What is llms.txt?
The llms.txt proposal describes a Markdown overview containing background information, guidance and links to detailed material. It is intended to help an agent find information while assisting a user. Think of it as a curated entry point into content, rather than a command that every AI system must obey.
A useful file should help someone choose the right document. It should not tell a reader that an unsupported product claim is authoritative merely because the publisher included it. The linked pages still need to be accurate, current and accessible to their intended audience.
Start with a documentation task
Consider whether an agent or reader has difficulty finding the correct setup guide, policy or product reference. A maintained overview may be useful for that task. Define what information it should lead to, then test whether it does. That is a more concrete purpose than expecting a new file to produce citations.
Google's AI-feature guidance says no new machine-readable files or AI text files are needed for AI Overviews or AI Mode. It also specifies no special schema.org requirement. Do not sell llms.txt as a prerequisite for those Search features or infer a ranking improvement from its existence.
Where the file belongs
The proposal permits an llms.txt file at the website root or at a subpath. It says the file covers URLs beneath its path and that an agent should use the most specific applicable file. A documentation section can therefore have its own overview without requiring its owner to control the entire host.
The proposal also recommends clean Markdown versions of useful pages and describes discovery links: rel="alternate" with type="text/markdown" for a page's Markdown version, and rel="describedby" for the covering llms.txt file. These are recommendations in the proposal, not a claim that every client will discover or follow them.
Before adding discovery links, check that the destinations exist and contain the expected content. Do not manufacture Markdown URLs by appending an extension to every page and assuming the server provides those files. A route that returns the normal HTML page is not evidence of a clean Markdown export.
Required and optional sections
In the inspected proposal, the H1 containing the project or site name is the only required section. A summary blockquote, other explanatory text and H2-delimited file lists are optional. File-list items contain Markdown links and may include descriptive notes. “Optional” is a convention for secondary material that an agent can skip when it needs a shorter context.
| Element | Proposal status | Editorial purpose |
|---|---|---|
| H1 name | Required | Identify the site or project. |
| Summary blockquote | Optional | Explain the scope of the material. |
| Additional text | Optional | Explain relevant context or how to interpret the documents. |
| H2 file lists | Optional | Group useful links with descriptive notes. |
| Optional section | Convention | Separate secondary links from the main reading path. |
The table describes the proposal's format, not a checklist for search eligibility. Required syntax alone does not make the file useful. Conversely, do not call a file invalid solely because it omits a recommended summary or contains a different number of links than a marketing checklist suggests.
A small illustrative file
Format illustration, not a deployed theStacc file: this example points to the two sources used in this definition. It demonstrates the Markdown structure using real reference URLs; it is not a claim that those HTML pages are Markdown exports.
# llms.txt reference notes
> Sources for understanding the proposal and Google's AI-feature requirements.
## References
- [llms.txt proposal](https://llmstxt.org/): Purpose, format and path scope.
- [Google AI features](https://developers.google.com/search/docs/appearance/ai-features): Google's site-owner guidance.For your own file, choose links from the content you actually maintain. Use descriptive link text so someone can distinguish a setup guide from a pricing policy. When a verified clean Markdown version is available, assess it for the agent-facing reading path rather than guessing its URL.
llms.txt, robots.txt and sitemaps serve different purposes
The proposal distinguishes llms.txt from robots.txt and sitemaps. It describes llms.txt as a curated overview used on demand, robots.txt as communicating acceptable automated access, and a sitemap as a page inventory for search engines. An overview does not replace either the inventory or the access instructions.
Do not put private pricing, customer records or credentials in a public file and rely on a “do not summarize” note to protect them. Apply appropriate access controls to private material. Stating an attribution preference in Markdown also does not demonstrate that a particular client enforces it.
The same caution applies to a full-text export. Check the intended audience and publication rights before making content public through a different endpoint. An export should not accidentally bypass the access restrictions of the original page.
Validate content, not just the HTTP status
- Fetch the intended file and inspect the response body. A successful status may still be an HTML fallback or error page.
- Check that the project name and scope match the material the file covers.
- Open the listed destinations and verify that their descriptions are accurate.
- Check any declared Markdown alternatives instead of assuming an extension creates an export.
- Test a real information-finding task with the client you intend to support, recording its configuration and what you supplied to it.
- Review the file for private information and misleading access or attribution promises.
Supplying the file directly to a test client can show that it helps with a task under those conditions. It does not show that the client would find it automatically, that all agents use it, or that a search service will cite your pages. Report those as separate questions.
Maintenance mistakes to avoid
Update when key content changes. Remove or correct stale links when a document moves, and review descriptions when a feature or policy changes. Assign an owner so the overview does not become a second, contradictory version of the website.
Avoid unsupported promises about crawler schedules, knowledge-graph weighting, training cycles or citation deadlines. The proposal does not establish those behaviors for every service. Choose a file size and level of detail that help the actual reading task, not a universal quota of “high-authority” links.
Keep the underlying content accurate
theStacc's content workflow documents research, drafting and publishing to a connected CMS, with manual, approval-first and automatic modes. Those functions can support maintaining the pages an overview points to. This definition does not claim that theStacc generates llms.txt files or that publishing one improves citations.
Review the AI Blog Writer and publishing controls for your underlying content process. Book a demo if you need to inspect the workflow and its current scope before choosing it.
Frequently asked questions
Is llms.txt a platform requirement?
The inspected source presents it as a proposal. Google explicitly says no new AI text files are needed for AI Overviews or AI Mode. Evaluate other clients against their own documentation or tests; do not infer universal support from a filename.
Must llms.txt be at the domain root?
The inspected proposal supports a root file or a file at a subpath, covering URLs beneath that path. It says the most specific applicable file should be used. Client behavior still needs verification for the implementation you intend to support.
Does llms.txt stop crawling or enforce attribution?
It provides information and links, not access enforcement. The proposal distinguishes it from robots.txt. Do not use a Markdown instruction to protect private content or assume that a client will enforce attribution preferences without evidence of that behavior.
When should citation or traffic results appear?
The sources used here do not establish a citation or traffic timetable. You can test whether a file is accessible and useful for an information task. Search citations, attributable referrals and business outcomes require their own observations.
Related terms
Continue with AEO, GEO, AI citations and AI Overviews for the broader content and measurement questions.
Sources and scope
The llms.txt proposal supports the format and purpose described here. Google Search Central's AI-feature guidance supports the Google-specific requirements. Both were reviewed September 14, 2026. Product statements use theStacc's first-party feature catalogue; the checklist and code example are editorial guidance, not test results.
Explore this topic: LLMs.txt Guide: What It Is and How to Set It Up (2026) · How to Create an LLMs.txt File: A 6-Step Guide for 2026.