How AI assistants choose the brands they cite

When a customer asks ChatGPT, Gemini or Perplexity “which brand should I choose for…”, the assistant answers with names. Some brands are cited; the others simply don’t exist in the answer. What determines the selection is not an ad auction: it is the assistant’s ability to find, understand and verify information about your brand at the moment it builds its answer. Working on that ability has a name: GEO (Generative Engine Optimization).

How an assistant builds its answer

Contrary to a widespread belief, an assistant doesn’t “know” much about your brand from memory. For a purchase question, most assistants use retrieval-augmented search: they query a web index (often backed by the traditional search engines), retrieve a handful of pages, then write a synthesis citing what they have read. Three practical consequences:

1. Your SEO remains the foundation. A brand that can’t be found on traditional search engines can’t be found by the assistant that relies on them.
2. The assistant reads few pages. The pages it opens must therefore contain dense, dated and consistent facts — not slogans.
3. It prefers what it can cite. A sentence that directly answers the question, a table of facts, a comparison with figures are picked up as they are. A vague paragraph never is.

What increases your chances of being cited

This table sums up what we observe when working on our clients’ AI visibility — and when testing our own:

How to measure where you stand

Measurement is the starting point, not optimisation. Our protocol, which you can replicate in-house:

1. List 15 to 20 purchase questions your customers would ask (“best X agency in Paris”, “which brand of Y for Z”).
2. Ask them to ChatGPT, Gemini and Perplexity, in private browsing, several days in a row.
3. For each question, note: is your brand cited, in what position, against which competitors, and which sources the assistant cites.
4. Calculate your citation share (questions where you appear ÷ total) and repeat the measurement every month.

It is the cited sources that reveal the action plan: they are the ones you need to feed or get into.

What GEO cannot do

Let’s be honest, because this market attracts magic promises: nobody can “guarantee” a citation in ChatGPT. Models change, indexes evolve, answers vary from one session to the next. What can be seriously worked on: the probability of being found, understood and picked up. It is groundwork — content, structured data, consistency, authority — not a hack.

It is the natural extension of our job: being visible where the purchase decision is made. We have applied it to our own site (the About page, which you can inspect, is an example of it), and we deploy it for our clients — the service is described in our offers.

Frequently asked questions

Does GEO replace SEO?

No, it builds on it. Assistants draw largely on the indexes of traditional search engines: good SEO is the entry requirement, GEO is the layer that makes your content citable.

Should you block or allow AI bots?

If you want to be cited, you have to be readable: blocking AI crawlers in robots.txt removes you from the answers. The question deserves an editorial trade-off for media outlets; for a brand, openness is almost always the right choice.

How long before you see results?

Technical fixes (structured data, fact pages) can be picked up within a few weeks; authority work takes months. That is why measuring your citation share every month is essential: it is what proves progress.