GEO: how to get into answers from ChatGPT and Perplexity

A growing share of buying decisions no longer starts in a list of links. Someone opens ChatGPT, Perplexity or Gemini, describes a problem in their own words and gets a short answer with two or three companies named in it. If you are not one of them, you never learn the enquiry existed.

The discipline that deals with this has a name: generative engine optimization. Below is what it actually is, how it differs from classic SEO, and what to do about it.

Why this is a different game than SEO

Classic SEO competes for a position in a list. The user still sees ten results and decides for themselves. A model does the deciding instead: it reads sources, picks the facts it trusts and produces one answer. There is no second page of results to fall back on.

The consequence is uncomfortable. In search you can be fifth and still win clicks. In a generated answer you are either named or you do not exist for that question.

How a model chooses which companies to name

There is no published ranking, but the mechanism is not a mystery either. A model reaches for content it can:

• fetch at all, meaning your site does not block its crawler,
• read unambiguously, so facts are stated rather than implied,
• corroborate, because the same facts appear in other sources,
• match to the question, because the page answers that question directly rather than circling it.

Six things that actually change the outcome

1. Unambiguous facts instead of marketing fog

"Comprehensive solutions for demanding clients" means nothing and cannot be quoted. "We automate invoice processing in Make for companies in Poland and across Europe" can be. A model repeats what it can extract without interpretation.

2. Structured data describing one subject

Schema.org is not decoration. It is a way of stating plainly: this is the company, these are its services, this is where it operates. What matters more than having markup at all is that it describes one coherent subject instead of several loose copies of the same entity. Entities are linked with @id, so a service page can point at the company rather than repeating its description.

3. Access for AI crawlers

GPTBot, PerplexityBot, ClaudeBot and Google-Extended are separate agents with separate rules. Some companies deliberately split crawlers that train models from crawlers that fetch pages while answering a user. If you block everything by reflex, you also block the second kind, which is the one that can cite you.

4. Answers written as answers

A question as a heading followed by a direct reply in the first two sentences is quotable. The same content buried in the middle of a long section is not. This is why FAQ blocks work so well: they match the shape of what a model is looking for.

5. An llms.txt file

A short summary of the offer in markdown at /llms.txt, following the llmstxt.org convention. It costs almost nothing and gives a model a clean, unambiguous description instead of forcing it to infer one from navigation and footers. It is no longer exotic: most serious AI companies now publish one.

6. Mentions in sources the model trusts

Your own site states what you claim about yourself. Third-party sources decide whether that claim is believed. Directories, industry media, profiles and case studies published elsewhere all feed the same confirmation loop.

How to measure it

Not as precisely as positions in Google. Model answers vary between runs and there are no official statistics. What works is a repeatable set of questions, asked before and after the changes, with the answers recorded. That shows the direction of travel, which is the honest claim to make, rather than a number pretending to be a ranking.

Where to start

Ask the models the ten questions a good client would ask before finding you. Write down who gets named. That list, compared against your own site, is the whole brief.

If you would rather go through it systematically, that is exactly what a GEO audit is: visibility check, implementation plan, then a second measurement with the same questions. Get in touch and we will walk through it on a concrete example from your industry.