GEO: getting cited by AI engines

For twenty years, being visible online meant one thing: ranking at the top of Google. Today, a growing share of searches ends without a click. People ask ChatGPT, Perplexity, Gemini or Google’s AI overview, and get a written answer, sometimes with a few sources attached.

So the question shifts: it is no longer just about being ranked, but about being cited. That is what GEO, Generative Engine Optimization, is about.

How a generative engine picks its sources

Without going into technical detail, most assistants that cite sources work in two stages:

1. They search: a query is sent to a web index, their own or a partner’s, to retrieve relevant pages.
2. They write: the model reads excerpts from those pages and composes an answer, citing some or all of what it used.

To be cited, you need to pass both tests: be found at search time, then be useful at writing time. The second one is the new part. A model will happily reuse a clear, factual, self-contained passage, and much less a vague marketing paragraph.

The four levers of GEO

1. Structured content, readable by LLMs

A heading that asks the question, a direct answer in the first lines, then the detail. Lists and tables where the information suits them. Each section should make sense on its own, because the model will often read an isolated excerpt.

Researchers from Princeton and IIT Delhi published an early study on the topic in 2023. It suggests that adding cited sources, figures and precise quotations increases the likelihood of being picked up in a generated answer.

2. Entities and structured data

Engines reason in entities: an organisation, a place, a service, a person. Schema.org markup (Organization, Service, FAQPage, Article…) makes those entities explicit. Consistency matters too: the same name, address and description everywhere you appear.

3. Authority: being mentioned elsewhere

A model trusts information more when several independent sources confirm it. Local press articles, professional directories, public talks, partner pages: every credible mention strengthens your presence in answers.

4. Measurement

GEO without measurement is a gamble. Define a list of questions your clients actually ask, for example “which agency can train my team in AI in Lausanne?”. Put them to the main assistants regularly, and track who gets cited, how, and how that changes.

The technical side: let the right bots in

An often-overlooked point: many sites block AI engine crawlers, sometimes without knowing it. If the goal is to be cited, your robots.txt should allow the ones that feed search, such as OAI-SearchBot and GPTBot (OpenAI), PerplexityBot, ClaudeBot (Anthropic) or Google-Extended.

Some sites also publish an llms.txt file: a plain-text summary written for models, presenting the organisation and pointing to its key pages. The convention is recent and engine support is uneven, but it costs almost nothing.

GEO and SEO: rivals or allies?

Allies. Clear, well-structured, fast content that other sites cite remains the foundation of both. GEO adds one requirement: write passages that stand on their own and genuinely answer the question, without detours.

SEO ranked you. GEO gets you cited.

Where to start

1. List twenty questions your clients ask before contacting you.
2. Ask them to ChatGPT, Perplexity and Gemini, and note who gets cited.
3. Check your robots.txt and your structured data.
4. Rewrite your key pages so they answer those questions directly.

We support businesses in French-speaking Switzerland through each of these steps. See our GEO offer or request an AI visibility audit.