Does Generative Engine Optimisation Actually Work: Live Benchmark Evidence & Verification

Introduction: does Generative Engine Optimisation actually work?

Paul Rowe:

Hello and welcome to the NeuralAdX GEO & AI Search Strategy Podcast, with your host Paul Rowe, who is the Founder, Chief Generative Engine Optimisation Officer and CEO here at NeuralAdX Limited.

So, for this very first podcast, I thought I would answer arguably one of the most important questions in this space: does generative engine optimisation actually work?

Obviously, I’m sure in your research you’ve got so many differing opinions from so many different sources. You would love to have firm conviction that it does indeed work. And from my own empirical experiments, I have evidence-backed proof that it does.

So I’m going to go through that with you now, but there are a lot of details that I have to remember to fit into this particular podcast. I hope I’m able to remember everything and go through everything in an orderly way. But I am human, so bear with me. I’ve got some notes here to support me along the way, but I’m just going to try and explain what I found out in my own investigations and share them with you.

Why NeuralAdX began testing GEO

So it all began on the 9th of March 2025, when I got my initial domain, which was neuraladx.co.uk, when I commenced building the company.

Then I found out in July through my research that AI engines have a preference for .com websites. So obviously, in that instance, I’d spent all this time building the .co.uk website and then realised, to some degree, the AI engines are a bit biased to .com. So no matter how well I optimise the .co.uk domain, I’m always going to fall a little bit short to a competitor that has a .com domain.

So I then had to bite the bullet, so to speak, and with the .com domain I had to commence building a website. And when that actually happened, I thought, well, let me turn a bit of a disaster into something useful.

So I thought, if I’m starting this afresh, this .com new domain, how about off the bat I test it via AI citation tracking software against the most established, well-known businesses in the industry at the get-go, and then I will GEO-optimise it and see how quickly I can compete with them and, if not, potentially surface above them in the rankings and AI engines?

Because in the reports from the Princeton study in GEO, that is indeed possible. But how does that actually work in the real world? Does it really work? And so I thought, because I was in that predicament, why don’t I just go for it? So that’s what I did.

So I commenced the .com website and I basically built up the foundations, just putting all the pages together, the basic infrastructure, not being particularly SEO– or GEO-optimised, just the foundations, as I say, to repeat myself.

And then I thought, okay, well, let me select AI citation tracking software. And at the time, the tracking software was still rolling out, but the best from my due diligence was Otterly.ai. So I signed up with them and decided to commence all of my optimisation and implementing the AI Citation Tracking Software and the AI Answer Visibility and Share of Voice software from Otterly.ai. It’s basically the same thing; it’s just how I term it. And I commenced it all from the 24th of September.

Building a measurable benchmark

So, just to give you an idea of how that works with the benchmarking: with the AI citation tracking software, you will type in your website and then you will put, for example, your 10 most important prompts that you want to win in AI chatbots for.

So somebody types in, “Which restaurant in London is the best?” and you have a restaurant in London, you’re obviously wanting to win for that particular query.

So I made 10 queries that were pertinent to generative engine optimisation: questions of who’s the best company, who’s got the most proof, what’s the pricing, and also to do with education on generative engine optimisation.

So those were the 10 questions that I put in the AI citation tracking software to test and see how my company would do against what I call the established guard.

So it was five companies that I had researched. I’d basically gone through AI engines and typed in all commercial-intent prompts, and whichever companies kept surfacing in all the AI engines — I think it was like five or so that I tested — the companies that kept surfacing at the top, I competed against those.

Because I thought, let’s just see how it really does work against the current best in the industry, against a complete outsider: a new domain with no SEO or GEO optimisation.

The academic foundations of the methodology

So, with that being done, I then commenced the GEO optimisation. Embedded in that is SEO as well, because there are a lot of SEO practices in that. But the GEO optimisation that I did implement was my own cultivated version.

Well, I should be careful when I say that, because all the information that I learned and then practically tested derived from the academic studies.

Now, the academic studies were published in June 2024 from Princeton University, and they worked with the Indian Institute of Technology in New Delhi. That latter particular university is harder to get into than MIT. So, if you can begin to imagine, the intellect of these individuals that are carrying out this study is beyond our understanding.

And they were collaborating together to actually find ways to get companies to surface prominently in AI platforms, and they published in June 2024 their findings on the best way to do that.

So I took that information and I studied it intensely, and then I tried and tested, I don’t know, a mock website on WordPress, and I just went through them step by step and figured out strategies on how to actually implement their findings.

From that, I developed my own methodology, which I call the 11-Factor GEO Framework.

The 11-Factor GEO Methodology

Now, to go through it, I hope I can remember it all because there’s so much information. I’ll go through it one by one.

So the first factor is citations. That basically means, with your content — well, not all of it, but with certain content where it’s relevant — you want to have an actual in-text APA-style citation that links to source information that is backing up your claim or statement wherever that is occurring. So that’s the citation factor. That’s number one.

