What is llms.txt?

A proposed convention (llmstxt.org, September 2024): a Markdown file at your site root that gives language models a curated, token-efficient map of what your site is and where the important pages are. Think of it as a press kit for machines — a short description, organized links, key facts — rather than a directive file like robots.txt. Nothing about it is enforced by any standard body; it's a handshake offered to whichever systems choose to read it.

The uncomfortable evidence (mid-2026)

We'd rather you hear this from us than discover it after paying someone for "llms.txt optimization":

  • Adoption is real but shallow. Adoption grew several-fold over the past year (SE Ranking measured roughly 10% of studied domains), yet a scan of the top 1,000 sites found almost none.
  • Most files are never fetched. Ahrefs' server-log study across 137,000 domains found 97% of llms.txt files received zero requests in May 2026.
  • Google is explicit. Search Advocates compared it to the old keywords meta tag, and Google's AI-optimization documentation was updated in June 2026 to state llms.txt has no effect — positive or negative — on rankings or AI Overviews.
  • No major AI lab has committed to it in production. The traffic that does arrive comes mostly from developer tooling — coding agents, IDE assistants, and MCP-connected apps that fetch it when a user points them at your domain.

Then why ship one at all?

Because the cost is fifteen minutes and the failure mode of not having one is invisible. Our reasoning, in order of weight:

  1. Agents that act on your site read it. The growing traffic isn't chatbots-answering-questions, it's tools-doing-tasks: an assistant asked to "find a supplier who ships same-day from Arizona" or a procurement agent comparing catalogs. A curated file is the difference between the agent parsing your mega-menu and reading your pitch.
  2. It forces entity hygiene. Writing one makes you state, in one place, exactly what the business is, where it is, and which pages matter. Inconsistencies you find here (old phone numbers, dead URLs, three different founding dates) are the same inconsistencies confusing every AI system that models your brand.
  3. It's a cheap hedge. If a major platform adopts the convention, you're done already. If none ever does, you spent fifteen minutes.

What we won't tell you: that llms.txt will get you cited by ChatGPT or ranked in AI Overviews. The measurable GEO work is elsewhere — crawlable content, entity-consistent structured data, being the source engines quote. That's the substance of our GEO service, and llms.txt is one small, honest line item inside it.

How to write one for a store

The spec is simple: an H1 with the site name, a blockquote summary, then H2 sections of link lists with one-line descriptions. For ecommerce, the structure that serves agents best:

# Acme Industrial Supply

> B2B distributor of rigging and lifting hardware since 1987.
> 12,000 SKUs, same-day shipping from Phoenix AZ, NET-30 for
> qualified accounts.

## Products
- [Chain slings](/slings/chain/): Grade 80/100, custom lengths.
- [Wire rope](/wire-rope/): Cut to length, tested, certified.
- [Hardware](/hardware/): Shackles, hooks, turnbuckles.

## Buying
- [Bulk & contract pricing](/wholesale/): Tiered account pricing.
- [Shipping](/shipping/): Same-day LTL and parcel cutoffs.

## Company
- [About](/about/): History, certifications, safety record.
- [Contact](/contact/): Phone, email, quote requests.

## Key facts
- Founded 1987 - family owned
- Ships from Phoenix, AZ - same-day before 2pm MST
- All rigging proof-tested and certified to ASME B30.9
  • Curate, don't dump. Ten to thirty links. It's a map for a reader with a token budget, not a sitemap replacement.
  • Lead with what an agent needs to act: what you sell, where you ship from, how buying works (wholesale terms, cutoffs, minimums). Facts an agent can quote beat adjectives it will ignore.
  • Every fact must match the site. Same phone, same address, same claims as your pages and your Organization schema. A contradicting llms.txt teaches machines to distrust the file that's supposed to orient them.
  • Verify every link resolves. Ours is checked automatically on every build; stale links are the most common llms.txt failure we see in audits.
  • Keep robots.txt in charge of access. llms.txt describes; robots.txt permits. If your robots.txt blocks GPTBot while your llms.txt courts it, you've written a welcome mat for a locked door.

Want the working example? Read ours — it's the same structure, live. And if you want your store's AI-readiness measured rather than guessed at, the AI Site Auditor checks llms.txt alongside the forty other things that actually move visibility.