Getting cited by ChatGPT and AI engines: what really matters

How AI assistants choose their sources

“GEO SEO” is searched 2,900 times a month in France, “référencement IA” (AI SEO) 390 times. GEO (generative engine optimisation) is the work of getting picked up and cited in AI assistants’ answers. The principle is simple: when ChatGPT, Perplexity or Google answer with a web search, they read pages, extract facts from them and cite their sources. To be cited, you have to be read, then be the clearest page on the question.

Google’s AI answers are already here: of the 25 trade searches we recorded on Google France, 6 showed an AI-generated overview.

Step 1: let AI crawlers in

Each assistant has its own crawlers, which your robots.txt file allows or blocks. OpenAI documents several, each with a distinct role:

• OAI-SearchBot powers ChatGPT search. OpenAI states that a site blocking it does not appear in ChatGPT’s search answers, except possibly as a plain navigation link.
• GPTBot collects content to train the models. Blocking it signals that you refuse this use; it is a setting independent of the previous one.
• ChatGPT-User visits a page when a user asks a question that requires it; it does not crawl the web automatically.
• For the AI features of Google Search (AI Overviews, AI Mode), Googlebot is what counts. Google-Extended only governs the use of your content for training and grounding in some other Google systems.

In our study, 2% of sites block at least one of these crawlers. Check yours: a copied robots.txt template or a hosting setting can close the door without you deciding it. Google points out that crawling must also be allowed by your CDN or hosting, not just by robots.txt. Finally, OpenAI says to allow about 24 hours after a robots.txt change for its systems to take it into account.

Step 2: get indexed and rank well

Google is explicit: to appear as a link in AI Overviews or AI Mode, a page must be indexed and eligible to show in Google Search with a snippet. There is no other technical requirement, and no special optimisation needed. Google adds that there is no need to create new machine-readable files, AI text files or special markup.

In practice, GEO therefore starts with SEO: a page that is not indexed, or that carries a nosnippet tag, cannot serve as a source. The good practices Google cites for its AI features are the usual ones: crawling allowed, content findable through internal links, a good page experience, important content in text form, structured data that matches the visible text, an up-to-date Google Business Profile.

Step 3: write pages that AI can cite

A generated answer assembles short sentences taken from several sources. The easiest pages to cite are those that give the answer straight away, with verifiable facts.

Research has measured it. In the paper that introduced the term GEO, presented at the KDD 2024 conference, Pranjal Aggarwal and his co-authors tested several ways of rewriting a page. Adding citations of sources, expert quotations and statistics is among the methods that increase a source’s visibility in generative engine answers, by up to 40% according to their measurements.

1. 1
   Answer the question in the very first sentence of the page or section, in a sentence that stands on its own.
2. 2
   Give precise, true facts: area served, lead times, opening hours, prices if you display them, certifications actually obtained.
3. 3
   Cite your sources whenever you state a figure or a rule: official text, study, organisation.
4. 4
   Add an FAQ with the questions your customers really ask on the phone.
5. 5
   Describe the business in structured data (LocalBusiness or a more specific type), with the same information as the page and your Google Business Profile.

The llms.txt file: useful, not mandatory

The llms.txt file is a proposal published in 2024 by Jeremy Howard: a Markdown file placed at the root of the site that summarises the offer and points to the useful pages, to help AI agents use a site. It is an emerging convention, not a standard, and Google clearly states that no such file is needed for its AI features.

Our audit checks it with a low weight, as a minor point: it only takes a few minutes and replaces nothing above. 80% of the sites in the study do not have one.

Measuring your presence in AI answers

Google states that visits from its AI features are counted in the search traffic of Google Search Console, in the Performance report, search type “Web”. For ChatGPT and Perplexity, look in your analytics tool for visits whose source is chatgpt.com or perplexity.ai.

Also test by hand, every month: ask the assistants your customers’ questions (“best accountant for a property holding company in Lyon”, “roofer in Nantes for a roof leak”) and note who gets cited. Those are your real competitors in AI answers.