James Dooley: GEO versus SEO. Is it the same thing? I always ask this question to different SEOs and marketers. And today I'm joined with Ben from AISO ISO and he has a lot of data and evidence related to artificial intelligence and large language models. So Ben, let's jump straight in. Is GEO just SEO and have you got any evidence to back this point of view up?

Benjamin Tannenbaum: Yeah. So that's that's the big question. Uh lots of debate on this. So uh I'll say a short answer and then we can maybe dive a little bit deeper on some aspects. the the the first uh answer to the question uh and we covered that in the first part of this series is that uh you first want to know if people are asking the same thing to AI than to Google because if they're asking completely different things um even if the techniques to rank are the same you're still aiming for different target and we talked a little bit about how the questions are you know in are different from the Google to summarize they are much more informationational much more specific and sometimes the conversation is the funnel the entire funnel going through different steps. Now once you know that um there's still a question on okay but maybe the techniques are the same so maybe the queries are a little bit moreformational but if I use my SEO tools you know my all the techniques to create uh to be high in the ranking then I just aim for that query and and and then I would be uh I will appear in AI. Uh and the short answer to this is that it's u it's actually 80% true. Uh so you know the techniques are are really similar. Um but what you really want to make sure is that you're aiming for the right target. Um and this is how we uh essentially discovered and the whole world of SU has discovered the concept of query fan out. um where essentially AI uh all AI in Gemini JB all of them when you send a a question uh very often the AI will summarize the question and send this as a web search to see what are the most recent results for that web search um so I'd say that yes uh you know Jio is SEO but you want to aim at those web searches which are called the query fan out and you want to be visible in you know those um uh as as a search as a result for this this uh this background web search aka query fan out. um little more nuance to it and I I'll let you uh then uh tell me which direction you want to go but little more nuance is that the ranking the the absolute ranking okay like being number one is much less important uh for AI because it's going to take the top uh 30 up to top 50 uh results for those query find outs and then depending on how present you are okay um you're more likely to be in the in the answer. So, first of all, you don't need to be number one, okay? If you're top 10, it's it's it's fine, okay? It's more than enough. And also, there's a premium. There's a bonus if you appear multiple times in this top 30. Okay? Uh so for instance, if you're half, you know, if if you're not in the top 10, but you own all the position from top 20 to to top 30, uh you're actually very likely to be in the answer even though you're nowhere to be seen in the in the traditional uh world of SEO.

James Dooley: So right, let me stop you there then because in one of the sayings you said, so GEO and SEO, they're almost the same thing, right? But I want to I want to throw a caveat into this because generally speaking within SEO, if you're not ranking top five, you're not getting any clicks. You're not being cited. Like you don't have that discovery pretty much is the top five results. But now you're telling me I could be ranked in position 12, 13, 14, 15, 16, 17. Now I'm getting in the AI overview. No one being cited more than position one result which makes it very different to SEO and organic rankings. Plus, you're telling me that with regards to GEO with the previous episode that the searches are a lot different. They're not short little three and four um word kind of searches. Now the more longtail multi- kind of query type questions. So I'm hearing that actually go and SEO is is actually very different in my opinion from what you're telling me. Would you agree on that?

Benjamin Tannenbaum: Um yeah, I think it's different. But then if you you you you aim for different queries and you aim for the fan out for instance then how to appear uh in the you know in the in the rankings for the fan out even though you may not need to be number one but how to appear there then it's going to be using all the tools of the world of SEO. So it's going to be creating content. It's going to be making sure your your page is easy to crawl doesn't spend too much budget. uh is going to be having you know enough authority that you're considered you know trustworthy by Google using backlinks. So all of these techniques still apply but they just apply to um slightly different uh targets.

James Dooley: So there's slight different nuances then but the same SEO practices if you're going after all the query find out terms is the same. It's still semantic SEO semantic triples to get you those kind of answers. But anyone who's listening to this, how do you personally go and find the query fan out queries on chat GPT and on Gemini? Is there a tool that you're using? Are you going in in the in the code of chat GPT and stuff like to find out what those queries are? I know there's different people that are using different tools.

