Skip to content
50% OFF $299 $599
Lock in
§ 6.0.3 ARTICLE
Published VerifiedEvery 6 weeks Sources3 named Authored bySquareRank Team

Glossary · § 6.0.3 · Defined term

llms.txt

llms.txt is a proposed plain-text manifest at the root of a site (e.g. example.com/llms.txt) that lists the URLs an LLM should prefer when answering questions about the site1. Jeremy Howard published the spec in September 2024. It is a publisher-side convention — no major AI engine has publicly confirmed that it reads or weights llms.txt as a retrieval signal as of May 2026.

Adoption is publisher-side: Anthropic publishes one2, Cloudflare publishes one, Perplexity publishes one. Adoption is consumer-side too in the sense that LLM-context tools (Claude Projects, custom GPTs, retrieval pipelines) accept llms.txt as an input. Whether the public AI search engines read third-party llms.txt files at answer time is unconfirmed.

Definition

llms.txt is a proposed plain-text manifest at the root of a site (e.g. example.com/llms.txt) that lists the URLs an LLM should prefer when answering questions about the site. It was proposed by Jeremy Howard in September 2024 and is a publisher-side convention — no AI engine has confirmed it as a retrieval signal.

The file is plain text formatted in markdown. It lives at /llms.txt on the domain root. The spec is permissive: one required H1 (the site name), an optional blockquote one-line summary, an optional prose section, and one or more H2 sections each listing markdown-formatted links to URLs the publisher wants the LLM to know about. A companion file at /llms-full.txt can contain a fuller content dump (often a concatenated markdown export of every key page) for use cases that want context, not just URLs.

What the file looks like

A minimal llms.txt is roughly five lines. A complete one is between 30 and 200 lines depending on site size. The structural skeleton: H1 site name, blockquote one-liner, optional context paragraph, then H2 sections like 'Docs', 'Articles', 'Pricing' with bulleted markdown links.

An example skeleton: # Acme Inc. on line one, then a blank line, then > Acme builds payroll software for European startups., then a paragraph or two of plain-text context, then ## Docs followed by a bulleted list of links to the key documentation URLs, then ## Articles with the equivalent for editorial pages. Each link is markdown-formatted with an inline summary. The whole file is human-readable; that is the point.

Where the proposal comes from

Jeremy Howard (cofounder of fast.ai, now at Answer.AI) proposed llms.txt on 3 September 2024 in a blog post and at llmstxt.org. The motivation: LLMs spend most of their context window reading content not intended for them — sidebars, footers, cookie banners — and a publisher-curated manifest of canonical URLs would let an LLM answer site-specific questions more accurately and more cheaply.

The spec is intentionally lightweight. It does not require structured data; it does not require schema validation; it has no consumer-side enforcement. The trade-off is that adoption is voluntary on both ends — publishers must ship the file, and LLM operators must choose to read it. As of May 2026, publisher adoption is widespread among developer-tooling companies (Anthropic, Cloudflare, Perplexity, Hugging Face) and modest elsewhere. Consumer-side adoption is largely undisclosed.

Adoption status in 2026

Publisher adoption: substantial in developer-tools and AI-infrastructure (Anthropic, Cloudflare, Perplexity, Hugging Face, Vercel, Supabase, MDN-adjacent docs). Publisher adoption in mainstream marketing sites: thin. Consumer adoption (AI engines reading third-party llms.txt at answer time): unconfirmed for ChatGPT, Perplexity, Claude, Gemini, AI Overviews. The closest to a confirmed consumer is the Claude Projects feature where users can manually attach a domain's llms.txt as context.

The honest position for a Squarespace owner: shipping llms.txt costs almost nothing once you know the URL Mappings workaround, and the downside is zero. The upside is uncertain but plausibly real, particularly for Perplexity (which has shipped publisher-facing tooling that suggests it cares about publisher-side signals more than ChatGPT does). We recommend shipping it; we do not promise it changes anything.

llms.txt on Squarespace specifically

Squarespace blocks root-level file uploads. The workaround: create a regular Squarespace page (e.g. /s/llms-txt) with the manifest body in a Code Block, then set up a 301 URL Mapping from /llms.txt to that page. The Content-Type header will be text/html rather than text/plain — most LLM tools accept this; some strict consumers will not.

The four-step install: (1) create a page in the Pages panel at slug /llms-txt; (2) paste the manifest body into a Code Block on that page wrapped in <pre> tags for monospace rendering; (3) open Settings → Advanced → URL Mappings and add /llms.txt -> /llms-txt 301; (4) verify by fetching https://yourdomain.com/llms.txt in a terminal or browser and confirming the manifest body loads3. The detailed cluster covers the content-type caveat and the validation steps; see Cluster 1.6 — llms.txt for Squarespace.

llms.txt sits inside the AI-search infrastructure cluster. Adjacent terms below.

  • GEO: llms.txt is one tactic inside the broader discipline.
  • Structured data: the schema-based sibling signal; structured data is consumed; llms.txt is read.
  • Cluster 1.6: the full install guide for Squarespace.
  • llms.txt template: copy-paste starters for service businesses, ecom, photographers, therapists.

Keep reading · § 6.0.3