How to Optimise for AI Mode in Google Search with Charles Floate

James Dooley: How to optimize the AI mode in Google search? In the previous episode, I was speaking to Charles Flo and Charles is saying that so-called this year in Q4, AI mode is going to become the default search engine. So, looking ahead of the game, we need to be looking at different optimization strategies to try and improve that LLM visibility. Charles, it's an absolute pleasure having you on and let's jump straight in like let's jump it with regards to the strategies that you would be recommending for people to do so future-proof getting their brand or business into AI mode.

Charles Flo: For sure. So, the very first thing we need to differentiate is the actual main brand domain versus the third-party sources that are getting cited and referenced and have the visibility themselves in AI mode. That changes slightly again from the personalization layer that we were talking about, but that personalization layer is kind of more tuned to what you were talking about, which is the everywhere optimization, where essentially personalization matters less and less if you are essentially the number one brand across the board. With with regards to trying to get the high visibility sources for your niche that are actually covering you, that third-party publishing might be very different depending on what niche you're in, right? And it also might be very different depending on if there's already huge players that are investing in their marketing or um brand awareness, that kind of thing. As a really good example, Calendly, which is again we were just talking about earlier. A lot of people associate Calendly with booking a call. Same thing with Googling something instead of searching something called searching on a search engine, you Google something, right? That kind of an association is the exact thing that we're A looking to reverse engineer from who is associated with those kind of terms and B trying build that association for our own brand and entity. For our main domain name, that's mostly about just reinforcing what the third-party opinions are going to say. So, if they say you have 100 employees, you need to say you have 100 employees. If they say you are based at this location, you need to say you're based at this location. It's all about reinforcing that consensus of information so that the machines can actually trust it, right? And they'll then, without having any sort of real verification, output it as fact in and a lot of users will end up believing that.

James Dooley: Yeah, for sure. I mean, for me, what I've seen a big shift on with regards to the white hat versus the black hat kind of SEO community, the black hats have always said third-party corroboration and backlinks and page rank distribution, link juice going through each of your website is really important. Now it seems the white hats, who only ever spoke about topical authority, like good quality content and all the rest of it, and it was all on-page, they're now starting to realize, through a model called claim, frame, and proof, that the proof part needs to be on third-party corroborative sources. And therefore, it doesn't matter how well they claim and frame something, if they don't get that third-party sources, they're realizing that they're going to lose the race. And even they're now are talking about acquiring links or trying to find ways to get this third-party corroborative source. How much do you feel, from a percentage point of view, for LLM visibility, would be on-page versus off-page? Is it like 20% on-page and 80% third-party corroborative sources, 10% and 90%? Is it a 50/50 split? What do you think is the split, if you were just putting an average on on-page versus off-page, specifically for AI mode visibility?

Charles Flo: So, it depends on how strong your initial domain name is on in the first place, right? If if you're a brand new company, brand new domain name, that kind of thing, the AI LLM is probably not going to source you at all anyway. So, it's going to be 100% third-party corporation in the first place to be able to get the AI to output anything because it won't trust your domain, it won't pick it up in retrieval. Um and it also there's under underlying algorithms like Google and Bing won't also surface you because you don't have the the history and authority and etc. etc. for the traditional SEO signals. Um when it comes to on-page for the sources, there's still a big kind of differential in the source picking versus the content side of things. So, if you have got Wikipedia for example, or if you've got the New York Times or BBC or something covering you, especially as an example, if you're trying to rank in ChatGPT and they have partnerships with like the Washington Post and things like that, those domains are significantly higher weighted in the AI LLMs than your domain will ever be, right? So, they will just intrinsically trust and extract and retrieve more data from those domain names than they will your own domain name. So, it's fundamentally better to be on those higher-trusted, higher-retrieval, higher-visibility domain names than it is to do it on your domain name almost all of the time. Unless you're, you know, Coca-Cola or Dyson or some massive, massive brand that has the authority to compete with those third-party sources.

James Dooley: So, let's talk about these third-party sources then cuz you're saying now that third-party sources are the most important element to influence LLM visibility for AI mode. So, when we now we're talking about AI AI mode SEO strategies, I'm going to start naming some things and you can tell me whether you think that this would influence AI mode visibility or not. Would doing press releases, which is on third-party sources, help increase AI mode visibility?

Charles Flo: Yes, but it only works with the lower competition this up basically. The lower competition the niche.