Benjamin Tannenbaum: Yeah. Um so yeah maybe to answer how how we made all these experiments to find if it's top 30 or 50 for instance uh I should explain a bit how we generated the fan out and um yeah there are different tools uh starting from the most there's different ways starting from the most manual to how we do it the most manual is to look into you know the you open the session you ask a question to to Gemini or or GPT you go into the network request you find where it is you know a specific JSON and you can find the the the find out Um after that there are some Chrome extensions uh some free tools you know free Chrome extensions that are working fine. Um now the issue we had is that we wanted to generate not just like the finals for one question we wanted to generate like thousands of them. Um and and and so we basically build a tool for ourselves at the beginning uh and then we basically added it to our app. Uh and the difference is essentially that um with our tool you can just generate like you know 10 you know 20 thou hundreds of prompts I mean you can put like as many prompts as you want and then you get all the fun for those um and then it's also a little bit easier for you because you see like some you know uh you're going to see for instance okay out of all the problems that you're interested in these are for instance um the most common brands that appear you know it might be a competitor or this is the most common source that appears right if you want to do a bit of digital PR you want to understand what are the source that are dominating the fan out um and you can have like a high level overview for many uh many prompt and many fan out instead of just doing one by one so that's how we generated them

James Dooley: so if if I'm inputting multiple prompts for anyone who's watching this and they don't know what your tool is what is the tool called and how can they go and import 50 different prompts to get hundreds different query find out stuff that they then should be optimizing for. What What is the tool called?

Benjamin Tannenbaum: Yeah, the the tool is called ISO. A IO and the website is getiso.com. Um so ge t aiso uh.com and there is a free trial. Um so you have like three credits I think. Um and uh that's one feature. Uh there's a bunch of other features but that's one feature called fan out and one credit gives you access to I think um 20 uh 20 prompts so you can try you know 60 prompt I think if you want in total um and for each of them you're going to have you know for um for GPT about two query fan out on average sometimes three usually two um and Gemini Gemini generates a lot more fan outs uh so you can you can uh you can generate a lot more for for that model um but uh you can try for free and then you So if you want to add more prompt uh you you can you know you can just buy credit and you can put the prompts you're interested in you know the one you think are the relevant ones uh but you can also use another feature of the app to see what might be the the real prompt that people are asking about you which is which is also helpful.

James Dooley: Yeah. So anyone who's listening to this with regards to the query fan out, you touched upon there where we had somebody else who was on that spoke about the query fan out and they said that generally speaking on chat GPT on average they found that the the average on chat GBT was three query fan outs and the average on Gemini was 10. Why do you think it is that Gemini does a lot more fan out queries than chat GPT? Do you think someone previously answered it that it's a lot cheaper for Gemini because obviously they can use the cash and stuff like that of Google. Do you think it is that or do you think that they're just looking to get more data? Like what why do you think one does so much more fan out than the other?

Benjamin Tannenbaum: Yeah, I think that's a simple I I in our analysis it was on average two actually for Gemini and 10 I agree it's about this for no two for JGBT and about 10 for Gemini and uh yeah I think the easy answer is that Chip needs to pay for each of the fan out search right um so there's been an evolution at the beginning they were using Bing then it looked like they were using Google and now maybe they look they use SE API which really is scraping Google so it's Google but it's they have to pay for it Um and uh so they need to pay for it, right? And uh and Google has probably access to, you know, much cheaper searches. That's my impression that it's just much cheaper for for for Google so they do more of it.

James Dooley: Yeah. So anyone who's watching this then with regards to the evidence base for you, is there anything that they should be doing to try to optimize for query fan out? So obviously let's let's just use Gemini. someone goes and puts a prompt in and Gemini comes back with 10 different query fan outs. Should they be like literally creating them as subheadings, new pages? What should should they try to literally because sometimes on the query fan out there's contradictory type fanouts like is your brand a scam? Um your brand complaints and all those type of terms. Should you still be trying to own that and framing that messaging to make certain that you're the one that's answering that? What would what would you give to anyone on the contradictory like fan out terms that's coming back?

