The Interview

Lili: Hello, James Berry, founder and CEO of LLMrefs, I'm super excited to have you here. This new area of GEO searches and GEO optimization is a little bit overwhelming for a lot of people. So we want to give them more hands-on ways of what to do with these tools and how to get the LLMs to rank their brand, website, or content. I would love to ask you a few questions if I can jump in right away. If a user takes a look for the first time at your platform, for example LLM Refs, what should they look at first and how should they try to interpret the data?

James: Yeah, well, firstly, thank you for having me on the show. I think it's really interesting. We're in a brand new market and there's a lot of misconceptions that people have when they start looking into AI search tracking and analytics. The first biggest trap that I see people doing when they're getting started is treating all of the data just like classic SEO rankings or web analytics. And the truth is it's just not that. LLMs are not ranking in a single list just the way that Google does. They're non-deterministic, meaning that if you ask the same question multiple times, you will get a very unique and completely different response every single time.

So the first thing I would say that you need to understand is that visibility in AI search is not the order in which you appear for one prompt, but instead the frequency in which your brand appears over many responses across a broad range of prompts. The second thing that someone new to GEO should understand is that AI models are using live web search to answer questions, which means SEO is absolutely not dead. Using LLM Refs, you can see exactly which web pages are being used by the LLM to generate those responses and you can also see how often they're getting cited.

Lili: So what you were saying is basically it's not always the same prompt. Obviously people are prompting in different ways. It's more like the meaning, right? Various questions that mean the same thing might show similar answers.

James: Yeah, it can depend. You could ask the same prompt in a different way and it could recommend a different brand or a different solution to your problem. But you could equally ask the same question twice and you can try this right now. If you go into ChatGPT and you ask a question and you just copy and paste exactly the same prompt, it doesn't always guarantee that the same brands will show up and it definitely doesn't guarantee that the same result will always be first.

So when you're looking into tracking visibility, it's really important that you're tracking across a huge range of different prompts and tracking those prompts multiple times so that you can get this statistically significant understanding of how you're being viewed in AI search.

Lili: Well, basically that means it doesn't help if I'm prompting for myself, right? I actually need your tool.

James: I'm biased, but yes. Sadly not. It's not like Google where the responses change once a day or however often depending on which keyword it is. The responses change every single time.

Lili: Can you give us three actionable tips website owners can do right now to make their content rank better? Not a big strategy, but more like I want to rank better and see results as soon as possible.

James: Yeah, for sure. This is the best question you could possibly be asking. The first thing that anybody needs to do when starting to think about AI search is ensuring that your current content is crawlable by AI. By that, I mean just making sure that AI bots can actually read your content.

I recommend first starting by checking that you're not blocking these AI crawlers in your robots.txt file, but you also need to be checking that your server is not configured to completely reject these bots. We're seeing this really, really frequently since Cloudflare flipped the switch and changed the default configuration to block all incoming AI bots on your site. We actually have a free tool available on the LLM Refs website that you can use to check this.

The second most important thing you need to be looking at is fan-out queries. This is the concept that when an AI makes a live web search, it doesn't just paste your full seven-sentence prompt you just typed into ChatGPT. It actually extracts this into smaller sub-queries, typically one to five of them, made up from keywords in your prompt.

So you need to make sure that you have content ranking for all of these sub-queries and that you're using terminology in your content that matches each of these phrases. There are Chrome extensions you can use to find this in ChatGPT. On Perplexity, they show you what the queries are. On Gemini, you would need to use an API to understand what they are. We track these all on LLM Refs, and we also have some free tools for synthetically generating these just to give you some ideas to get started.

The final most important thing you need to be doing is ensuring all of your content is fresh. AI has this huge recency bias. What we're encouraging people to do is ensure that you're revisiting your own content at least once per quarter and refreshing it with current updates. What I can tell you from our data is that the moment your content becomes over three months old, the AI citations to that page just drop off a cliff. So it's really important that you're keeping things fresh and refreshing your content with recent information.

TRY LLMREFS FOR FREE NOW - 7 day free trial with 10 prompts and ALL LLM breakdown

Lili: I think that totally makes sense because how would the LLM know that I'm the original source or that maybe I've just copied it from everybody else? As soon as I generate content that is very unique with very new insights and it comes out, others will copy it from me and then the LLM won't know which one was actually the original person saying that and it will not quote me, but just somebody who quoted me.

