How to Create an llms.txt File

An llms.txt file is a Markdown document, usually placed at the root of your site, that summarizes what your brand does, which pages matter, and which resources an AI should read to understand you better. It doesn't replace your sitemap or your robots.txt; it works as a curated editorial guide for language models.

What is an llms.txt and why would I use one?

llms.txt is a recent proposal for making sites more readable to language models. The core idea is simple: if an AI can't read your entire site because of context limits, hand it a short, clean, useful map. That map is written in Markdown and lives at an easy-to-find URL, typically yoursite.com/llms.txt.

I like to explain it this way: a sitemap tells a search engine "these are my URLs"; an llms.txt tells an AI "this is what you need to understand first." They're not enemies, and they don't do the same job. A sitemap can list dozens or hundreds of pages. An llms.txt should curate what matters: who you are, what you do, which pages define your expertise, and which documents can help answer questions more accurately.

For a site belonging to a designer, consultant, or studio, this makes a lot of sense. An AI doesn't need to read every landing page, every old post, and every portfolio variation to understand your positioning. It needs to know your specialty, your core services, your relevant case studies, your authority topics, and the right paths to deeper content.

The structure proposed by llms.txt is pretty human: an H1 with the project name, a short summary in a blockquote, additional context, and sections with lists of links ¹. That matters, because we're not talking about a file full of obscure code. We're talking about a technical editorial piece: clear for humans, easy to process for machines.

My advice is to not treat it as an AI trick. See it as a brutal exercise in strategy. If you can't summarize your site in a clear llms.txt, your content architecture probably isn't clear either. That's where this practice becomes useful even before a model ever reads it.

How does it differ from robots.txt and sitemap.xml?

robots.txt defines crawl rules, sitemap.xml lists URLs, and llms.txt offers curated context for language models. Each file answers a different question, which is exactly why they should coexist instead of compete.

The confusion comes from assuming everything technical serves the same purpose. It doesn't. A sitemap can be complete without being strategic. A robots.txt can control access but doesn't explain who you are. An llms.txt, on the other hand, forces you to prioritize.

The official llms.txt site states that the file can complement existing standards like robots.txt and sitemap.xml, because it offers a curated summary and links to content that's important for language models ¹. The key word is "curated." If you turn your llms.txt into yet another endless list, you've missed the point.

For personal brands, studios, and B2B companies, my suggestion is to use it as an editorial cover page for AI. It has to explain what you do, from what position of authority you do it, and which pages should be consulted to understand your most important services or content.

What should a good llms.txt include?

A good llms.txt should include the site name, a short description, editorial context, priority links, and optional resources. What matters is that it stays brief, specific, and easy to maintain.

The structure recommended by the original proposal is fairly clear: an H1 with the project name, a blockquote with a summary, additional text without required headings, and H2 sections with lists of links ¹. In plain terms: title, description, context, and paths.

Here's the practical structure I use for design, tech, or consulting sites:

If you're building the file for a blog, don't dump everything in. Choose pillar articles, evergreen guides, important comparisons, and content that explains your point of view. If a post was written purely around a temporary news cycle, it probably doesn't belong there.

How do you create an llms.txt step by step?

To create an llms.txt, first define the site's narrative, select the essential URLs, write the file in Markdown, and upload it to the root of the domain. Then test it by reading it as if you were someone who's never heard of the brand.

Here's my process, without overcomplicating it:

1. Define the main entity. Before you open an editor, write in one sentence who you are or what the site is. If you can't, the problem isn't technical.
2. Choose the pages that actually matter. Don't include the entire blog. Select the home, services, about, case studies, pillar articles, and resources that explain your expertise.
3. Write in clean Markdown. Use an H1, a blockquote, short paragraphs, and lists with links. Don't add unnecessary HTML.
4. Add descriptions to each link. A URL alone isn't enough. Explain why that page matters.
5. Separate the required from the optional. The "Optional" section is for resources that can be skipped when a shorter context is needed ¹.
6. Publish it at /llms.txt. The path should be easy to find and consistent.
7. Keep the file alive. Every time your services, pillar articles, or positioning change, update it.

