How to appear in ChatGPT: what works and what does not

“¿Por qué ChatGPT no recomienda mi empresa?” is the question this page answers, without hacks. The short version: ChatGPT assembles its answers from sources it trusts, and most of those sources are not your website. There is no file to add and no markup that flips a switch. What works is being present in the sources the models actually cite. Which sources those are, in your category, is a measurement question, and we answer it with a study.

Request the study

A model answers in a paragraph, not a list of links. It retrieves from sources it has learned to trust, reads them itself, and writes the answer. Your position in a search results page does not decide the outcome; your presence in the retrieved sources does.

Two facts about retrieval, straight from the documentation:

The sources are mostly third parties. One widely quoted analysis [VERIFICAR: identify and link the original study before asserting] found that earned media contributes 48% of what assistants cite, third-party commercial content 30%, and the brand’s own site 23%. We keep this figure marked until we can name its source, because a number without a source is exactly what we criticize. Even without it, the practical conclusion holds: your site is one source among many, and the decision happens mostly outside it.
Crawlers decide whether you can be read at all. OpenAI’s publisher documentation distinguishes GPTBot, which governs training, from OAI-SearchBot, which governs ChatGPT search (https://help.openai.com/en/articles/12627856). If you want to be cited, do not block OAI-SearchBot. Note the detail almost everyone misses: pages disallowed via robots.txt can still surface as a link and title through third-party providers; noindex is the actual opt-out. So robots.txt is not a visibility switch in either direction, and noindex is the off button, not robots.txt.

If you sell through retailers, pharmacies or distributors, there is a bigger trap: measuring the wrong instrument. Grading your website tells you about your website. Your buyers walk a path that may never touch it: they ask an assistant, open a retailer app, read two reviews, and pick up the product. Visits to your brand’s site on that path: zero. The category conversation is the frame that matters, not your domain.

What does not work

Google’s own guidance closes this door, and we quote it rather than argue with it:

llms.txt. Google names it as a myth: no AI-specific text file changes how your site is treated. Source: https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
Special schema for AI. There is no extra markup for AI features. The requirements are the same as they have always been: make your content indexable and technically sound. Source: https://developers.google.com/search/docs/appearance/ai-features
Chunking tricks and structured-data overfocus. Named as myths in the same guide. Source: https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
Guaranteed appearances. Nobody can guarantee what an assistant will recommend. Answers vary by user, by context and from run to run. Anyone selling a guaranteed position in ChatGPT is selling something they do not control.

The same guide states the frame we work with: optimizing for generative AI search is still SEO. If an agency sells you a secret GEO framework, they are selling what Google says does not exist.

What actually moves it

What is left is unglamorous and effective:

Be present in the sources the models cite. For consumer brands this is usually retailer product pages, review platforms, consumer media and regulators. For B2B it is comparison content, industry media, documentation and the communities where buying committees ask. The list differs by category, which is exactly why it has to be measured, not guessed.
Keep the content you want cited recoverable. Open to OAI-SearchBot, not noindexed, technically sound. Boring, decisive, and the same work good SEO already requires.
Keep your entity consistent. The model has to know who you are to credit you with anything: one name, subbrands attributed correctly, facts that agree across your site, your retailers and your press. When a model repeats an outdated study or a wrong claim about you, that error is living somewhere in the sources it trusts.
Earn the third-party mentions. This is editorial work, PR and retail content, and it is slow. It is also the lever that survives every model update.

None of this is llms.txt. All of it is the open web.

Why measure before you optimize

Ask an assistant your category question five times. You will not get the same list of brands every time, and your brand may appear in one of the five answers. If you had screenshotted the first one, you would have concluded your brand is invisible. If you had screenshotted the third, you would have concluded everything is fine. Both conclusions would be wrong, because one screenshot is an anecdote, and anecdotes fail in both directions: false comfort and false panic.

This is also why generic advice stays generic. “Be present in the sources the models cite” is only actionable once you know which sources those are in your category, how often they appear, and who controls them. That is not an opinion. It is a count.

From generic advice to your category

The study makes this page specific to you:

We build a prompt bank from your category and documented demand: the questions your buyers actually ask, unbranded, brand-named and competitive.
We run it across AI models, with repetitions, and read every response.
You get the ordered source list with owners: the retailer page you can rewrite this quarter, the earned media to pursue, the regulator page nobody displaces and you work around.
Every action item above stops being a category and becomes a name.

The method, the deliverables and the sample reports are on the study page: /auditoria-visibilidad-ia/.

Start with the sources, not the hacks

There is no file that turns ChatGPT into a fan of your company. There is a source list, and it is measurable. Tell us your category and we will scope the study that names it.