Benjamin Tannenbaum: Yeah, that's a very good question and uh it's as always with the SEO it's it's about tradeoff and you know how much you know uh how much budget you want to put and what and where do you want to be in the funnel you know in terms of return on investment. So that the you know you have um if you look at a typical fan out so uh let's say uh the user is entering um I'm preparing for a trip to New York. Uh I'd like to uh uh eat pepperoni pizza uh but I I I I hate spicy food. Is there any restaurants in New York with you know not too spicy pepperoni? So that's actually a typical thing that can actually a real realistic query um for AI. Then um Chip or Gemini is going to take the question and they're going to summarize it to something like pepperoni not spicy in New York 2026. Okay. Then you might have another fan out which might be uh best pepperoni best restaurants you know with a repetition of the word best. Um the idea of this you know is basically given the way um essentially the information is used by the AI if you repeat a specific world you put more weight into this world. So you're more likely that the AI is going to uh consider that attribute. So you might have like best people only best restaurant uh you know allergic conscious uh New York. So that might be true and then you know for Gemini you might have like more variation around those you know with with with other aspects. Um, and the the the first thing you want to think about is okay, what are all the different ways you have leverage on this. So, you look at the search the Google search results for those find outs. You you you know, you should take the most recent one, top 30, I think it's a safe bet. Um, you know, you can go all the way to 50, but you should start, I think, with with 30. And then you want to see what's in there already, right? And you're going to have really three kinds of website. Usually, you're going to have like brands. So you might be there or your competitors might be there. That happens. Uh we saw that for many many users. Um then you're going to have uh media. So you're going to have for sure, you know, an article uh about, you know, the top 10 uh pizza in uh in New York City uh in some, you know, in in a famous sort of, you know, lifestyle um paper. So you're going to have like media like this and uh you're going to have um let's say social media, okay? you're going to have like a red thread on the best paper on your pizza, right? So, it's very very likely. So, then you have those three options and you have a trade-off in each of them in terms of effort and reward. So, if you see your websites, well, that's the lowest effort. Um, you know, you're already there and maybe that's fine. You have nothing to do or maybe you just want to make sure that you're addressing the specific query a little bit more, right? That's it. You know, you make one edit to your page. So lowest effort decent reward. Now if you go into the media into the paid media uh so like the listicles you know like the top 10 pizza here it's you know it's higher cost because those things are not cheap to to be featured in you know it's digital PR but it's likely that this specific media has not just one presence in the fan art but maybe like two or three articles. That's usually how it happens. And so then you basically multiply your likelihood to be picked up by two or three if they have two or three mention. It's almost as simple as that. So it's much you know the reward is two or three times higher. Then if you go into social media there depends okay if it's a LinkedIn article it's you know cheap and low low risk. If it's a Reddit thread it's high risk right? Because if you go on there and say, "Hey, you know, the person is asking about best pizza." You say, "Hey, this pizza is best and you're picked up as someone you you notice, you know, by the other Redditors as someone who works for that company, you're going to have like a push back, right? The Redditors are going to say they're very mean on this platform on Reddit, right? They're very rude. They're going to tell you, hey, it's obvious you work for that company. Get away from here." And AI is sensitive to this. So the Gemini and Gupta are going to are going to notice that you're trying to uh you know that you have like um dubious practice slightly you know suspicious practice and this might come up in the answer uh when you know uh when you mentioned so it's basically a double-edged sword right those platform um so I'd advise not to do it you know we always say uh uh you know well a bad ready strategy is worse than no ready strategy so so don't don't go there if you're not ready to, you know, play the long game, spend 12 months on this, do it properly, you know, if you're not ready to do this, just don't even bother. All right? Uh it's it's not uh so I think that's kind of how you can think about it and and and depending on your budget and and what you expect in terms of impact, you should aim for one of those, you know, channels.

James Dooley: Yeah. So, I've got a few more questions on query fan out, which is obviously related to AI SEO, LLM SEO, or GEO or whatever you want to call it. Somebody on a previous episode told me that with regards to and they said they had data on this with regards to the query fan out. They broke the query fan out down into six dimensions is the way that they called it um or six categories. And the six categories of what they said was entity query fan out, attribute query fan out, freshness query fan out, consensus query fan out, reputation query fan out, and contradictory query fan out. So just to explain this for anyone who's listening to this, entity is about the brand. Attribute is about like the services that the brand might offer. Freshness, freshness of testimonials, freshness of case studies and so on and so on. Consensus, how many times are you being repeated being mentioned? So not ranking in position number one. Have you got multiple listings um in the top 50? Reputation comes back down to case studies, awards, testimonials, reviews and stuff like that. Contradictory, is there any complaints or negative reviews or lawsuits against the company and stuff like that? And they said that every time it always fitted in those six dimensions. One, have you heard that? And two, do you agree or is there any more categorization that you find with the query find out?

