I've worked with websites and SEO for years, but recently I've been trying to get my head around GEO — Generative Engine Optimisation — and what businesses may need to do differently as more people use ChatGPT, Gemini, Perplexity and AI-powered search to find information.
I'm still learning. In fact, some of the most useful things I've discovered have come from experimenting on a real UK service website rather than reading another list of GEO tips.
I started working through some GEO recommendations and initially approached them with an SEO mindset. Is the site structured properly? Are the pages clearly labelled? Does the content explain what the business does? Are there useful external links and signals?
Some of that still matters. Google itself says that the fundamentals of SEO remain relevant to its generative AI search features. GEO isn't a replacement for SEO.
But as I worked through the recommendations, I noticed something that changed the way I was thinking about the website.
The questions weren't only about whether an AI could find the business.
They were increasingly about whether it could understand enough about the business to use it in an answer.
Being found isn't quite the same as being understood
Traditional SEO has always involved much more than keywords. Technical structure, useful content, links, authority, relevance and user experience all play a part.
But there is still a fairly familiar journey behind much of it. Somebody searches for something, the search engine presents possible destinations and the user chooses where to go next.
Generative AI can change that journey.
Someone might ask an AI a much more detailed question containing their circumstances, requirements, budget or deadline. The AI can potentially break that question into parts, gather information from different sources and construct an answer.
Google says its AI features are designed for these more complex questions and can use a technique it calls "query fan-out", issuing multiple related searches across different subtopics and data sources before generating a response.
That made some of the GEO recommendations I was seeing make more sense.
The system wanted greater clarity around things such as:
How much does the service cost?
How long does it take?
Exactly how does the service work?
Where is it available?
Who is providing it and what are their credentials?
What independent evidence exists to support what the business says about itself?
Initially, some of those questions can look like ordinary FAQ material. But together they provide something more useful: context.
If an AI is helping someone understand their options rather than simply presenting ten links, it needs enough information to work out whether a particular business is actually relevant to the situation.
Could your business become part of the answer?
This is where I've started to see GEO slightly differently.
Imagine somebody planning a project through a generative AI assistant. They might ask what they need, what order they should do things in, how long different stages could take, roughly what they should budget and what kind of supplier they need.
For a business to become useful in that conversation, simply stating "we provide this service" may not give the AI very much to work with.
Price ranges, timescales, geographical coverage, processes, limitations, qualifications and evidence potentially help an AI understand where that business fits.
I'm deliberately saying potentially because I don't think we're at a stage where anyone can confidently give businesses a universal checklist of GEO ranking factors.
The research is still developing. A recent critical review of GEO research found that the evidence remains narrow and that no reviewed technique had yet demonstrated a stable, long-term, cross-platform causal effect on organic discoverability and subsequent user behaviour.
Different generative systems also retrieve, use and cite information differently. Recent research has even proposed separating whether information is selected as a citation from whether that information is actually absorbed into the generated answer.
So I'd be suspicious of anyone presenting GEO as a solved formula.
Removing ambiguity may matter more than adding content
One of the more interesting problems I encountered wasn't a lack of content. It was ambiguity.
The optimisation tool repeatedly generated questions based around a particular location, even though the business provides a postal service across the UK.
A human looking around the website could probably work that out.
But it made me think about what happens when an AI has to decide whether that business is suitable for somebody hundreds of miles away.
The answer shouldn't be another twenty slightly different FAQ questions mentioning different towns. That starts heading towards exactly the sort of repetitive content I don't want to put on a website.
Google's latest guidance is useful here. It explicitly warns against creating separate content for every possible variation of how people might search simply to influence rankings or generative AI responses. Instead, it recommends useful, original, people-first information.
So I took a more selective approach.
Where a question exposed a genuine gap, I tried to answer it. Where several questions were essentially asking the same thing, I didn't see much value in repeatedly answering them just because a GEO tool suggested it.
That's where I think human judgement still matters.
An AI may want evidence, not just claims
Another thing that caught my attention was external validation.
It isn't difficult for a business to write flattering things about itself. We're experienced. We're trusted. We're experts. We provide excellent service.
But why should an AI believe us?
The recommendations I encountered were interested in reviews, credentials and other information that could help substantiate what the business claimed.
There is emerging research suggesting that third-party sources can play an important role in AI search. One large experimental study of generative search found substantial differences between traditional Google results and AI-search sources, with the AI systems in that study showing a stronger preference for earned, third-party media.
I wouldn't turn that into a rule that every business now needs to get mentioned by newspapers to appear in ChatGPT. The systems differ and the evidence is evolving.
But the underlying question seems sensible:
Can something outside your own website support what you're saying about yourself?
That could make reviews, recognised credentials, consistent company information and genuine third-party references increasingly interesting parts of the picture.
Don't optimise for the score
After making changes, I started seeing signs that the site's AI visibility was changing.
The GEO score itself remained relatively low. AI-related query activity increased, and some tests suggested improved visibility on generative platforms.
Initially, that looked encouraging.
Then I looked more closely.
Some of the ChatGPT activity could have been generated by me because I'd been using ChatGPT while working on and reviewing the site. More importantly, the analytics showed no attributable site sessions from ChatGPT at the time I checked.
So I can't say, "I improved the GEO and AI traffic went up."
I don't know that.
And that's probably another useful lesson.
AI visibility, AI crawling, being cited in an answer and an actual person visiting your website aren't necessarily the same measurement.
Even the measurement tools are evolving. Google rolled out dedicated generative-AI performance reporting in Search Console during 2026, allowing site owners to examine visibility within its generative Search experiences separately.
We may need to become more sophisticated about what success means rather than replacing "Where do I rank on Google?" with "What's my GEO score?"
So what should businesses do now?
I'm not about to suggest rewriting your entire website for ChatGPT.
If anything, what I've learnt so far points in the opposite direction.
Keep doing the fundamentals well. Make the site technically accessible. Explain what you actually do. Produce useful original content based on real expertise. Make important information available as text rather than hiding it from crawlers. Google explicitly says these existing SEO fundamentals continue to apply to its generative AI experiences.
Then perhaps look at the site through a slightly different lens.
If an AI had to decide whether your business was suitable for someone's particular situation, what would it struggle to understand?
Would it know your prices or at least realistic starting points? Would it understand your geographical coverage? Could it work out the process? Are your timescales clear? Are qualifications or credentials explained rather than merely claimed? Is there independent evidence that supports what you say? Are important limitations clear?
Those are useful questions for people too.
And perhaps that's the reassuring part.
Preparing a website for a more generative-AI-led world may not require filling it with strange phrases written for machines. It may mean doing a better job of making the business clear, specific, useful and verifiable.
I'm still experimenting with GEO, and I expect my understanding will change as the technology and evidence develop.
But I've already changed the question I ask when looking at a website.
It's no longer only:
Can somebody find this business?
I'm increasingly asking:
Have we given them — and the AI helping them — enough information to understand when this business is actually the right answer?