llms.txt Practical Guide: How to Create It and What to Realistically Expect

Quick answer: Creating an llms.txt is free and takes about ten minutes, but there is currently no public evidence that it improves how often generative engines cite you. Treat it as a low-cost add-on, and never as the top priority in your GEO strategy.

What is llms.txt in one sentence?

It's essentially robots.txt, but aimed at AI agents rather than search engine crawlers: a simple Markdown file located at tudominio.com/llms.txt that describes what your site is, what sections it has, and which links are most relevant, so that models can quickly locate things when deciding what to consult during the grounding process for an answer.

In other words, it's a "site navigation map" for large language models, not an index file for traditional crawlers. It was designed to help models understand your site's structure in seconds, not to enumerate every URL.

Does it really improve AI engine citations?

There is no verifiable public evidence

No mainstream generative engine has confirmed using it. It's a community-proposed standard, not a verified factor.

Google's Gary Illyes has publicly confirmed that Google does not use it and does not plan to support it. According to the GenScore blog, it is currently a community-driven proposal with no confirmed adoption by any major generative engine.

This doesn't mean it's useless—it means you shouldn't expect measurable changes in AI share of voice just because you published it.

The signal it tries to influence: authority

According to GenScore's GEO Score framework, llms.txt aims to influence the "authority" signal—whether AI cites your domain as a source. Among the five GEO Score metrics (presence 66%, prominence 80%, share of voice 64%, authority 55%, technical readiness 70%—illustrative data, structure based on the product), authority is neither the highest-weighted nor the only one that needs optimizing.

How to create your llms.txt step by step

No special tools required—it's just a plain text file.

Create the file in your domain root

At the same level as robots.txt or sitemap.xml, accessible via https://tudominio.com/llms.txt.

Quick action: Check whether you already have one: open tudominio.com/llms.txt in your browser.

Start with an H1 and a one-sentence summary

Use your brand name as the heading, followed by a sentence explaining what you do—no marketing jargon, plain language. This is the first thing the model reads.

Quick action: Write this sentence as if you were introducing yourself to someone in an elevator.

Add key links by section

Your docs, your most important product pages, your blog. One Markdown-formatted link per line, with a sentence describing what it contains. If the site is large, group with H2 headings.

Quick action: If it takes more than one screen, cut it in half. Less is more.

Keep it short and publish

It's not a complete sitemap or a dump of all content: it's a directional summary. After publishing, only update it when your site structure changes—new sections, rebranding—not on a weekly routine.

A real minimal example

Here's the structure GenScore's own llms.txt roughly follows:

# Your Brand Name

> A clear one-liner describing what you do, no marketing jargon.

## Product

- [Pricing page](https://tudominio.com/precios): what each plan includes.
- [Free trial](https://tudominio.com/registro): how to get started.

## Docs

- [Docs](https://tudominio.com/docs): master index.

That's it. A heading, a description, and a few link sections pointing to what truly matters on your site.

Common mistakes when creating llms.txt

• Don't just copy-paste your sitemap.xml: it's designed for crawlers, contains hundreds of URLs, and isn't meant to help a model understand your site in seconds.
• Don't treat it as a keyword list. It's not content for ranking positioning—it's a navigation map.
• Don't expect measurable results. No mainstream engine has confirmed using it.
• Don't put it ahead of anything that has supporting evidence.

What to prioritize (or do at the same time)

If you have ten minutes, create llms.txt—it's free and harmless. But if you have to choose where to spend your time this week, here are the things with actual evidence behind them:

Lever	Evidence	Effort
Structured data (Article, FAQPage)	Confirmed, reduces ambiguity during extraction	Low
Citable content with numbers and sources	Confirmed, models cite specific data	Medium
Real domain authority	Confirmed, strongest observed correlation	High
llms.txt	No evidence, no engine confirmed using it	Low

You can read more about structured data and AI citations. To understand where you currently stand on these signals, that's exactly the composite metric GEO Score calculates.