Number two is statistics, and AI engines love statistics, so you want to implement them in your content where possible.

Number three is quotations. You want to implement quotations because that helps give you credibility. Because if you’re referring to somebody that is respected in the industry and you’re linking to that person, then the AI engine can verify that that’s a stable source and you’re referencing them, and it just helps bolster your authority in the AI engine’s eyes.

The fourth element of the 11-Factor GEO Framework is fluency. And so fluency, in its simplest form, means that when somebody is reading the content, it doesn’t cause them cognitive load. It’s just kind of a smooth, easy process to absorb the information.

The fifth factor is easy to understand, and it’s also, in GEO terms, called content decomposition. And that just basically means, for example, say you had a big paragraph on your website in your content, you would just break that down into maybe three individual paragraphs, or better still, you could break it down into, say, six bullet points.

So you’re just making it, again, really streamlined for the user to be able to read and understand, and kind of coming back to fluency, causing less cognitive load, so it’s just easier for them to understand.

There’s a lot more involved with that particular factor, but I’m just sort of going through briefly each one to give you an idea of what they are.

The sixth one is authority. I’m sure you all know about authority and how that works. That’s to do with topical depth on your particular subject. It also, to some degree, is about your third-party mentions and the quality of those, and it is also connected with your website content structuring and many, many other factors as well. I’m just mindful not to deviate too long on one particular factor. So that’s number six.

Number seven is technical terms and unique words. So obviously, in whatever industry you’re in, where you have your content, you want to make sure that you are using technical terms where necessary and unique words.

So the technical terms are obvious in your particular field, what that would be classified as. But with the unique word, what the AI engines are looking for is where you might say something like “good”, you would say “beautiful”. That’s kind of the variation of what they’re looking for there, if that helps you understand it. So, yeah, that’s technical terms and unique words.

Then the eighth one is schema markup, which I’m sure I probably don’t have to tell you about. Everybody knows about JSON-LD machine-readable schema markup. So we’ll move on from that one.

The ninth one is author bios. So that is basically, with the AI engines, they want to know, with the content that they’re processing, who actually wrote it. They want the security and trust value of being able to isolate an individual person that has credibility rather than an anonymous source. So that’s a key factor that the AI engines look for as well. So that’s the ninth factor.

The tenth factor is source diversity, and that’s just, again, to improve the quality of the content. Because you could have one particular page that is only referencing one particular article. So you might have a few citations backing up your content, as I say, with one particular article.

But the way that AI engines see it, it would be better if you were referencing three or four different articles, because you’ve got all different people, professionals and viewpoints getting brought into that kind of information, if you will. Rather than it just being the narrative of one, it’s got the kind of synergistic narrative of maybe three or four different sources. It’s just providing better content for the reader, so it’s a thumbs-up in the eyes of AI engines. So that’s the tenth factor.

And the eleventh factor is recency. That’s that the AI engines want to see that your content has, self-explanatorily, been updated recently.

And this is really quite important, and it makes a lot of logic because, as you’ll understand, in any industry, everything is always evolving and changing so quickly.

So, when an AI engine is looking for a source to reference, if it looks at a source that has an answer to the question they’re looking for that was six months ago, compared to one that was done and dated, I should say, a week ago, the AI engine knows, well, from an intelligent point of view, the most recent one will probably have information entwined in it that the one from six months ago didn’t — just industry updates and slight changes.

So it’s going to have an intelligent preference for the most recent information. So that also makes a lot of sense.

So, yeah, I’m hoping I haven’t gone too far into the 11-Factor Framework, but I just wanted to explain that that’s what I looked into, and that’s what I skilled myself on, and then I was implementing that from the 24th of September 2025.

Implementation and the 104-day timeline

Now, if you’re with me watching this on video, I want to take you into the actual benchmarking itself, just to validate it and show proof that what I’m saying is actually true.

So when I was doing the optimising from the 24th of September, I didn’t really notice that much to start with. But then, at one particular point on the AI citation tracking software, I think, I believe it was the 4th — yeah, I think it was the 4th of January. We’re going to find out now. The 4th of January, 6th of January, on the AI citation tracking software, we saw a boost, and that was precisely 104 days after I had begun optimising with the 11 factors of GEO.

So I’ll take you into the AI citation tracking software to show you what actually happened, to give you real-world evidence.

What the benchmark showed

So if we select the date being the 24th of November, which is when I started the actual AI citation tracking software tracking the actual prompts, you can see here — and let me do a refresh for you on the video so you can see it’s authentic and it’s not AI-generated or anything like that, unfortunately, because of the modern day.

So, as you can see here, where I started on the 24th of November, there was literally nothing. It had problems sort of picking up myself and the five other companies.

So you can see our company listed there, and these were the five other established guard, as I call it, that I was competing against. And, I mean, they’ve got sort of, you know, 5, 10, 15, 20 years’ experience, and they were marketing themselves as GEO service specialists, so that’s why I competed against them.