James Dooley: Okay, so not on the high competition, low competition press releases can work. What about guest posts? So you can go and get on a third-party source on a guest post, you can write it in a specific way. Does that influence AI model visibility?

Charles Flo: As long as the domain is being cited and sourced that the guest post is on, and the guest post is uh written correctly and structured and formatted so that the LLM will pick it up and cite it and it has a higher retrieval rate and etc. 100% yes. Um

James Dooley: So on there then, you're basically saying that the choice of the um guest post domain is important. Ideally, you want a trusted source that's already been cited in the LLMs and the micro semantics in the way you do semantic triples and feed the LLMs via the micro semantics is important.

Charles Flo: Mhm. Yeah, exactly.

James Dooley: How many How many get How many So wait, I'm going to throw something out here. How many do you feel link builders out there that are selling guest posts understand that and are doing that in comparison to let's say press releases? Cuz that obviously now you're saying that's the most important part. So many people just seem to sell on DR, not looking at is it a source and not looking at the micro semantics and the importance of that.

Charles Flo: Yeah, it's it's significantly more important, right? Like we will for every new client or campaign that we're doing for Ahrefs right now, we're extracting tens of thousands of citations at a minimum. So that's tens of thousands of sources for outputs in AI mode in ChatGPT, in Perplexity, whatever AI the client is trying for us to target. That's the other side of things is that AI mode is going to be the future default experience. But for example, if you're in B2B, then right now Claude and to some extent ChatGPT is uh used a lot more than Gemini is for B2B use and things like that. So a lot of people are looking at which AI mode which AI model or which AI model to target in the first place.

James Dooley: And then AI modes Go on, sorry. Go on.

Charles Flo: No, yeah, with AI mode specifically though, the higher DR stuff doesn't really work anyway because it is fundamentally powered by Google's underlying search algorithm. And the search algorithm doesn't have a direct connection to DR at all. You could have a DR 30, but it's 100% niche relevant and it'll rank number one. But you have a DR 80 that has no niche relevancy that's been hit by a bunch of penalties, been all that smashed to pieces, that kind of thing. It won't rank even in the top 100, right?

James Dooley: And then what about let's move away from backlinks for now or the third-party sources on web pages. What about is there is there anything people could be doing with image search to try to get in AI mode or are you not bothering with that? Like what's your thoughts on image SEO as an AI mode SEO strategy?

Charles Flo: So, we've been trying to do image corroboration mostly just for ORM and brand elements. So, trying to get the the model to understand and associate our logo as an example, right? It doesn't When you get ChatGPT or AI Gemini or anything to create an image, it doesn't hallucinate a logo for you. It It puts your logo in the in the image for you. That takes some training to do. It also takes some association with like your color scheme and your planning and the design and elements and entity and all sorts of things around that sort of image optimization. In other ways, there are some techniques that you can do to get LLMs to associate elements and things with images and things like that. But most of the time, we're trying to work on the kind of highest ROI techniques. And whilst we're testing a bunch of stuff right now, there's kind of very linear pathways that you can go down to get the highest uh visibility increases.

James Dooley: And then let's move on from on visual semantics from images to videos. Are you doing anything with regards to long-form videos, short-form videos? Which one do you think performs the best or do they both perform well as an AI model SEO strategy?

Charles Flo: Yeah, so video we actually are working on. We're working on trying to get clients to do podcasting series, trying to get them to do active YouTube channels, trying to get them to do short-form video content for TikTok and things like that. Just even from the fact that you could get sub-features and AI model picking up the actual short-form TikTok videos, especially if there's company updates or company changes and things like that. Just create a short video and that might get pulled up and referenced and it might not hallucinate, you know, random changes and things that have happened to the business. In regards to short-form versus long-form, right now short-form 100% just because of context windows. So, most AI models, with that being said, Gemini is very, very good at large context windows, but it will still try and tune to have as little token usage as possible and reduce token usage so that they're not costing huge amounts of processing power for for the actual underlying models. But, they will try as hard as they can to absorb as much of the information and as the transcripts of those videos as possible. However, the bigger the transcript, the less likely they are to take the actual extracted elements that you want them to. So, the shorter form video you can make it, the and the shorter the transcript or the more optimized the transcript that you can make it, the better output that you'll have from that content.

