An llms.txt file is a short Markdown file at /llms.txt that tells an AI agent what your site is and where to find the detail — a table of contents for language models, the way robots.txt is a table of rules for crawlers. For a store, that means: what you sell, grouped into categories, plus links to the pages an agent needs to answer a shopper’s question or complete a purchase. The proposal just got its first major revision in two years, which makes now a reasonable time to check whether yours (or the lack of one) is still current.

What changed in v2

The llms.txt proposal, authored by Jeremy Howard, shipped a v2 revision dated 2026-08-10 — the first substantive update since the original September 2024 proposal. The headline change is discoverability: v2 adds two standard link relations so an agent landing on any page can find its Markdown twin and the llms.txt that covers it, without guessing:

  • rel="alternate" type="text/markdown" — points from an HTML page to its Markdown version.
  • rel="describedby" — points from a page to the llms.txt file covering it.

Both can be served as HTML <link> elements or as an HTTP Link: response header, and the header form works on non-HTML resources and can be added at the CDN or server-config layer without touching page templates. v2 also formally allows both page.html.md and page.md as valid Markdown-URL forms (v1 only specified the former), and confirms that a file at a subpath — /collections/hiking/llms.txt, say — covers everything under that path, with the most specific file taking precedence when more than one applies. See the full changelog for the complete diff from v1.

The format, in the order it’s required

A spec-compliant file is Markdown, in this exact order:

  1. An optional byte-order mark.
  2. An H1 with your store’s name — the only required section.
  3. A blockquote with a one-line summary — what you sell, where you ship, anything an agent needs before reading further.
  4. Any number of plain sections (no headings) with more context.
  5. H2-delimited sections, each a Markdown list of [link](url) entries — your “file lists.”

An example, for a store

# Acme Outdoor Gear

> Acme Outdoor Gear sells hiking, camping and climbing equipment, ships to
> the US and Canada, and offers a 60-day return window on unused items.

## Categories

- [Hiking boots](https://acmeoutdoor.example/hiking-boots): Trail, backpacking
  and mountaineering boots, sizes 5-14.
- [Tents](https://acmeoutdoor.example/tents): 1-6 person tents, 3-season and
  4-season.
- [Climbing gear](https://acmeoutdoor.example/climbing): Harnesses, ropes,
  carabiners and protection.

## Policies

- [Shipping](https://acmeoutdoor.example/shipping.md): Rates, carriers and
  delivery windows by region.
- [Returns](https://acmeoutdoor.example/returns.md): 60-day window, prepaid
  label, refund timing.

## Optional

- [Size guide](https://acmeoutdoor.example/size-guide.md): Boot and clothing
  sizing conversion tables.

Every link needs real Markdown link syntax — [text](url), not a bare URL on its own line. That single requirement is also what Chrome’s Lighthouse Agentic Browsing audit checks for, and it’s the most common way real llms.txt files fail; see our breakdown of what Lighthouse actually checks if you’ve hit that failure. Keep the file itself small — it’s meant to fit in an agent’s context window whole; put the detail behind the links, not in the file.

Where this sits next to what else you publish

llms.txt is prose guidance, not structured data — it has no fixed schema for prices, availability or variants, which is what Product JSON-LD is for. It’s also distinct from newer, JSON-schema discovery files like ai-catalog.json, which lists callable resources (MCP servers, APIs, agent cards) rather than readable pages. A store can reasonably run all three: llms.txt as the readable index, JSON-LD on the product pages it links to, and ai-catalog.json if the store also exposes agent-callable tools.

FAQ

Does llms.txt help my Google ranking?

No — that’s a separate question from whether it’s worth publishing. Google’s own generative-AI optimization guide lists “LLMS.txt files and other ‘special’ markup” under things you can ignore for Google Search, stating plainly that “Google Search itself doesn’t use them” and that maintaining one “will neither harm nor help your site’s visibility or rankings in Google Search, as Google Search ignores them.” It’s read on demand by agents that already support it, which is a different, growing audience than Google’s crawlers.

Do I need a separate llms.txt for every category page?

No, only if you want more specific coverage than a single root file gives you. Per the v2 spec, one file at /llms.txt covers your whole site; a file at a subpath like /collections/hiking/llms.txt covers only that path, and where both exist, an agent should use the more specific one. Most stores only need the root file.

What’s the minimum viable llms.txt for a store?

An H1 with your store name and a one-sentence blockquote summary is spec-valid on its own — that’s the only required section. In practice, add at least one H2 section linking your main product categories, since a file with no links gives an agent nowhere to go next.

Sources

Whether a store publishes a well-formed llms.txt is one line item in the larger question AgentReady exists to answer — can an agent actually find, understand and buy here — which is what a free scan checks end to end.