llms.txt

The convention was proposed in September 2024 by Jeremy Howard of Answer.AI, with a specification published at llmstxt.org. It is a community proposal, not an official web standard, and whether any given AI engine reads or honors the file varies, so it's best treated as low-cost hygiene rather than a guaranteed ranking lever.

The file itself is Markdown served at /llms.txt. A typical file opens with the site name and a one-line description, then lists curated links to the pages that matter most, often pointing to clean Markdown versions of those pages. The idea is to hand a model a high-signal index of your best content instead of making it crawl and parse noisy HTML to figure out what your site is about.

It's important to be clear about what llms.txt is not. It is not an access-control or ranking mechanism, that's what robots.txt and your normal SEO signals do. llms.txt doesn't block or grant crawling; it's a curation and wayfinding aid. Adoption is still early and uneven across engines, so the honest framing is "cheap to add, potentially helpful, not a silver bullet."

For a store, llms.txt is a chance to point AI systems straight at the content that drives buying decisions, top collections, sizing and care guides, shipping and returns policies, in one clean, machine-friendly file instead of hoping a crawler finds them among thousands of product URLs.

The caveat matters as much as the tactic: because engine support is uneven and unconfirmed, llms.txt should be part of a broader GEO program (retrievable, citable, well-structured pages), not a substitute for it. It's a hygiene step, not the whole strategy.

Illustrative scenario: a store publishes an /llms.txt that lists its size guide, shipping and returns policy, and top three collection pages, each linked to a clean Markdown version. When an assistant is asked "what's their return window," the canonical policy page is one hop away and easy to parse.