James Dooley: Yeah, we found exactly the same as what you're saying there. So, I feel like it's to do with like So, on like if you got like the site radius within the Google leak and it's like don't go too far away from the the core topic. Same with videos, normally an hour-long video talks about multiple topics. It's all over the place, where a short-form is answering a very, very specific question. It's very on point to that and therefore can be being used, where some of these longer form, the cost of information retrieval can be expensive. For such a small little 1-minute clip in a 60-minute one, it's easier for them to go with the shorts. But away then from video, you mentioned their podcasting, how important you feel podcasting is for AI mode SEO strategies?

Charles Flo: Yeah, I think it's getting more and more important just because it's so easy to associate entities and people and everything with a podcast, right? You can very easily say, "Okay, this podcast is run by this person, operated by this company, it's had these guests that are all associated with these topics." Again, it kind of just reinforces expertise and trustworthiness for the AI model. And again, we're trying to basically create machine verifiable information so that the machine trusts and understands this information. And one of the easiest ways to do that is by having an expert-level podcast like we're doing right now with maybe shorter-form video content that the AI can transcribe, understand, and that is about specific segments and topics that it can then output as well for you.

James Dooley: Yeah, for sure. And then with you guys too, is there anything else for AI literally just for the AI mode SEO visibility? If you're looking to get that better LLM visibility in the AI mode in Google, is there anything else that's working well? Like, what about social media? Is there any specific social media platforms that's working well or forums like Reddit? What's your thoughts on social media as a whole and forums like Reddit?

Charles Flo: Reddit is fantastic because it gets sourced routinely and there's partnerships from Reddit to Google and OpenAI. So, you're kind of hitting both of the biggest two two of the biggest birds with one stone, right? Um in regards to other UGC, I saw a lot more use or a lot more retrieval of forums and that kind of stuff at the beginning of AI and it tends to have fallen off to more trusted domains and trusted information, especially with Gemini, uh more so than Claude or or OpenAI. With AI mode specifically, because it's got such a high personalization layer, you can actually manipulate or you can get the outputs personalized via your photos. So, they can actually go and go into your photo gallery that's connected to your Google account, and you could take a picture of yourself in a cafe, and you could be eating a red velvet cake. Well, the next time you start talking to them about the cake, they could actually source that image, source it to red velvet cake, and put that into your personalization layer, which might edit that output. As a really interesting one, Garrett Sussman, when he realized how kind of invasive the personalization was, that Google as the Google's personal intelligence layer understood that Garrett had a daughter. So, when Garrett went and searched best streaming services into Google AI mode, it came up with one of the sources as best streaming source streaming services for kids, because he knew he had a daughter, and he knew how important it was for him his daughter to have some sort of control over that viewing experience. Did he ever at any point mention his daughter, kids, anything? No, Google just inferred that information. Now, yes, that's creepy and weird, but that sort of personalization is showing you how much of a difference that sort of understanding of you as an account, you as a user, you as a profile, you as an entity will have on the actual the end output, even if you didn't necessarily put it in the context of your prompt.

James Dooley: Yeah, I want to cycle back to the start where we've mentioned about third-party corroborative sources for influencing AI mode in Google. And I want to talk about backlinks. So, we've mentioned DR a little bit, and we've mentioned the third-party corroborative sources, and the people in the black hat community used to always think of third-party sources as backlinks, link juice, increasing the DR, increasing the domain authority, sending like power through to a specific page, the anchor text being important. But now there seems to be a bit of a shift where a lot of people are talking about branded mentions being almost as good or just as good as having a backlink with regards to LLM visibility. What's your take on backlinks versus branded mentions specifically not for ranking in Google but for AI mode and trying to get cited in AI mode in Google search?

Charles Flo: Yeah, we we mostly just use mentions, right? It's it's effectively gone from If specifically if you're targeting AI SEO, AI mode, that kind of thing, then 100% it's gone from link building to mention building. Google can to some degree with the traditional algorithm connect entities back to domain. There's some discrepancy amongst how much link juice you're going to get for all these mentions versus if you had them as links in the first place. They still can sort of connect the dots. With AI mode, it there's not really any difference between having a link or having a mention. It's mostly and in fact some from our own testing, sometimes having the link can act especially if you're the only link in a listicle for example, it can actually cause the AI to distrust that specific listicle. So a lot of the time we are just doing mention building and we're not doing link building whatsoever and that specifically is a big big change that wasn't really a thing previously or at least it wasn't a thing unless you were doing crazy corporate SEO where they're spending, you know, 250k a month or something on SEO and that's one of the smallest channels in their marketing department.

