There is no procedure that "registers" a company into ChatGPT, Claude, Gemini or Perplexity. Their answers are built from two materials: the models' training data, and the web pages consulted at question time. Appearing in AI answers therefore means becoming an obvious presence in those sources. That can be worked on with observable levers; it cannot be bought, and it cannot be guaranteed.
Where the names cited by AIs come from
When an engine answers "which provider would you recommend for...", it does not consult an official register. Depending on the case, it draws on what its training data retained from the web (slow to evolve, impossible to edit directly) and on a live web search (fast, observable: that is where the short-term game is played). The engine reads a few pages, extracts names and facts, then writes.
The practical consequence: your visibility in AI answers is a function of your visibility in the pages they consult. Those pages can be observed: that is the starting point of any serious work.
The observable levers
- Be present where engines draw from. On your questions, engines cite their sources: comparison articles, specialised directories, press, review platforms, reference pages. Every source you are missing from is a concrete project; tracking cited sources gives you that list, question by question.
- Publish pages that answer real questions. Engines readily pick up content that answers, defines, compares or quantifies. One page that plainly answers a question your customers ask is worth more than ten presentation pages.
- Be consistent everywhere. The same name, the same description, the same location on your website, your directory listings and your public profiles: what is consistent extracts and restates well; what contradicts itself cites poorly.
- Provide citable material. Original data, sourced figures, reference content about your trade: that is the kind of material engines (and the people writing the pages they consult) pick up together with its author's name.
- Cover your customers' languages. A question asked in French pulls French-speaking sources. If your customers query AIs in several languages, your written presence must exist in those languages.
What does not work
- Guarantee promises. Nobody controls AI answers: not you, not any provider. An offer that "guarantees" your citation by ChatGPT promises something it does not control.
- Stuffing. Repeating your name or piling up keywords does not create picked-up material: engines synthesise sources that corroborate each other, they do not count occurrences on a single page.
- Testing once and concluding. The same question returns different answers from one run to the next. An isolated appearance (or absence) says nothing; only a rate measured over repetitions is meaningful.
- Expecting an immediate effect. Training corpora evolve slowly, consulted sources a little faster. The work pays off, but it shows over weeks and months, not overnight.
Measure before, measure after
Before optimising anything, establish a baseline: the questions your customers would ask, asked continuously to the engines that matter to you, with repetitions. You will know where you appear, at what rate, and above all which sources feed the answers: that is your work list, in priority order.
Then re-measure over time, with the same protocol: it is the only way to separate the effect of your work from an accident of the draw. The full method (repetitions, frequency, extraction of names and sources) is described on our how-it-works page; the ranking tracker applies it to your questions, and every figure stays verifiable in the archived full answers.
Read next
If your competitors already appear and you do not, "Why ChatGPT recommends your competitors" details the mechanisms at play. For a first manual diagnosis, see "How to know if ChatGPT recommends your business"; for the vocabulary (GEO, AEO, generative engine optimization), see "What is GEO?".