If you have been following the world of GEO, you have surely heard this: «you need to create your llms.txt for AIs to find you». Agencies say it, LinkedIn gurus say it, half of the articles that rank on Google say it. The problem: almost no one shows the data. And the data tells another story. In this article, we explain what llms.txt is, how to create yours in 10 minutes, and what almost no one tells you: what Google says and how many bots are actually reading these files.
What is llms.txt?
llms.txt es una propuesta de estándar creada en 2024 por Jeremy Howard, cofundador de Answer.AI. La idea es simple: un archivo de texto plano, escrito en Markdown, que vive en la raíz de tu dominio. Funciona como un índice de tu contenido más valioso, pensado para que lo lean los modelos de IA. Un detalle de escritura: se llama llms.txt, con «s» (por «LLMs», modelos de lenguaje en plural). Mucha gente lo escribe «llm.txt», pero esa versión no existe en el estándar. Pensalo así:
- robots.txt tells bots what they cannot look at.
- sitemap.xml lists all your URLs, without hierarchy.
- llms.txt tries something different: to tell the AI «of all my site, this is important.»
Why Markdown and not XML? Because AI models read Markdown more easily: fewer tags, less noise, fewer tokens.
How does llms.txt work (in theory)?
The flow proposed by the standard has three steps:
- Your site publishes the file at the root of the domain: tudominio.com/llms.txt.
- The file lists your key content: a title with your brand, a short summary, and links with descriptions (guides, prices, policies).
- The AI model queries the file and uses that index to go directly to your correct pages, without navigating blindly.
Steps 1 and 2 depend on you. Step 3 depends on whether the models actually query the file. And that’s where theory and reality diverge.
Who uses llms.txt today? What the data says
Here comes the part that most articles in Spanish omit. In May 2026, Ahrefs published a study on 137,000 domains to measure the actual adoption of llms.txt. The results:
- The 28% of domains already publish a valid llms.txt file.
- The 97% of those files received not a single visit (neither from bots nor from humans) in the entire analyzed month.
Read it again: one in four sites did the homework, and almost no AI came to check it out. It’s like the whole block put up a sign at the door for a mailman that doesn’t exist yet.
What does Google say about llms.txt?
Google was very direct on this topic, on two fronts. First, John Mueller (Google Search Advocate) compared llms.txt to the keywords meta tag from the 2000s. He added: «no AI service said it is using llms.txt, and if you look at your server logs, you can verify that they don’t even consult it».
Second, the official Google guide on AI optimization (2026) put it in writing:
«You don’t need to create new machine-readable files, AI text files, tags, or Markdown to appear in Google Search. […] Google Search does not use them.»
And what works for Google’s AI features (AI Overviews, AI Mode)? The same as always: unique and valuable content, indexable pages, and structured data where appropriate.
So, should I create my llms.txt or not?
Our stance, with the data on the table: create it, but as a 10-minute task, not as a strategy. 
In favor:
- Cuesta casi nada y el riesgo es cero. Google confirmó que no perjudica.
- Algunos crawlers y herramientas menores sí lo leen hoy. Si el estándar despega mañana, ya estás listo.
- The exercise itself is valuable: it forces you to choose your 10 most citable pages and describe them in one line. That question is real work GEO with or without a file.
Against:
- It will not improve your position in Google or in AI Overviews. Word of Google.
- There is no evidence that ChatGPT, Gemini, Claude, or Perplexity consult it.
- Every hour spent on «optimizing your llms.txt» is an hour not spent on what truly drives your visibility in AI.
How to create your llms.txt in 10 minutes
If you decided to create it (good job, it’s the right decision as long as you don’t spend more than 10 minutes), the format is this: 
Step by step:
- Open a plain text editor (not Word: a real .txt or your code editor).
- First line: an H1 with your brand.
# YourBrand - Below, your summary in one or two lines: what you do and for whom. This is the description you want an AI to repeat. Write it with that intention.
- Then, sections with H2 (
## Guides,## Product) and a list of links with a short description:- [Título](URL): what you find there. Choose your 5 to 15 most important URLs, not the 500 from the sitemap. - Save it as llms.txt and upload it to the root of the domain. In WordPress, you can do it via FTP or with a plugin; several SEO plugins already generate it automatically.
- Check that it loads by visiting yourdomain.com/llms.txt in the browser. Plain text, no design: that’s correct.
A technical detail we learned the hard way: if your hosting has aggressive anti-bot protection (like SiteGround’s captchas, for example), make sure the file is accessible to crawlers and not just to humans. A llms.txt behind a captcha is a sign inside a drawer.
llms.txt vs robots.txt vs sitemap.xml
| Aspect | robots.txt | sitemap.xml | llms.txt |
|---|---|---|---|
| What it does | Controls bot access | Lists all your URLs | Selects your key content for AIs |
| Format | Plain text with directives | XML | Markdown |
| Who respects it | All serious search engines and AI bots | All search engines | Without official adoption by major AIs |
| Mandatory? | Yes, in practice | Highly recommended | Optional |
| Proven impact | High | High | None demonstrated (2026) |
The conclusion of the table: robots.txt and sitemap.xml are infrastructure, llms.txt is a gamble. Have all three, but in that order of priority.
What to do instead of obsessing over llms.txt
If the goal is for ChatGPT, Gemini, or Perplexity to mention your brand, this is what works:
- Citable content with original data. AIs cite sources that provide numbers, comparisons, and real experience, not summaries of what already exists. We have a 7-step model to optimize content for AI search engines that brings this down to earth.
- Brand authority outside your site. Models build their recommendations by reading listings, reviews, media, and communities. If you are the only one talking about your brand, there is no signal. This is the E-E-A-T applied to GEO, and the key to appearing in ChatGPT.
- Fast, indexable pages with a clear structure. To appear in Google’s AI features, you first need to be indexed. Without that, there is no magic file that helps.
- Measurement. Everything above is blind if you don’t know if AIs are already mentioning you. Optimizing without measuring is guessing. If you’ve never measured it, start with our AI visibility checker.
Del último punto nos ocupamos nosotros: en CreceRank configurás los prompts relevantes de tu industria y la plataforma los consulta cada día en ChatGPT, Gemini, Perplexity y Google AI Overviews. Así ves si tu marca aparece, en qué posición y quién te está ganando terreno. Datos, no sensaciones. → Empezar 14 días gratis, sin tarjeta de crédito
Conclusion
So, what is llms.txt? A Markdown file at the root of your site that proposes an index of your content to AIs. The key word is «proposes»: in 2026 it remains a standard without confirmed adoption, with 97% of files that no bot visits and Google explicitly stating that it does not use it.
Our recommendation:
- Create it today, in 10 minutes, with your 5 to 15 most important URLs. Zero risk, and you’re covered if the standard takes off.
- Don’t spend another minute, nor buy «llms.txt optimization» services.
- Invest that time in what really moves the needle: citable content, authority, and measurement. If you’re building your stack, here we compare the 15 best GEO tools in 2026.
If you want to know if AIs are already mentioning your brand (with or without llms.txt), at CreceRank you can measure it in minutes.
FAQs
Sources
- What Is llms.txt? (Ahrefs): the study on 137,000 domains and the statements from John Mueller.
- Google’s Guide on Generative AI in Search (Google Search Central): Google’s official position on files for AI.
- llmstxt.org: the original proposal of the standard, by Jeremy Howard (Answer.AI).