FAQ

Does llms.txt replace robots.txt or sitemap?

No. Each serves a different function: robots.txt controls what bots can crawl, sitemap.xml lists all indexable URLs, and llms.txt is a directional summary for models to read quickly. You can—and should—have all three, with no conflict.

Is it an official standard from Google, OpenAI, or Anthropic?

No. It's a proposal driven by the AI tools community, not a standard officially released or supported by any generative engine provider. That doesn't make it invalid, but it explains why its actual adoption rate remains uncertain.

Can a poorly done llms.txt hurt me?

Not directly—search engines won't penalize it based on file content, unlike a misconfigured robots.txt that can accidentally block crawling. The real risk of a poorly done llms.txt is just wasted time, not punishment.

This article was written by the GenScore team, covering AI engine brand visibility based on real scan data from our own tools. When there's no evidence, we say so.

Background

Grounding is a key mechanism in how large language models generate answers: rather than relying entirely on knowledge memorized during training, the model retrieves and cites external sources in real time, making answers verifiable and reducing hallucinations. In web-search or RAG (Retrieval-Augmented Generation) scenarios, the model first fetches web content, then organizes its answer based on that content and attributes sources. The design intent of llms.txt is to play a role precisely at this retrieval stage—when a model or retrieval agent decides whether to crawl a site, a clear, short navigation file may help it more quickly judge where the site's relevant parts are, thereby influencing whether it cites and which page it cites. But note: the specific implementation of grounding is decided by each engine, and whether an external standard gets adopted depends entirely on the engine's engineering choices, not on the website owner's wishes.

GEO (Generative Engine Optimization) is an emerging field relative to traditional SEO. Traditional SEO seeks high rankings in search result lists, while GEO seeks to have brands appear in the body and cited sources of AI-generated answers—users may not click links at all, so "being cited" matters more than "being clicked." Frameworks like GEO Score typically break visibility into multiple dimensions: presence refers to whether your content is crawled and indexed by the engine; prominence refers to the degree to which you're mentioned in answers; share of voice refers to your mention proportion relative to competitors on similar questions; authority refers to whether the engine cites your domain as a trusted source; technical readiness refers to whether your site structure is friendly to AI crawling. Understanding these dimensions helps you invest limited energy in the highest-yield areas, rather than being led around by a single term (like llms.txt).

Structured data (Schema.org markup) is semantic annotation embedded in web pages in the form of JSON-LD or microdata, declaring in a machine-readable way "this is an article," "this is an FAQ list," "this is a product price." For AI engines, its value lies in reducing parsing ambiguity: the model doesn't have to guess from messy HTML which part is the title and which part is the answer—it can read fields directly. For example, FAQPage markup declares questions and answers in pairs, and Article markup indicates author, publication date, and body text—exactly the high-confidence sources engines need most during grounding. Compared with llms.txt's voluntary proposal, structured data is a format long supported by Google, Bing, and others, backed by official documentation, so its return on investment is clearer, and it's a higher-priority technical action in GEO.

One-line Summary

llms.txt is a low-cost AI navigation file, but there's no evidence it improves citations—it should rank after structured data and other priorities.

Paragraph Summary

This article discusses llms.txt, a navigation file aimed at AI agents, and clarifies its positioning and value. It's similar to robots.txt, using a short Markdown description of site structure and key links to help models quickly locate content during grounding. But the article emphasizes: no mainstream generative engine has confirmed using it; Google's Gary Illyes has publicly said it doesn't support it and doesn't plan to, and it's just a community proposal—publishing it brings no measurable citation boost. Creation is simple: create the file in the root, add an H1 and one-sentence intro, list key links by section, keep it short. Common mistakes include copying your sitemap, treating it as a keyword list, and expecting measurable results. The levers with real evidence are structured data, citable content with numbers and sources, and genuine domain authority; llms.txt should be a low-priority add-on, with the worst case being wasted time rather than punishment.