llms.txt Examples: 4 Real Files, Annotated

If you already know what llms.txt is and just want to see real files, this is that page. Below are four llms.txt files fetched directly from live sites on September 1, 2026, quoted and dated, an annotated walkthrough of what each part of a file actually does, a template you can copy, and the mistakes people make when they copy an example without thinking about their own site. For the definition, format basics, and adoption data, start with What Is llms.txt? — this page assumes you've already read that or don't need to.

What makes an llms.txt file good, quickly

Every good example below shares the same four traits. This doesn't replace the full breakdown in the pillar guide's "What to include in llms.txt" section — it's the short version so the examples make sense:

• One H1 title naming the site or brand.
• A short blockquote summary — one or two honest sentences about what the site does.
• H2 sections of Markdown links, each with a one-line description of what a reader or model will find there.
• Canonical URLs only — no redirects, no duplicates, no dead links.

Nothing below dumps a full sitemap. That's the biggest difference between a good llms.txt and a bad one — curation, not completeness.

4 real llms.txt files, checked September 1, 2026

GeoReady — geoready.dev/llms.txt

Ours, so no third-party attribution needed beyond the date: accessed 2026-09-01. It opens like this:

# GeoReady

> AI visibility audit and monitoring platform built on the open-source GEO Optimizer engine. Score 0-100 based on 8 scoring categories derived from peer-reviewed research (Princeton KDD 2024, AutoGEO ICLR 2026).

From there it moves into detail sections (scoring categories, CLI commands, MCP tools) before the actual link sections — Main sections, Tools, Guides, Research sources, Companion files. It's long because the site is: dozens of guides, several tools, a monthly research report. Every link still gets its own one-line description; nothing is a bare URL. It also points to a companion llms-full.txt for systems that want expanded context instead of a curated index.

Supabase — supabase.com/llms.txt

Accessed 2026-09-01. Supabase takes the opposite approach to length — short and pointer-heavy:

# Supabase Docs

For the complete documentation in a single file, see [Full Documentation](https://supabase.com/llms-full.txt).

## Documentation
- [Supabase - Auth](https://supabase.com/docs/guides/auth.md)
- [Supabase - Edge Functions](https://supabase.com/docs/guides/functions.md)

Two things worth noticing: the file leads with a link to its own llms-full.txt for anyone who wants everything in one fetch, and its per-language API references (JavaScript, Python, Dart, Swift, Kotlin, C#) each get their own linked .txt file rather than being described in prose. It's built for machine consumption first, human skimming second.

Resend — resend.com/llms.txt

Accessed 2026-09-01. Resend's file goes beyond the bare format with a "when to use this" section aimed squarely at AI agents:

# Resend

> Resend is the email API for developers. Send transactional and marketing emails at scale with a simple, modern API.

## When to use Resend
Reach for Resend when a task involves sending, receiving, or managing email programmatically:
- Send transactional email... via POST https://api.resend.com/emails

Nothing in the llms.txt spec requires a "when to use X" section — Resend added it because their actual audience is coding agents deciding whether to call an API, not a person browsing docs. It's a reminder that the format is a floor, not a ceiling: the required part is the H1; everything else is a judgment call about your own audience.

Perplexity — docs.perplexity.ai/llms.txt

Accessed 2026-09-01. The most compact of the four:

# Perplexity

> Perplexity API documentation for building with the Agent API, the default for web-grounded AI and multi-provider applications, plus Search, Embeddings, and Sonar APIs.

- [Overview](https://docs.perplexity.ai/docs/getting-started/overview.md): Perplexity API Platform...
- [Quickstart](https://docs.perplexity.ai/docs/getting-started/quickstart.md): Generate an API key and make your first call in under 3 minutes.

No H2 section headers at all here — just a blockquote and a flat list of links, each with a description. That's still valid: the format doesn't require H2 sections, only that link lists, when present, are organized clearly. With one focused API-docs use case, Perplexity skipped the category layer entirely.

Annotated walkthrough: how GeoReady's own file is put together

Using our own file as the example, since it's the one we can explain with certainty:

• # GeoReady — the H1. This is the only element the format strictly requires. It names the entity the rest of the file describes.
• The blockquote — one sentence stating what the product is and the research it's built on. This is what a model reads first if it reads nothing else.
• The paragraph after the blockquote — expands the summary in plain language (who it's for, what problem it solves). Optional, but it's where nuance that doesn't fit a one-line summary goes.
• Detail sections like "Scoring categories" and "CLI commands" — not link lists. Prose or data placed between the summary and the link sections, useful when a short description isn't enough context (our scoring weights, for instance, need more than one line).
• "Main sections," "Tools," "Guides," "Research sources," "Companion files" — the actual file-list sections: H2 headings, each containing Markdown links in the format - [Title](url): one-line description. Every llms.txt needs at least one section like this.
• "Companion files" — points to llms-full.txt, an extended version for systems that want more than a curated index. Not required, and most sites don't need one.

The lesson that transfers to a smaller site: sections exist because there's enough distinct content to justify grouping it, not because a "complete" llms.txt needs six H2 headings. A five-page site doesn't need "Research sources."

Common mistakes when you copy an example

• Copying the size, not the judgment. A 60-link file makes sense for a site with 60 pages worth linking. It doesn't make sense for a five-page marketing site — you'll end up padding with pages that don't matter.
• Leaving someone else's URLs in. It happens more than you'd think when people copy a template fast. Every link in your file needs to resolve on your own domain.
• Linking redirects or non-canonical URLs. Point to the final address, with your site's actual trailing-slash convention.
• Treating optional additions as mandatory. Resend's "when to use" section and GeoReady's "companion files" pointer aren't part of the required format — borrow the idea only if it fits what your file is actually for.
• Blocking the pages you just linked. If robots.txt disallows a URL, listing it in llms.txt doesn't override that block.
• Publishing once and never touching it again. An example is a starting point for one moment in time; your file needs to track your site as it changes.

A template you can adapt

# Your Site Name

> One or two honest sentences: what your site does and who it's for.

## Docs
- [Getting started](https://example.com/docs/start): what a reader finds here.
- [API reference](https://example.com/docs/api): what a reader finds here.

## Product
- [Pricing](https://example.com/pricing): plans and limits, in plain terms.
- [Changelog](https://example.com/changelog): what shipped and when.

## Optional
- [Blog](https://example.com/blog): background reading, not required.

Swap in your own sections — Docs and Product won't fit every site. Keep the shape: H1, blockquote, sectioned links, one-line descriptions.

Honest limits

None of the four examples above earned extra visibility by publishing an llms.txt — no AI vendor has confirmed that any of them are cited more often because the file exists. Adoption is also not a steady climb: GeoReady's own State of GEO benchmark showed llms.txt adoption at 58.3% in the June 2026 edition, dropping to 54.2% in July (−4.1 points), then rebounding to 62.8% in the August 2026 edition (282 domains audited, average GEO score 56.4). Read that as a moving, uneven metric — not a settled best practice everyone has already adopted, and not a checkbox that guarantees anything once you tick it. Google does not use llms.txt as a ranking input.

Want the deeper mechanics — what the format actually requires versus what's just common practice, and where it's documented? That's covered in llms.txt standard, explained (publishing September 28, 2026).