So, yeah, back to the results. As you can see here in yellow, while I was doing the optimisation from September, October, and then obviously we started tracking in November, I was not having much success at all.

And then there was a little bit of movement in December, but still I’m way down in the dumps. And then, all of a sudden in January, there was a significant surge which boosted me above all of the competition.

And then, as you can see here, moving from January 2026 all the way up to the current date — well, it’s listed up to the 26th of August there; today we’re the 28th of August 2026 — but you can see here, just in this brand mention section, that the generative engine optimisation allowed me to rise above all of that competition.

So you have to understand, it was a new domain and a completely new website that, within the space of optimising it from the 24th of September up until early January, was nowhere to be seen, and then all of a sudden we’re completely beating all of that competition.

AI citations versus brand mentions

Now, this section here is what I call the AI Answer Visibility and Share of Voice section.

If we move down to the citations — which is what, sorry, this basically covers the AI engines mentioning your name and showing some of your content, etc., in the answers — and further down here is the citation.

So this is the amount of times when an AI engine gives an answer and it includes you: does it include a link to your domain?

And so, if we follow the results here, you’ll see again, sort of in November when I started the tracking, there wasn’t — well, actually, it was half reasonable compared with the others. But as we move along, you can see it significantly improves.

And then on the 4th, I believe — yeah, look, 4th of January — there’s a massive surge, an unbelievable surge there, where I go up to having 87% domain coverage in citations.

And, as you can see, from January 2026 moving all the way up again to date — I mean, the figures only show you the 27th of August. Well, then again, we’re the 28th of August, so it’s always about a day behind, this AI citation tracking software — but you can see it’s absolutely dominated it.

So, from this evidence, I can extrapolate that generative engine optimisation does work, and my 11-Factor GEO Framework, when applied, clearly works.

Why repeated evidence matters

Now, I’ve also bolstered this where, on our website, I have our live AI retrieval testing on the Proof Generative Engine Optimisation Works page.

And that’s where I’ve got three particular prompts that I’ve tested, and I’ve done that for a year now. It comes up on, once again, the 24th of September. I started the live AI retrieval testing on the 24th of September 2025, and we’re now just about to come up to September 2026, so it’ll be a whole year soon.

But I’ve already done five intermittent live AI retrieval tests, and it shows you our performance increasing and keeping the same high level of result, and then sort of dipping a little bit more recently. But clearly, we are surfacing prominently in AI platforms.

So what I’m doing is I’m cross-referencing the results of the AI citation tracking benchmark and then validating it in real-world evidence via the live AI retrieval testing that I’ve done.

So, if you want to have a look at that, I’m going to have all the links to a page that explains what I’ve told you today about our investigation to prove GEO works.

And there’ll be a link also to the live AI retrieval test, and also links to my other live AI retrieval tests where I’ve got short videos, like 25 of them, showing our results.

So, yeah. In essence, that’s what I really wanted to cover in this podcast: just to give you a real-world, honest account of my investigations into it.

And, you know, not just taking my verbal confirmation — take it from the evidence.

Over the period of time that this benchmarking has been done, you can look on our website. We’ve got the AI Citation Benchmark and the AI Answer Visibility and Share of Voice Benchmark, where I’m documenting each month.

And each month I do a live screen recording of the results of the benchmarking, with me talking like this and showing the date. I think on most of them I do. Some of the dates I forget, because there’s just so much to do sometimes.

But monthly I’m coming and showing you that to give you live validation, like: this is really here, this is really happening, there are no sort of edits or any sort of foul play taking place.

So you can see, as open and transparent as I can do. I’m teaching in public, in a sense, to just show you: here are our findings, and here is what is happening, transparently as much as possible.

So, does GEO work?

And so, from this, I have unquestionable belief that generative engine optimisation does work through my own empirical investigation.

So, yeah, I thought I would just cover that today in this particular episode of our podcast because everything encompassed in what I’ve spoken about, and so much more, will be the main core intent of this podcast.

It’s to enrich the knowledge of generative engine optimisation and AI search visibility, and how organisations can do everything possible to ensure that they surface in AI platforms, and hopefully I’ll be attributary to helping companies with that.

Podcast purpose and closing reflections

Okay. All right, I think that’ll be all from me in this episode.

I thank you so, so much if you’ve taken the time to watch, and I hope that my, somewhat I could say, my life’s work for the last year has been interesting and has given you some security and some sense of an honest, evidence-backed view, which seems to unfortunately be so difficult to find in the modern world.

If anyone has any questions or queries, feel free to ask me if this is appearing on YouTube and other channels, or you could — well, not email us, because that’s for work queries — but you get my point.

Any communications where I can communicate with you, I would like to, as I’m trying to grow the podcast and everything associated with the company, as you would expect somebody like myself to do if you’re in my position.

Closing

Okay, thank you so, so much indeed for watching, and I look forward to seeing you in a new podcast.

Thank you so much indeed. Take care now. Bye-bye.