James Dooley: Yeah, for sure. And is there anything else? Have I missed anything? Any questions that I could be asking you that you think would be a good question specifically as an AI mode SEO strategy to try and get in Google's AI mode that you're saying in Q4 this year that Google's kind of shareholders have come out and said Q4 this year, AI mode is going to become the default. Is there anything else that people should be doing who's watching this or listening to this should be doing to increase LLM visibility in AI mode in Google search?

Charles Flo: I think that the first thing is play around with it, right? A lot of the SEOs that I'm speaking to who are worried about AI abuse, AI mode, etc. replacing Google search have done very little actually inside AI mode or AI overviews to understand what it's like, what the experience is like, what everybody knows a Google search. Everybody understands that. Everybody knows what it's like, what it what the what to expect when you put this a query in. Not many people actually know what to expect when they put a query into AI mode so far. No nobody knows what the underlying system is doing to create that output and how you can tune to that and what it likes, what it doesn't like, what it prefers, what it doesn't prefer, etc. etc. So, the number one thing would just go and spend 20 hours actually using it and see what is different, how you think it changes, what changes for the queries that you're putting in. Make sure when you're using it, use it on a logged-in account, use it on a non-logged VPN account, use it on a brand new device, use it on a smartphone, use it on your partner's phone, use it on a a different person's phone. Try and have different types of personas engaging and see what the outputs are and register all that data and understand that data. In terms of just any other techniques and things like that, there's a bunch of stuff to try and get in Gemini grounding. So, actually get the underlying Gemini Flash model inside of its training data. As an example, with Coca-Cola, there's probably thousands of mentions in the training corpus of Coca-Cola of Gemini mentioning Coca-Cola. We would try to at least get a dozen mentions inside the training corpus for our own brand, right? Those kinds of things. Getting training data ingestion, so getting your actual data to be ingested by the models is itself a bit of a tricky thing, but Common Crawl, there's a bunch of um file-sharing websites that they use routinely. There's a bunch of like open-source PDF websites. There's all of these things that end up kind of actually forcing the the uh adoption from the training model ingestion. And that is the number one thing that will future-proof everybody is actually having the underlying model not even having to search for you, but then the underlying model itself know what to know what to say about your brand.

James Dooley: Yeah, and then the last question is So, AI mode um powered by Gemini, AI overviews powered by Gemini, and then you've got the Gemini app. Is there anything different that you're seeing come back? If I was to go and do a search in the Gemini LLM app, and if I was to go and do a search just in normal web search and the AI overview shows up in Google versus AI mode, so the the tab of AI mode, are they bringing back the same I know I know they're always different like per thing, but are they If you was on the same history and stuff like that, are they bringing back similar results to the Gemini app to the AI overview versus AI mode? Are are all them different answers that's coming back with?

Charles Flo: They're all different. So, specifically AI overviews use a custom tuned model of Gemini Flash or Gemini 3 Flash. Um AI mode uses a different Gemini model, um and then the AI the app uses a different one that's specifically designed for apps and mobile usage. So, just the underlying model itself and the architecture is slightly different depending on the app and the experience. And they're all tuned for their specific use cases. So, that in itself ends up creating the outputs are slightly different. Now, with AI overviews, they use a set query and then query fan out underneath. And it's usually like three, four, five query fan out. So, it might be like, you know, best CRM tools. It might also think that HubSpot versus Salesforce or something like that might be one of the subset queries. With AI mode, the subset queries are normally 14 to 16. So, they do 14 to 16 fan out instead of a four to five one for AI overviews. So, you get a lot broader information, a lot more citations, a lot more sources, that kind of thing. And then the mobile app uses a completely different sourcing methodology, but still uses grounding and the query fan out itself. I don't think they even do query fan out. It's just the one query for for mobile. So, it's all very different across the board. Um and I'm sure that it will continue to change as the specific underlying models get more reinforced to their experiences and especially where they get more data. This is the big thing. Google has got so much data, so many users that as you've said earlier, if they know that you're clicking on that tab every time that you load this experience, then that tab will become the default, right? Like these things will change over time as we as users force the adoption and force the change.

James Dooley: Yeah, for sure. Anyone who's watching this, are you excited for AI mode? Is AI mode going to become the default search engine? If you don't think it is, make sure you check out the link in the description where me and Charles are speaking about AI mode becoming the default search engine. Make sure you also check out the next one where how to future proof yourself to get in AI mode, to get in Gemini mentions, and to also get in them AI overviews that's powered by Gemini. Charles, it's been an absolute pleasure. Thank you very much.

Charles Flo: Thanks for having me.