What is llms.txt, and do you have to make one?
SHORT ANSWER
llms.txt is a markdown file at the site root that gives AI a summary of the site's key information. It is not yet an official standard and not every engine reads it, but it costs almost nothing to make and the act of organising the facts is useful in itself.
What llms.txt is
A markdown file placed at the site root (https://example.com/llms.txt). It sets out, in a form AI finds easy to read, what the site does, which pages matter, and what the key information is.
The idea is simple. Rather than leaving AI to crawl the whole site and work it out, you supply the summary yourself — much like handing a briefing note to someone visiting the company for the first time.
Honestly — it is not a standard yet
This part belongs first.
llms.txt began as a community proposal and is not a specification any search engine or AI company has committed to supporting. It has not settled into a de facto standard the way robots.txt has.
Stated precisely as of July 2026: no major engine has officially confirmed that it reads llms.txt. Google goes further and classifies it under “things you do not need to make”, stating that its search ignores it. OpenAI, Anthropic and Perplexity have not denied it — their documentation does not mention it at all, and silence should not be read as support.
One common confusion. It is true that these companies publish an llms.txt on their own documentation sites. But that is them presenting their own documentation in a form LLMs can read easily — it does not mean they read anyone else’s llms.txt.
If you see a claim along the lines of “just make an llms.txt and you will appear in AI”, treat it with scepticism. It is not a problem one file solves.
Why make one anyway
Two reasons.
First, it costs almost nothing. It is one text file. If it gets read, that is a gain; if not, there is no loss. Choices with that shape are usually worth making.
Second, the act of making it helps. Writing an llms.txt forces you to settle “what is the one sentence that describes this company” and “which questions can we answer”. That sentence does not stay in the llms.txt. It has to appear in the same words in the home page’s opening paragraph, the structured data description, the about page and the blog profile — and that consistency is what gives AI the basis for recognising the brand as one entity.
So the gain from the tidying is larger than the gain from the file itself.
Format
The specification is loose, but the commonly used structure is this.
# Company name
> A one-sentence definition. What this site is for, finished here.
- Website: https://example.com
- Email: hello@example.com
- Phone: 02-0000-0000
## Core concepts
Term definitions or a description of the business, in paragraphs.
## Main services
- Service A: one-line description
- Service B: one-line description
## Pages
- [Page name](https://example.com/path): what is on this pageThree things to hold to:
- The first line is
# Company name, followed by a>blockquote holding the one-sentence definition. Those two lines get read most. - Attach a description to every link. A bare list of addresses is no different from sitemap.xml.
- Make it agree with the facts on screen. If the llms.txt says the phone number is A while the website says B, you are worse off than with no file.
What to do after making it
llms.txt is not generated automatically. If the site content changes and this file stays as it was, you are handing AI wrong information.
It needs the habit of being edited whenever a service name changes, a contact detail changes, or a page is added. Navirang always includes updating llms.txt in the task list for any change to site structure.
Read it alongside robots.txt
Some sites build an llms.txt while blocking AI crawlers in robots.txt. That is writing a briefing note and locking the door.
If the goal is being cited in AI answers, the main AI crawlers have to be explicitly allowed.
User-agent: GPTBot
Allow: /
User-agent: ClaudeBot
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: Google-Extended
Allow: /That judgement depends on the business. For a publisher whose content is the product, blocking training crawlers can be reasonable. But it is worth being clear before deciding that “we want to appear in AI” and “we would rather AI did not read our content” do not both hold at once.
Summary
- llms.txt is a site summary given to AI, and it is not yet an official standard
- Treat descriptions that exaggerate its effect with scepticism
- It costs almost nothing, and the brand definition it forces you to settle is the larger gain
- Make it agree with the facts on screen, and update it whenever the content changes
- Check at the same time that robots.txt is not blocking AI crawlers
For reference, you can read this site’s llms.txt as it stands.
Frequently asked questions
Q If we make an llms.txt, will AI definitely read it?
A No. llms.txt began as a community proposal and is not a specification any search engine has committed to supporting. As of July 2026 no major engine has officially confirmed that it reads one, and Google's guidance states that its search ignores such files. Since a single file costs almost nothing and there is no loss if it goes unread, we recommend keeping it as an optional item rather than a requirement.
Q How is it different from robots.txt?
A robots.txt instructs what may and may not be read; llms.txt summarises what matters. The purposes are entirely different, so both are needed. robots.txt is a de facto standard; llms.txt is still a proposal.
Q We already have sitemap.xml — is it still needed?
A They do different jobs. sitemap.xml is a list of page addresses; llms.txt explains in human-readable sentences what the site is for. The point is to let AI grasp the essentials without crawling the whole site.
If you need this done rather than read
This article belongs to Structured data and technical. The pages that handle the same subject as work are below.
Related reading
- Which websites does ChatGPT find? AEO read from OpenAI's official documentation Taking OpenAI's ChatGPT Search documentation as the primary source: how ChatGPT searches the web (query rewriting, follow-up searches, source citation), the official conditions for a site to be found, and what those facts mean in practice.
- Can AEO or GEO top placement be guaranteed? What to check before choosing an agency Can top placement in ChatGPT, Gemini or Perplexity answers be guaranteed? Starting from ZDNet Korea's August 2026 report on overselling, here is why it cannot, what to measure instead, and a checklist for choosing an agency.
- How do you measure AEO and GEO performance? Mention rate, citation rate, recommendation share How to measure how far a brand is found in AI search across ChatGPT, Gemini and Perplexity — the definitions and calculations for mention rate, citation rate, recommendation appearance and share of voice, plus question set design and the repeated-measurement principle.