Visibility in AI search: a GEO guide for ChatGPT and Gemini

To be visible in AI search, your site first has to be sound in classic search — and then its content has to be written so that an AI tool can quote it easily: pages that answer the question directly in the first paragraph, with clear definitions and a visible author and date. This is called GEO (Generative Engine Optimisation). There is no secret trick — only solid technical foundations, plainly written content and consistent information about your brand.

People no longer put their questions only to Google. They ask ChatGPT, Gemini and Perplexity too, and these tools often list their sources beneath the answer. In this article we explain what it takes to become one of those sources — and what it does not.

What is GEO, and how does it differ from SEO?

GEO is the work of getting your content used as a source in the answers that AI-assisted search tools generate. It does not replace classic SEO; it is built on top of it.

	Classic SEO	GEO
Goal	Ranking high in the list of search results	Being cited as a source in the generated answer
What the user sees	Ten links, each with a title and description	A single answer and a handful of source links
Valuable content	A thorough page that matches the query	Short, clear, quotable passages
Foundation	Crawling, indexing, links	The same foundation, plus clarity and trust

One important fact: most AI search tools find current web pages through a search index or their own crawlers when building an answer. A page that is invisible in classic search, cannot be crawled or loads slowly has little chance of being cited by an AI tool either. That is why we always begin GEO work with a technical SEO checklist.

How do AI search tools choose their sources?

On its own, a large language model knows only the data it was trained on. To answer a current question, AI search tools generally follow these steps:

Turn the question into several search queries.
Find pages relevant to those queries.
Pick out the passages within those pages that best answer the question.
Combine those passages in their own words and link to them as sources.

The third step is what GEO is really about. The model uses parts of a page, not the whole of it. A page whose answer is buried halfway through the third paragraph — or that opens with a vague introduction, or tries to persuade you to buy before it informs — is far less quotable than one that starts with a clear answer.

Writing quotable content

These are the principles we apply in our own articles:

A direct answer in the first paragraph. If the title is a question, the first two or three sentences answer it. Detail and context follow. This habit also serves Google’s featured snippets.
Sections that stand on their own. The text under each heading should make sense without the rest of the page. A paragraph that begins “as we explained above” loses its meaning when quoted.
Clear definitions. Short, self-contained sentences of the form “X is…”.
Tables and lists. Comparisons and steps are far easier to parse than running prose.
Concrete knowledge. Your own experience, examples and measurements. Rewriting what other pages already say gives an AI tool no reason to cite you.
Frequently asked questions. Questions phrased the way people really ask them, with short answers.
Author and date. Who wrote the page and when it was last updated should be plain to see. Information that ages should be refreshed regularly.

An example: imagine a page answering “When should I see a psychologist?”. The weak version opens with the history of psychology and gives its answer halfway down. The strong version lists, in the first paragraph, the situations in which seeking support is advised, then covers each under its own heading and closes by saying where to turn in an emergency. The second page is more useful to the reader — and far better suited to being quoted.

The technical side: crawlers, structured data and llms.txt

Allowing AI crawlers

Each AI company runs its own crawlers. Some gather data to train models; others find sources for search answers. You should choose deliberately what you allow in your robots.txt file. You may not want to contribute to training, yet still want to appear in search answers — and most companies use separate crawler names for those two purposes. Blocking all of them blindly can switch off your search visibility as well.

Google is a slightly different case. Its AI-assisted search features run on the ordinary Google Search infrastructure, and Google states that no extra markup or special file is required to appear in them — the fundamentals of SEO apply. The details are on Google Search Central’s page on AI features.

Structured data

Structured data describes, in machine-readable form, who wrote a page and for what purpose: organisation details, person, article, service, frequently asked questions. It clarifies meaning for classic search and for other systems that parse content. The markup must say the same thing as the content visible on the page.

llms.txt

llms.txt is a proposed file, placed at the root of a site, that summarises the site’s most important pages for language models in plain language. It is easy to set up and describes the site’s structure clearly. But in honesty: there is not yet any general commitment that every AI tool reads or takes account of it. We see llms.txt as a useful but complementary step — it does not stand in for content and technical foundations.

Brand and trust signals

When deciding whether a source can be trusted, AI tools look not only at the page but at how the brand appears across the web.

Consistent information. Business name, address, services and contact details should match on your site, your social profiles and your business listings.
Visible expertise. It should be clear who the author is, what field they work in and why they are writing on this subject.
Being mentioned elsewhere. Mentions of your brand in trade publications, directories and communities build trust here, just as they do in classic search.
Depth on a topic. A set of linked articles that treat one subject from several angles shows that the site is a genuine source in that field — far more than a single post.

Measuring the effect

GEO is harder to measure than classic SEO, because not every tool shares detailed data. Still, there are places to look:

Referral sources. Visits from AI search tools may appear as distinct sources in your analytics. Grouping that traffic separately shows the trend.
Manual testing. Draw up a list of questions your customers are likely to ask, put them to different tools regularly and note whether your site is cited.
Search Console. Impressions and clicks from Google’s AI features are included in the general web search data, so changes can be tracked page by page.
Conversion. Whether these visitors go on to act — a contact form, a booking, a sign-up.

We share the results in a short monthly report. Appearing in a given tool’s answer is not something anyone can guarantee; we measure what can be measured and do more of what works.

Frequently asked questions

Should we drop classic SEO for GEO?

No. AI search relies heavily on pages that are crawlable, indexed and trusted on the web. Classic SEO is the foundation of that visibility; GEO is a layer added on top.

Is adding an llms.txt file enough?

No. llms.txt is a useful proposal that describes your site’s structure to language models, but there is no general commitment that every tool uses it. The real effect comes from crawlable pages, clear content and correct structured data.

No — and nobody can honestly promise that. How AI tools choose their sources depends on their own systems, which keep changing. We do the work that makes your site a strong candidate, and we measure the results.

What happens if we block AI crawlers?

If you block the crawlers that find sources for search answers, your site becomes less likely to appear in those tools’ answers. If you simply do not want to contribute to model training, it is usually possible to block only the training crawlers.

If you would like to look at how your site appears in both Google and AI search, see our SEO and maintenance service or write to us. In the first review, we show plainly where the largest gains can be made.