Benjamin Tannenbaum: Um, yeah, so similar categories, maybe not exactly those um and maybe it's covered by this description, but there there's one thing um the the there's one thing that often appears which is and you know people who've looked at those have have probably seen this. It's like um taking your the request that's been done like the the prompt, summarizing it and then adding uh you know a location and a date. So you know if if I'm looking for you know best uh pepperoni pizza in New York uh I'm already giving like the location but I might just ask like best pepperoni pizza and uh you know for both Chip and Gemini whenever you send a prompt the location where you are is added to the prompt by default. Okay. uh there's actually no way to take this out. It's not a setting. It's just the way the prompt is sent. So whatever your your prompt is, the location is added to it based on your IP address. Um so often the fan out is going to be if I just say best people pizza, it's going to say best people pizza New York and then date. Um then some other fan out are not going to have the date and those um the the ones that are you know you can say time dependent fan out like this is a category that we had um and maybe it's related to the category of freshness that you had um but it's basically just a way for the AI AI to look at like the most recent results right and to see what's like currently the the best in this category. So, of course, that's apply that applies to SEO, but you know, if you can make sure that your key, you know, content is updated as to the latest year, uh, it's a good idea. Um, because you're going to you're going to be more likely to to to feature in this in this time dependent or freshness, uh, fan out. So, that's a quick win and it's a very good idea to do this and, you know, assuming, of course, that you rank for for that one uh in Google. Um and the the the the other category and I think this might be related to the entity or or the attribute maybe one that you described is really like this repetition of words in the fan. So you're going to say like best pepperoni pizza and it's going to say you know I'm I'm I'm pushing it. It's not never going to be that obvious but it's going to be like best best best best you know pepperoni pepperoni pepperoni. Um and and and and those um the idea is if you look at the space of all the text that the AI has access to and that includes like the training data in addition to the web searches by the way because you know you want to be in the fan out but you're so also going to look at the training data right and this we're not going into it because it's much harder to influence but there is it's it's you know we have to be aware they exist. So you have the training data which is kind of static and then you have the live search which is done you know on the on the fly and when they say best best uh they basically want to take all this information and put more weight on what's uh you know um uh likely to uh describe the uh the best option right so they they want to put more weight into it could be you know that you ask I have like celiac allergy uh or sensitivity And it's a it's it's super important for me, right? Um maybe even you said in your in your settings that this is something that the the LM should really pay attention to. Um and we'll talk in the next episode about personalization. But if you put if you talk about this and you say it's very important, the the LM I repeat like five times celiac because it's going to overweight that point. So in terms of how you want to address this, I think is that as a business, if you have an edge, if you have something you're very good at and you differentiated, which is usually is the case, you want to make sure that it's super clear so that when a query is heavily weighted for that specific feature, it's it's obvious to the land that, you know, this is what you're about. So you really want to kind of double down and basically just dep prioritize all the generic description on your website and really overweight and put at the top and put in the most popular articles and especially those that are ranking and ranking for the final put like those things where you differentiated because you're much more likely to be picked up for those weighted query. So those at at the moment are my kind of um I I did these categories like you said for instance you might have like reviews in the finite if it's about perception you might have uh specific brands that are mentioned even if so it's a branded search effectively even if you didn't ask for any brand um I I think it's you can go into this and it really depends on how to because it's essentially they take the query the prompt then there's a little mini LLM right that that looks at the query and generates the final and depending on where this mini LM has in this training data is going to trigger a branded search or time search etc. I I think I would focus on the first two like the time dependent one and the and the and your differentiating feature because those are the ones you're most likely to have an impact on. Um the other ones where it's like a branded search um uh you know things like this you you might have an impact on it but I think it's it's much more difficult. Um so that that would be my kind of you know approach for this.