A common mistake is treating the llms.txt like a legal document: long, cold, and full of corporate phrasing. Don't. If your site has personality, the file should reflect it too, even if it's brief.

What does a realistic example look like for a consultant or designer?

A realistic example should be short, clear, and oriented toward explaining authority, services, and useful content. It shouldn't promise things the site doesn't deliver.

This example is illustrative, not an official file for any client:

```md

Israel Piña

Israel Piña es diseñador UX/UI mexicano, AI Design Expert y Webflow Premium Partner. El sitio reúne servicios, artículos y recursos sobre diseño con IA, Webflow, neuromarketing, branding y conversión digital.

Este sitio está escrito para equipos de marketing, producto y dirección que necesitan crear experiencias digitales más claras, medibles y escalables. La posición editorial del blog es: IA como copiloto creativo, no como reemplazo del oficio.

Servicios principales

• Diseño UX/UI: Servicios de diseño de interfaces, estrategia digital y experiencia de usuario.
• Webflow: Desarrollo de sitios premium en Webflow para marcas y empresas.

Artículos clave

• AEO: qué es y cómo se diferencia del SEO: Guía para entender la optimización para motores de respuesta.
• JSON-LD para diseñadores: Introducción práctica a datos estructurados sin código intimidante.

Optional

• Prompts de diseño UI: Recursos para usar IA como apoyo en exploración visual y prototipado.

```

The point isn't to copy this example as-is. The point is to understand the intent: give an AI a short map, with selected links and a clear narrative.

How would I validate whether my llms.txt works?

I'd validate it with a simple test: an outside person should be able to read it in under two minutes and explain what the site does, who it works for, and which content is a priority. If a person doesn't get it, there's no reason an AI would get it right either.

Then I'd run a second test with language models. I'd copy the contents of the llms.txt and ask: "What does this site do?", "Which topics does it specialize in?", "Which pages should I read to understand its services?", and "What's still unclear?" If the answers come back vague, the file needs more precision.

I'd also check maintenance. An outdated llms.txt can do more harm than good, because it communicates an old version of your brand. If you change categories, services, URLs, or positioning, you have to update it. This shouldn't be an annual project; it should be part of the site's editorial maintenance.

Checklist for publishing your first llms.txt

Before you upload it, review this:

1. The file lives at /llms.txt or at a clearly documented path.
2. It has an H1 with the correct brand or site name.
3. The blockquote explains the value proposition in under two lines.
4. It doesn't list every URL; only the most important ones.
5. Each link has a brief, useful description.
6. It includes pillar articles, not just commercial pages.
7. It contains no false promises or exaggerations.
8. It's written in clean Markdown.
9. It reads well for humans.
10. It has someone responsible for keeping it updated.

This checklist looks basic, but it prevents most mistakes. And yes: AI can help you generate a first version. But the final curation has to come from someone who understands the site's strategy.

Frequently asked questions

Is llms.txt an official standard?

It's an open proposal, not a universal standard adopted by everyone. Even so, it's useful as an editorial practice because it organizes a site's key context for language models.

Do I need an llms.txt if I already have a sitemap.xml?

It can absolutely make sense. The sitemap lists URLs; the llms.txt explains which ones matter most and why. They're files with different jobs.

Can I use llms.txt with Webflow?

Yes, though it depends on how you publish static files in your setup. If you can't upload it directly to the root, you can document an alternative path and link to it.

How long should an llms.txt be?

It should be short enough to serve as a map, not as a full copy of the site. For many personal or B2B sites, somewhere between 400 and 900 words is usually enough.

Can I put all my pages in the llms.txt?

You can, but I don't recommend it. If everything is a priority, nothing is. Use the file to curate, not to replicate your sitemap.

A gentle CTA

If you want to design a useful llms.txt for your site, or turn your blog into a clearer source for answer engines, write to me at hola@israelpinapol.com or visit israelpina.cool. We'll review it with no fluff and real editorial judgment.

---

1. llms.txt, "The /llms.txt file", proposed structure and usage for files readable by language models: https://llmstxt.org/ ↩