The map cannot replace the house.

llms.txt is an open proposal for providing a human- and machine-readable Markdown document. It can be placed at the site root:

https://example.com/llms.txt

V2 also allows /docs/llms.txt: each file describes URLs under its path, with the most specific file preferred when scopes overlap. It recommends Markdown page versions and rel="describedby" links to the index, plus rel="alternate" type="text/markdown" links to those versions, through HTML or HTTP Link headers. These are proposal recommendations, not guarantees of support across agents. Location and discovery in v2.

The file can summarize the website, provide interpretive context, and link to its most useful pages. Version 2 of the proposal recommends keeping the main document relatively small and placing detailed information behind links that an agent can retrieve when needed. The llms.txt proposal.

A simple example for a fictional Miami law firm might look like this:

# Rivera Legal

> A bilingual immigration law firm in Miami, Florida.

## Services

- [Immigration consultations](https://example.com/immigration):
  scope, limitations, and consultation process.
- [Team](https://example.com/team):
  attorneys, licenses, and verifiable experience.
- [Frequently asked questions](https://example.com/faq):
  general information, not legal advice.

There are no magic keywords here—only context, hierarchy, and links.

robots.txt belongs to a widely recognized protocol for communicating crawler access. sitemap.xml gives search engines a list of URLs. Structured data uses established vocabularies to describe entities and content.

llms.txt currently occupies a different category: it is a community proposal gaining adoption, not a universal requirement imposed by major search and answer systems.

Its own documentation describes it as an open proposal designed to coexist with robots.txt and sitemaps, not replace them. Relationship to existing standards.

The distinction matters:

• robots.txt communicates which crawling you permit.
• A sitemap lists pages you want search engines to discover.
• HTML contains what visitors receive.
• Structured data describes elements present on a page.
• llms.txt can provide a curated path for agents that know to look for it.

Five separate pieces. None should claim to perform the others’ work.

The official documentation we reviewed does not present llms.txt as a requirement for generative search experiences.

Google says its AI features use the same foundations as Search: a page must be crawlable, indexed, and eligible to appear with a snippet. It also states that AI Overviews and AI Mode have no additional technical requirements. Google: AI features and your website.

OpenAI documents OAI-SearchBot as the crawler publishers can manage for possible inclusion in ChatGPT search experiences. Its publisher guidance focuses on allowing that crawler and avoiding blocks in robots.txt, the server, or the CDN. OpenAI crawlers.

Those instructions do not identify llms.txt as an eligibility requirement.

This does not prove that no agent uses the file. It supports a narrower conclusion: we should not sell it as a mandatory passport or a confirmed ranking signal.

Optional does not mean useless.

A well-maintained file can help an agent, coding assistant, or documentation tool that knows it exists and chooses to retrieve it. Instead of inferring the organization’s structure from navigation, the system receives an explicit selection of resources.

It may be particularly useful for:

• extensive technical documentation;
• catalogs with many sections;
• projects whose terminology needs context;
• author websites with work spread across many pages;
• bilingual or multilingual organizations;
• agents that need to locate policies, services, or references quickly.

Documentation platforms and content-management tools have already adopted the proposal. That demonstrates adoption, not universal effectiveness. Proposal integrations.

Our interpretation is deliberately modest: llms.txt may reduce friction for a compatible reader. It cannot require that reader to enter.

The file adds another surface that must be maintained.

If the website offers five services while llms.txt lists four, the map is outdated. When a price, address, credential, or commercial term changes, the file must also change if it repeats that information.

The risk grows when the file becomes a large copy of the entire website. The current proposal favors a concise primary document with context and descriptive links. Detailed information can remain on individual pages or separate Markdown versions. Scope of the proposal.

For a small business, we would impose three limits:

1. Do not duplicate every paragraph on the website.
2. Do not include claims unsupported by the linked pages.
3. Do not maintain the file unless someone owns its updates.

An outdated map can be worse than no map. It confidently points toward the wrong address.

When time or budget is limited, llms.txt should not come first.

We would first verify:

• that important pages respond correctly;
• that primary content exists in accessible HTML;
• that titles and headings explain each page;
• that links allow the site to be traversed;
• that robots.txt does not block important access accidentally;
• that the sitemap contains valid URLs;
• that canonicals and language versions are coherent;
• that structured data represents visible content;
• that business information is accurate;
• that the site works with keyboards, assistive technology, and mobile devices.

Google applies these same foundations to its generative Search experiences. Meeting them does not guarantee inclusion, but ignoring them to create an experimental file reverses the priorities. Google’s requirements and best practices.

Build a house that can be visited first. Then, when useful, provide a map for machines.

Before creating llms.txt, we ask:

1. Does the site contain enough information to benefit from a curated route?
2. Can we summarize it without inventing or expanding promises?
3. Do the links lead to public, clear, current pages?
4. Will someone review the file whenever the site changes?
5. Do we understand that third-party use is not guaranteed?

When the answers are yes, the cost may be small. A carefully written text file weighs very little and adds no JavaScript, fonts, or requests to the normal page load.

When the foundations are missing, the file can wait.

llms.txt is a technical courtesy. It is not a complete GEO, AEO, or SEO strategy.

We can review how people, search engines, and agents find your information and decide whether llms.txt clarifies that path or merely duplicates content that will need further maintenance.

Our technical SEO service begins with the foundations. In The web of machines, we explain why automated readers are now part of that review.

Let’s talk →