Getting cited by ChatGPT: how to position your company in AI search

For twenty years the goal was to come first in a list of blue links. Today a growing share of your customers never sees a list: they see an answer, with two or three sources cited underneath. Either you are in those sources or you do not exist.

Getting cited by ChatGPT: how to position your company in AI search

This has a name now: GEO, generative engine optimization. The term comes from research out of Princeton, Georgia Tech and IIT Delhi that measured what makes a model cite one source over another, and found visibility gains inside generated answers of up to 40%.

It does not replace SEO; it sits on top of it. The difference is what is being contested. SEO competes for a position on a results page; GEO competes to be the source the model quotes inside its answer.

1. First: check you are not blocking them

AI search engines use their own crawlers. The ones that matter today: GPTBot and OAI-SearchBot from OpenAI, Google-Extended for Gemini, PerplexityBot, and Anthropic's ClaudeBot.

Here is the most expensive and most avoidable mistake in the whole subject: around 7% of sites accidentally block at least one major AI crawler, almost always through an inherited robots.txt rule somebody copied years ago. Open yoursite.com/robots.txt right now and look. It is a text file and the check takes thirty seconds.

2. llms.txt: an index written for models

The llms.txt convention is a plain text file at the root of your domain pointing at your most important, best-explained content — like a sitemap, but curated for language models. Nearly 70% of sites have none.

Let us be honest about what it does and does not do: on its own it will not win you citations. What it does is remove ambiguity — making clear which page is your canonical explanation of a service, instead of letting the model choose between seven similar ones. This site has had one from the start, at silverostudios.com/llms.txt.

3. Write so you can be quoted

A model cites passages, not pages. Content that quotes well has a specific shape:

  • Answer before context. If the headline is a question, answer it in the first paragraph. Three paragraphs of preamble before the number is a passage nobody can extract.
  • Information only you have. A real price, a real lead time, something you measured. Repeating what ten other sites say makes you the eleventh.
  • Headings that name things, not clever ones. «What a website costs in Spain» can be quoted; «Let's talk budgets» cannot.
  • Structured data in JSON-LD, which is what turns prose into machine-readable facts.
  • Visible dates. Perplexity in particular strongly prefers recent content, and an undated article is an article of unknown age.
Writing to be cited by an AI and writing well for a person have turned out to be the same thing: say what you know, say it early, and say where it came from.

4. Knowing whether it works

The awkward part: many of these visits do not leave the trail you are used to. Somebody reads your name in a ChatGPT answer and types your domain straight in. That shows up as direct traffic and looks like magic.

What you can do is tag model referrers in your analytics so sessions from ChatGPT, Perplexity, Gemini and Claude are separated out. And the simplest thing of all, which most people have never done: ask ChatGPT and Perplexity about your own service in your own city and see who they cite. That is your starting position.

If you want us to look at this on your site — robots.txt, structure, data — get in touch. It is a short review and something usually turns up that can be fixed the same day. And if you came from classic SEO, the old discipline still rules locally: we covered that in our piece on local SEO.

Sources

  1. GEO: Generative Engine Optimization (Princeton, Georgia Tech, IIT Delhi)
  2. llmstxt.org — the llms.txt convention
  3. OpenAI — GPTBot and OAI-SearchBot
  4. Anthropic — Claude documentation
  5. Google — crawlers and Google-Extended
Back to blog

Shall we talk aboutyour project?