James Dooley: So a follow on question um consensus you've mentioned that consensus is really important right so what I mean by that is let's say um someone does the query fan out and I've got position one rankings so I'm ranking position number one but I'm I'm my brand is not being repeated anywhere else but I'm ranking position number one and somebody somebody else's brand is ranking in position 21, 22, and 23. So, they're ranking, they've got three spots in the top 30, but they're only on page three in position 21, 22, and P 23. I'm ranking in position number one, but I've only got one. In your opinion, what would be more likely to be cited? the one that's ranking in position number one but only got one mention or the one ranking in position number 21, 22 and 23 with three kind of mentions. Which one's more likely to get cited in the AI overview?

Benjamin Tannenbaum: Yeah, that's a very good question and we we didn't I run lots of tests on this but my our anecdotal impression is that the number of mention is more important than where you rank. So the person who's ranking multiple times um because you know as you know if I send one query if I send the same query like a thousand times and we'll talk about this in personalization the answer is going to be slightly different. So the way you can look at it and range fishkin came up with this concept of like lottery like you know next token lottery u you know which is quite you know kind of porative in his mind but in a way it has like you know it's partly how it works. But if you look at to simplify it, if you look at like the 30 results in the fan out, it's a little bit of a lottery which one it's going to take out of those. Okay? And you don't have a bigger weight if you're number one. You might have a bigger weight if like semantically you're closer to the query. So you're answering something that's much more closer. You know, the snippets say like an important keyword, especially one of them, one of the ones that are that is repeated in the final, right? for instance, that's that's a better, you know, that's going to increase your weight. But if you're number one or number 11, you still have one shot out of 30. And if you're like, you know, 10 times out of 30, well, you've got 10 out of 30. So you got like 30% or 33% chance now. Um, so that's that's our anecdotal observation.

James Dooley: Yeah. And then one last question with regards to chunking or passages of content and stuff like that. So someone was telling me where when they get the query fan out that because of the cost element to open up the page and the time constraints to open up all the pages that a lot of the time they actually don't even open up the pages. What they'll just do is they'll just use the SER so the search engine results page and get the the meta title and the meta description. And a lot of time the metad description itself might not be used. It might pull in a passage of content with as part of the metad description. So like Google would choose its own metad description so to speak and without opening up the page they'll get the f the top 30 results if they've got enough answers within them top 30 results and a consensus. So let's say one brand's been mentioned five times within that. They've got enough information. They don't open up any of the pages. They're just using the metatitles and the descriptions. How true is that? Or do they ever open up the pages and get content from what's on the page that might not be in the search engine results page?

Benjamin Tannenbaum: Yeah, that's a super interesting question and we've done a lot of research on this, but uh for clients, not something we publish, but I'm hoping to publish one thing. And we looked at this in the context of YouTube content for instance. Uh and the question that you can that is like an analogous to what you said is is AI watching the video or are they just uh getting some kind of sum summary of it right and what is the summary and here it's interesting that I can give you a couple of pointers first of all there's a difference between chat GPT and Gemini uh so similarly to why Gemini does 10 query find out compared to two for GPT Gemini also uh looks much more in detail into the actual content of the page than chipity. Uh many many cases many very often what does um and also there's some kind of differences if it's thinking or not etc. But to put it simply very often what HP does is giving you the impression that it read the page without doing it. Okay, that's like 99% of the cases. And so if as you say it's going to look at the the the snippets like the just the search results without opening the page and there's enough information in there. It's going to tell you in the answer. It's not going to tell you only looked at the summary. It's going to tell you based on this page. So it's going to look like it opened the page. But really it's only going to look at the search results. That's for tragicity. Gemini, it's a little bit more subtle because they actually um read the entire thing and and so it might actually be content in there, but they also try to give you the answer as fast as possible. So sometimes they just stop at the at the search result. Sometimes they go a little bit deeper into it. This depends on the query and you know how spec like how much they need to get information. But uh but for for tragic they make it look like the open pages but they almost never do.

James Dooley: Yeah, that's crazy. So, anyone who's watching this, obviously I've picked Ben's brain with regards to GEO versus SEO. Is it the same thing? Is the different nuances? Make certain you check out the link in the description where on a previous episode we talked about how searches are very different when people are doing it in AI search and the large language models like Grock, Perplexity, Chat, GPT, Claude or Gemini. In the next episode, we're talking about the personalization of LLMs. He touched on it there before about how you can be using the exact same question on the exact same machine and get different answers, but also from one person to the next, you can get different answers as well because of the history. You be can you could be using different LLMs and stuff like that. So, make sure you check out the links in the description.