James: Exactly. It's a really tough problem to solve. I think just keeping up to date, relying on the fact that all of these AI systems are backed by traditional search engines to surface the most relevant content. Just making sure your content is genuinely helpful. I know that's the worst piece of advice you could ever receive, but building helpful content is a massive help.

Lili: Perfect. So now we've optimized, we've taken these low-hanging fruits and we want to go a little bit deeper and look into your platform or any of these GEO tool platforms. What would be the most important metrics that actually tell us we're doing well or we have a lot of potential to improve?

James: How I think about this is I've split it into two halves. I've split this into what the webmaster, what the agency that builds your website should be looking at to understand that you're doing really well in AI search, and then what you should be looking at as a brand owner to understand how you're performing in AI search.

Starting with the webmaster side first, as we mentioned already, you need to be sure that you're not blocking these AI crawlers. But secondly, you need to see if your site is actually getting cited in AI search. You can do this really easily by checking your server logs for this user agent called ChatGPT-user to see if there are any incoming bots reading your site. The other thing we see really, really frequently is you need to make sure that the content on your site is not hidden by JavaScript.

AI bots don't browse like normal humans do. They can only read the text that's in the HTML returned by the server. That means if you're building a single-page app, or you have a button that you have to click or a tab or a dropdown in order to see some content, AI will not be able to read that.

An example we had is I worked with a customer recently where their second most cited page in AI search was their pricing page. But all AI could see on this page was literally just the loading spinner. That caused the AI systems to pull outdated information from a competitor's landing page and it actually made them look like a far worse value for money option. Really low effort, high impact things you can do to improve your visibility.

Looking at the flip side, the brand side, this is the most interesting part. Sadly, most of AI search is just zero-click. So tracking click-through rates with the ChatGPT referrer in Google Analytics or using traditional attribution models sadly just doesn't work anymore. The first thing you should be tracking is share of voice. We can think of this as a mention rate. How frequently is my brand being mentioned in AI responses? The higher this percentage goes, the more impressions you are receiving in LLMs.

The second thing you should be tracking is rank. It's all good knowing your share of voice, how frequently you're appearing, but you want to know how you compare against your competitors. This can help surface optimization opportunities and insights. Importantly, you need to be tracking these over time to understand if your AI search optimization efforts are helping your brand grow or if they're less effective and you need to go back to the drawing board.

TRY LLMREFS FOR FREE NOW - 7 day free trial with 10 prompts and ALL LLM breakdown

Lili: Which LLM mention is the most valuable? If ChatGPT mentions me, is that the big crown because it's maybe still the most common LLM, or do we have to think about this in another way?

James: It's a really great question. I think about this in two different ways. Firstly, ChatGPT, yes, 100%, you're bang on the money. It's the largest LLM. They currently have 70% market share. So we need to be taking advantage of that and focusing most of our efforts on ChatGPT. Secondly, Gemini. This is the fastest-growing LLM and the distribution that Google has—we can't forget about it. They own Chrome, everybody's on Gmail.

Gemini can be everywhere at just a click of a button. The other one that I think is really important and don't want to forget, especially if you're working in SaaS, is Perplexity. From the data we're seeing, the highest conversion rates come from Perplexity users. So you should definitely be optimizing for that one as well.

The second answer to this, I know it's not really what you asked, but something I also try to think about when thinking about which LLM mentions are most valuable is what sort of prompts or mentions are most valuable to brands. I would encourage our customers and users to really only be interested in bottom-of-funnel prompts.

These are the prompts where somebody's comparing one product to another, trying to make a purchase of something. Basically anything that is high intent for the user taking a next action. These are the prompts we encourage people to track and optimize for.

Lili: That's really good advice. Totally makes sense. So now let's say we are getting a lot of traffic, but how are we going to turn that into money? All this work has to get paid.

James: Quite right. It's a really difficult question because it depends on who you are, who your organization is and the niche that you're currently building content for. It's also a really difficult question because the industry is changing really fast and whatever answer I give you today might not be true in a week's time.

That being said, I had a little think about monetization models, and one that I think can be quite difficult that I am seeing is paywalled content. If you're blocking AI crawlers, they can't read your content. This is sort of like an intentional block by paywalling your content and therefore the LLMs can't read it. And if they can't read it, they can't cite it. And if they don't cite it, you're not getting mentioned in AI responses.

So if you were already building a site and you already had impressions, I would probably refrain from starting to paywall content.
That being said, if I were a content creator right now, my go-to strategy would probably be building listicles and monetizing via affiliate links.

Lili: Like my page, you mean?

