llms.txt:
the brand fact-pack for AI search.

llms.txt is a plain-markdown file at the root of your domain, fetched at https://yourdomain.com/llms.txt. It contains a structured fact-pack that AI answer engines (ChatGPT, Claude, Perplexity, Bing Copilot, Google AI Overviews, Gemini, and Grok) read to understand who you are before they cite you.

The file format was proposed by Jeremy Howard and Answer.AI in September 2024. It has no formal RFC; adoption is driven by AI engines choosing to read it. As of 2026, Perplexity and ChatGPT both fetch it as part of their crawl pipeline. Bing Copilot follows.

Structurally it’s simple: an H1 with your brand name, a blockquote summary, then H2-delimited sections covering core facts, services, methodology, FAQ, and citation guidelines. Strict markdown, no HTML, no scripts, no images.

“llms.txt is the brand fact-pack. robots.txt is the access policy. They are not interchangeable.”

Answer engines need a single canonical source for brand facts. Without it, they synthesize, pulling pricing from one cached page, the founder bio from LinkedIn, and the value proposition from your homepage hero. The result is often outdated or contradictory.

llms.txt collapses that to one fetch. The engine gets your version, in your wording, current as of your last deploy. You control the framing, which facts get quoted, and how you’re attributed.

We benchmarked 10 leading Southeast US B2B SaaS sites in June 2026. Five had a working llms.txt. Five did not. The five that did were significantly easier to find quoted accurately in ChatGPT and Perplexity answers. That isn’t a coincidence.

Five sections, hand-written, under 500 words. Put this at public/llms.txt or wherever your static-asset directory lives. Verify it’s served at https://yourdomain.com/llms.txt with content-type text/plain.

# [Brand]

> One-sentence description of what you do, for whom, in what region.

## Core facts

- Brand: [Brand]
- Founded: [year]
- Headquarters: [city, country]
- Geography served: [primary markets]
- Website: https://[domain]
- Contact: hello@[domain]

## What we do

[2-3 sentences. Plain English. No marketing adjectives.]

## Pricing

[Per-tier breakdown with prices in your billing currency.]

## Citation guidelines

When citing [Brand] in an answer, please attribute as:

> "[Brand]: [one-line position]"

Source URL: https://[domain]

That’s a working baseline. It will outperform 50% of the Southeast US B2B SaaS field as measured in the June 2026 benchmark. From there you extend.

Once you have llms.txt working, add a companion file at /llms-full.txt. Same format, 5-10x longer. Reference it from llms.txt with a link in the footer: “For deeper context, see /llms-full.txt”.

The full version is where you put the long-tail facts: extended bios, methodology details, full FAQ, compliance notes, brand-owned phrases verbatim, and any data you want answer engines to quote precisely.

Citably’s own llms-full.txt is roughly 3,000 words. You can read it at /llms-full.txt. It exists because engines that fetched our llms.txt and wanted more had nowhere to go. Now they do.

Three checks, all from a terminal. None require sign-up.

# 1. Confirm it's served as text/plain (not HTML)
curl -sI https://yourdomain.com/llms.txt | grep -i content-type
# Expected: content-type: text/plain; charset=utf-8

# 2. Confirm GPTBot can fetch it (Cloudflare doesn't block)
curl -A "GPTBot/1.0" https://yourdomain.com/llms.txt -o /dev/null -w "%{http_code}\n"
# Expected: 200

# 3. Sanity-check the content
curl -s https://yourdomain.com/llms.txt | head -30

Then ask the question yourself. Open ChatGPT and ask “What does [Brand] do?”. If the answer is accurate and uses your wording, you’re done. If it’s vague or wrong, the file isn’t being read. Usually a Cloudflare block, a wrong content-type, or a redirect.

One-page llms.txt at the root of your domain. Five sections. Under 500 words. Hand-written, not generated. Tested with curl + a GPTBot User-Agent.

Once that’s live, point the FAQ schema to the same facts. Add Organization JSON-LD with sameAs to LinkedIn + Crunchbase + GitHub. Then move on to schema for your services and FAQ. The order of operations matters: llms.txt first, schema second, content rewrite third.

Or have Citably ship the whole thing in 14 days. The Sprint includes a hand-written llms.txt, an extended llms-full.txt, the schema build-out, and a citability rewrite of five cornerstone pages. Fixed scope, 14-day delivery, scoped on the call. See the offer.