface, and the overall type of answers that it provides.

Something you said a bit earlier, you said that, you know, the top of the funnel, do we need to still optimize for it. And we need to because that's kind of helping us with our branding and everything. And then if you scroll to the first chart that was shared here, it shows the overlap between AI search and traditional Google search. And then Perplexity has like the biggest overlap there. It kind of makes sense. There was an interview with Lex Fridman and the CEO of Perplexity, and they were talking about how does Perplexity work. And pretty much Perplexity's ranking system prioritizes sources based on combination of factors, just like Google does. But they also include authority, trustworthiness. So already it sounds like EAT, freshness, which we know Google always had a freshness boost. And then of course, relevancy to users' query, which of course it's a search engine at the end of the day. So it was somewhat expected that it would match so closely to Google. So this kind of works well for the people saying it's the same search. You don't really have to do anything different because In this case, yeah, you do it well for Google, it will help you with Perplexity as well. But then some of the other AI engines were not really overlapping as much. So there's differences, of course, and the more we can understand those, the more we know what to do to optimize for them.

We need to recall how Google AI Mode was tested very few months, like 3 months or so. Like, I think that this was the shortest period of time that I have seen Google testing something like this with AI Overviews, they had much more time. I believe this because of course there's pressure for them to compete with a similar product that can be considered like a competitor to ChatGPT, but also at the end of the day, they need to test a few things out to balance a little bit the act of trying to keep sending traffic to a diversity of domains, to a diversity of sites. They also have The index of information, unlike the other players that have had to practically like partner with other parties in order to have access to this sort of information, like OpenAI ChatGPT with Microsoft. But Google has all this data directly in any case. I believe that it will highly, highly depend on the monetization, how well they are able to monetize this new AI mode with ads. Let's move on to the next resource, this insightful and very well-timed guide from Glen Gabe about how AI search currently drives less than 1% of the traffic to most sites. Google is still dominant and yeah, it's very risky to overlook it or ignoring it completely just because we're chasing the shiny new AI search thing, right? Well, at least for a while until Google potentially decides to push it as default mode after testing it much, much better than what they have so far, this won't happen. So I think it is a very well-timed article showing also how to monitor and configure. There is this video, Dana Di Tommaso shared this how-to to show us how to monitor, to track AI traffic. And yes, it's trivial. If compared with the one that we currently get from organic search results. And I have also seen many decision makers literally have a completely different attitude on investing on AI tracking, search tracking platforms, or AI search-related priorities based on what usually attracts their attention in SEO while SEO still brings 99% of things, right? I think it is an important reminder here of it's good to prepare and start testing and be proactive with AI search, but we cannot overlook what is already bringing the value. And it is important to configure well this also in our own analytics, because unfortunately this is much more difficult to see with the lack of specific segmentation by Google and the Google Search Console. So this type of configuration is fundamental for sure. And then Based on that, we should be able to make better informed decisions and prioritizations in our efforts accordingly.

I think this article is worth reading just because Glen has a way of kind of storytelling and sharing his experience, but also because it acts as a word of warning that you shouldn't put all resource towards AI traffic because it's not yet there. It's not yet going to be the majority of your traffic. It's not yet a leading channel. So you still have to pay attention to traditional search because that's probably still making up quite a big percentage of your traffic. And even, even though AI engines will get more and more people to your website, as you said, like things like ChatGPT, it's not really an AI engine. It's more like a virtual assistant that has access to a search engine. So the intent of users using it will be different from what you're getting from other search engines. So you can't really replace one with the other. You just have to start building for both.

Also taking into account that a lot of what we're already doing in SEO sets a really good ground on what matters for LLM visibility, right? Yet another reason of why we should continue taking care of our SEO and also doing things in the more mature, sophisticated ways. For example, developing and building campaigns. Well, citations also matter, as we can see here. They don't need to be direct links, right? The authority, the relevance of where it's placed and the sentiments, all of that is definitely a much more direct impact now. But yes, it's not that we should have been doing that in any case in the past already, right? And this takes us to the last resource of today, which is the coverage by Roger Monti in Search Engine Journal about Google's new Movera algorithm that improves search speed and performs better on complex query. This is an announcement of Google on a new multivector retrieval algorithm called Movera. This new algorithm intends to speed up retrieval and ranking, improving accuracy too, and can be used for search recommender systems like the one of YouTube and natural language processing as well. Just in case there's an explanation. About what is vector embeddings and how Moovera focuses on solving problems in them. Recently, beginning with the landmark ColBERT paper, multivector models, which produce a set of embeddings per data, have achieved marked superior performance for IR tasks. So here, the vehicle challenge is the bottleneck from a performance standpoint, which is what it seems to have been tackled with the new algorithm.

So besides SEO, I also do search, which is not the digital search, but it's more a website search. So I kind of handle the search engine from Alamy that has over 400 products going into it. And once you start building the search engine and trying to understand it, you realize every search engine will try to solve the same problems. And then search engines like Google having more money and engineers and everything, they will solve them in a more complex way with greater results. But then maybe an e-commerce search engine being limited to just a website will solve it in a less complex way. I find that this kind of algorithm just solves one problem, and that problem is the cost and performance of search. So you always want results fast and accurate, but also you don't want them to cost too much. And I can imagine with Google having so many users, that is a big, big problem, right? When you have like a big corpus of documents that you're searching from, the bigger it is, the least accurate your results will be. You do like an approximate neighbor algorithm search where you kind of get the result, but you don't check every single document to make sure you haven't missed anything. And when you're Google and you have results in the million, that's more than perfect. It's enough. But if you're like a small website search looking in some help files, for instance, you want to be as accurate as possible. So you do a nearest neighbor algorithm search instead that literally searches every single document to find that one thing you might be missing. So Google tried to do a nearest neighbor KNN algorithm search on a big corpus, and this is what Movera feels to me solved. Because if you do that kind of very specific search on so many documents, it would take forever, it would cost a lot. No user would wait for that result. So they found a way to still search that big universe of documents by making it smaller to search in. And that's all they did. They, instead of having millions of vectors coming back, they found a way to concatenate those vectors and kind of bring them, sum them up or average them into a single vector. So they reduced millions of documents to maybe tens or hundreds. So when you have Fewer documents, then you can do a more accurate search that's more performant. So they solve that problem of performance and accuracy versus cost. And this article from Roger says, we don't know if this is in search yet. Let me tell you, it should be because it's more efficient and probably more cost efficient. So why wouldn't it be in search?

Yeah, this is definitely something that actually solves a problem that exists in search for sure. It's important to take into account that there wasn't an announcement of the release yet.

They probably have like a testing phase where we don't even know it's happening and then they will have an official release to search and so on. But that doesn't mean we're not seeing results coming from this yet. We just don't know it. It was definitely the best read this year.

And I love how you went through all of this in depth so you could share it. Today and your analysis, quite insightful. Really appreciate it. And I really appreciate all of your insights, uh, Roxana. It has been a pleasure to have you today. Remember that you don't need to wait until the next week to watch our SEO FOMO 200 read. You can see the latest news in real time by going to the SEO FOMO news. You should also totally subscribe to the newsletter, and I want to thank Again, to Similarweb, who's our today's sponsor. And as you can see, it's a tool that I'm using every day, a lot lately, especially with their AI-related insights, prompts, information, traffic engagement. Very, very useful to better understand the differences, the nuances, the impact that AI traffic is actually having on our sites and competitors also to identify opportunities. So go and take a look at it. Roxana, thank you again. Remember to like the video if you like it. Remember to subscribe to the channel and until the next edition, bye-bye.