James: For sure. Yeah, you're doing it right. I'm sure there are hundreds of ways to monetize these impressions. Attention is currency nowadays. That being said, we know that listicles are working phenomenally well at this moment in time in AI search. That might not always be the case. Affiliate links on those products—you could argue paid opportunities for brands to be included in them could work out as well.

Lili: Perfect. Then what's a good GEO strategy for website owners and content creators? How to look at it in an overview and more in a strategic way?

James: I actually had three things immediately come to mind. The first one is something not to do. I would absolutely recommend against mass automated AI-generated content and posting pages. I see this absolutely everywhere. One, it's just absolutely inauthentic and a real user reading this content can see through it a mile away. They're not going to trust your site and they're going to bounce immediately.

But secondly, I actually think this is going to be very bad for traditional Google SEO. They will come back at you and we're already seeing it, especially over the last week with increased volatility in the SERPs. They will be fighting back at sites that are just abusing AI-generated content with little to no value for true users. So definitely avoid that.

On the flip side, something you should continue investing in, as we all have for the last 20 years, is building backlinks. I've got two reasons for continuing to do this. Firstly is the obvious one—the LLM connects to traditional search via search grounding. So increasing the number of backlinks increases your rank in the SERPs, improving your chances of getting cited.

But secondly, there is new data showing that the more links your domain has, the more your brand will appear in the LLM training dataset called Common Crawl. This is a dataset, an index of the entire web. Effectively the more links you have, the more frequently your brand will appear in that dataset. And the more you appear, the more the LLM knows about your brand and therefore it's more likely to trust you and recommend you in AI answers.

Lili: That is so clever. The training set absolutely totally makes sense.

James: It's amazing, really cool. My final tip for AI search optimization is to just honestly find out what's already being cited frequently in real time and then get your brand into it.

This could be literally as simple as commenting in a Reddit thread that's already regularly being cited, or finding blog posts from other publishers and emailing the authors asking for your brand to be included in the article. This works really well and it's the fastest way to lift your visibility. I've seen brands go from completely invisible to getting their first mentions in AI search in under 60 minutes using this method.

TRY LLMREFS FOR FREE NOW - 7 day free trial with 10 prompts and ALL LLM breakdown

Lili: Amazing. I love these nuggets. Another thing I found when I was testing with your tool—it's so weird. How come for the same search term or prompt query, maybe I'm cited a lot more in America than in Britain? It's the same language. How does that happen and how can I optimize for that?

James: Why do you appear for the same prompt in the US and the UK very differently even though they're the same language asking the same question? This is again because of the search grounding process. When the AI goes to Google, it'll be going to the UK version of Google and then the US version of Google, and they can have very different search results.

Lili: We've just recently moved my website from Squarespace over to Webflow. With this move, my SEO guy said you need to add a lot more structured snippet or markup schema codes in the backend. So we've created so much of these codes. Do you also think that's a really relevant thing for the LLMs? Are they looking a lot into the backend of the website?

James: This is a highly contested topic and I'll give you my honest opinion. What we know is that LLMs, when they come and search your site in real time and visit your site, they're not reading the structured markup on your page. They're just reading the text. But what we also know is in order for the LLM to find your page, you need to be appearing high in the SERPs.

To get there, we need to appease Google and Bing and whatever other search engine we're using for grounding. I think Claude is running with Brave Search at the moment. In order to get there, I would recommend adding all of that structured schema to increase your ranking. What's unclear, and this is where there's a lot of speculation, is if that structured schema is used at the training stage of the models.

Lili: Thank you so much, James. I have to say I've tested five GEO tools and LLM Refs was so far ahead. You could tell from what we've been talking about—you know the stuff. Last tips and tricks and maybe some nuggets if you have any for us?

James: There are hundreds of things to recommend. The key thing for AI search is firstly getting an understanding of where you currently sit. When you've got an understanding of where you currently sit and why you rank and why you do not rank, then you can start making changes and affecting the AI responses.

There are a hundred different rabbit holes here, and I'm sure there will be more as time goes on. This is the most exciting space in my opinion. Just keep up to date with what's coming out on LinkedIn. We've got a blog where we're posting as much content as possible. Come and say hi if you ever want to chat about AI search.

Lili: Amazing. Sounds perfect. I will come back with all these new insights. I will implement them and then I will have new ideas and new questions for you. Thank you very much for having me on the interview.

TRY LLMREFS FOR FREE NOW - 7 day free trial with 10 prompts and ALL LLM breakdown