Sabilytics

GuidesAI Discoverability

AI Discoverability

What is llms.txt? How to Write One That Helps AI Assistants (2026)

llms.txt is a plain-text file at your site root that briefs AI assistants on what you offer and which pages matter. Learn the format, what to include, and how Sabilytics checks for it.

5 min readPublished Jul 25, 2026

Written by

Ayodele Samuel Adebayo

Founder, Sabilytics

llms.txt is a plain-text file you publish at https://yourdomain.com/llms.txt. It is a short briefing for AI assistants: what your product is, which pages matter most, and how to learn more.

Think of it as a curated README for your website. Assistants that can fetch your site may use it instead of guessing from scattered homepage copy, blog posts, and marketing slogans.

This guide covers what llms.txt is, why it showed up in AI Discoverability conversations, how to write a useful one, and how Sabilytics checks for it.

If you are new to the broader idea, start with What is AI Discoverability?.

Why llms.txt exists

AI assistants often answer questions like:

  • "What does this product do?"
  • "Where are the docs?"
  • "Is this analytics tool privacy-friendly?"

Your homepage is built for humans scrolling a page. An assistant may only pull fragments: a hero line, a nav label, a stale meta description. Those fragments can be incomplete or wrong.

llms.txt gives you a place to say, in one fetchable file:

  • who you are
  • what you sell or offer
  • which URLs are the source of truth
  • where to contact you

It does not replace SEO, a sitemap, or clear page titles. It adds a dedicated context layer for agents that look for it.

What llms.txt is (and is not)

It is

  • A public file at /llms.txt on your domain
  • Usually Markdown-flavored plain text (headings, lists, links)
  • Controlled by you, and easy to update when the product changes
  • Complementary to crawl access, sitemaps, and identity tags

It is not

  • A guarantee that ChatGPT, Claude, or Gemini will cite you
  • A replacement for robots.txt or sitemap.xml
  • A secret API key or private briefing (it is public, like any other page)
  • An official rule enforced by every AI vendor the same way

Treat it like Open Graph tags for AI context: useful when present and honest, ignored or underused when missing or vague.

A simple structure that works

There is no single required schema, but a clear pattern travels well:

  1. Title (# Product name)
  2. One-line summary (a short blockquote or paragraph)
  3. What you do (2–4 sentences)
  4. Product highlights (bullets)
  5. Key pages (named links to home, docs, pricing, legal, signup)
  6. Contact (email or site URL)

Keep it short enough to skim. Prefer facts over slogans. Link to real URLs you maintain.

Example skeleton

# Your Product

> One sentence: who it is for and what problem it solves.

A short paragraph with a bit more detail. Stay concrete.

## Product

- Feature one: what it helps someone do
- Feature two
- Feature three

## Key pages

- [Home](https://example.com/): Overview
- [Docs](https://example.com/docs): How to install
- [Pricing](https://example.com/pricing): Plans
- [Contact](mailto:hello@example.com): Support

## Contact

- Email: hello@example.com
- Site: https://example.com

Use your real copy. If a detail is not ready (pricing, docs), omit it rather than inventing it.

How to publish llms.txt

  1. Create the file in your project as public/llms.txt (Next.js, Vite, and many static hosts), or the equivalent static root for your stack.
  2. Deploy so https://yourdomain.com/llms.txt returns 200 with text/plain (or plain text Markdown).
  3. Open it in a private window and confirm the content is what you expect.
  4. Re-scan with an AI Discoverability check so you know the file is reachable from the public web, not only on localhost.

Tips:

  • Prefer the apex or primary hostname you want assistants to treat as canonical
  • Update the file when you rename the product, change positioning, or ship major pages
  • Do not paste secrets, internal URLs, or unfinished roadmap claims

What to put in (and what to skip)

Put in

  • Accurate product name and category
  • Audience in plain language ("indie hackers", "agencies", "SaaS founders")
  • Differentiator you can defend (privacy, speed, live view, and so on)
  • Links that always work
  • Contact that a human can actually reach

Skip

  • Keyword stuffing
  • Superlatives you would not put on your homepage
  • Every blog post you have ever written (link to a docs index instead)
  • Competitor attacks
  • Temporary promo codes or campaign noise

A good llms.txt sounds like a careful founder explaining the product to a smart intern in two minutes.

How Sabilytics checks for llms.txt

Sabilytics AI Discoverability requests https://your-domain/llms.txt during a scan.

  • If the file is reachable, the llms.txt finding passes. The report notes that assistants can get a curated overview of your product.
  • If it is missing or unreachable, you get a warning with impact copy and a fix hint: add /llms.txt with a short product summary and links to key pages.

That check is one weighted signal in the overall score, alongside crawl rules, sitemap, homepage identity, Open Graph, organization schema, canonical URLs, and H1. Passing llms.txt alone does not make a weak site "AI ready." Missing it is still an easy, high-leverage gap for many sites.

Common myths

"llms.txt replaces SEO."
No. Crawl rules, sitemaps, titles, and clear pages still matter for search and for assistants.

"If I publish llms.txt, ChatGPT will recommend me."
Publishing improves the briefing. Recommendation still depends on questions people ask, competition, and whether an assistant chooses to cite you.

"I should block AI bots in robots.txt but keep llms.txt."
You can, but if crawlers cannot fetch your site, a public briefing file may never be read in the workflows you care about. Be intentional about crawl policy and context together.

"Longer is better."
Usually not. A focused page beats a dump of every marketing paragraph you have.

"Only big companies need this."
Small, clear products often benefit more. Ambiguous landing pages are harder to summarize than a tight indie tool with an honest llms.txt.

Frequently asked questions

Is llms.txt an official standard?

It is an emerging convention for publishing AI-oriented site context at a well-known URL. Vendor support varies and will keep changing. The file is still useful as a human-editable summary you control.

Should it be Markdown or plain text?

Markdown-style headings and links are common and readable. What matters most is that the URL returns readable text, not a login wall or HTML app shell.

Do I need llms-full.txt too?

Some proposals mention larger companion files. Start with a solid /llms.txt. Add more only if you have a clear audience for a longer dump (for example extensive docs indexes).

Can I generate it with AI?

Yes, as a draft. Then edit it yourself. Assistants should not invent features, pricing, or company facts. Sabilytics fix prompts ask coding agents to use real repo content for the same reason.

Where should I start if my score already fails elsewhere?

Fix crawl access and sitemap first if those fail. Then homepage identity. Then llms.txt. A briefing file helps most when assistants can actually reach your site.

Write yours, then verify it

A useful llms.txt is short, honest, and easy to fetch. Publish it at the site root, keep the links fresh, and re-check after deploys.

Sabilytics scans for /llms.txt as part of AI Discoverability, with plain-language impact and fix hints you can hand to a coding assistant.

When you are ready, run a free AI Discoverability check and confirm the file shows up on your domain.

← All guides