What Is llms.txt? What It Actually Does for AI SEO (and What It Doesn't)

llms.txt is a proposed standard: a markdown file at your site's root that gives AI systems a curated map of your most important content — a table of contents written for language models. The pitch is compelling, the file costs almost nothing to add, and it has become one of the most hyped "AI SEO" tactics of the past two years. The measured reality is humbler, and worth knowing before you assign it any weight in your strategy.

What the file is

The proposal (from Answer.AI's Jeremy Howard in 2024) specifies a markdown file at /llms.txt containing an H1 with your site name, a short blockquote summary, and sections of annotated links to your key pages. A companion file, llms-full.txt, can inline the full text of those pages so a model gets your content without crawling at all.

Note what it is not: it is not an access-control file. robots.txt tells crawlers what they may fetch; llms.txt suggests what is worth reading. It grants nothing and blocks nothing — if you want to control AI crawler access, that still happens in robots.txt via user agents like GPTBot and ClaudeBot.

What the data actually shows

The honest picture, from 2026 measurements:

• Adoption is roughly one site in ten. An SE Ranking study of 300,000 domains put adoption at about 10%; a June 2026 Rankability analysis found 8.7% of the top 1,000 sites had one. Notable after eighteen months of advocacy — but not a stampede.
• AI crawlers barely fetch it. Server-log analyses are blunt: in one 90-day study, requests for llms.txt were about 0.1% of all AI-bot traffic — the crawlers overwhelmingly fetch your HTML directly. OpenAI's and Microsoft's bots are the ones observed requesting the file with any regularity.
• Google has said no. Google states it does not support llms.txt and has no plans to; John Mueller has compared it to the long-discredited keywords meta tag. It does nothing for Google Search or AI Overviews.

So the strong claim — "add llms.txt and AI engines will cite you" — is not supported by anyone's logs. If a vendor sells it as a ranking lever, discount accordingly.

Should you add one anyway?

A reasonable case for yes, with expectations set correctly: the file takes minutes to create, carries no risk, and a minority of real systems (notably OpenAI's) do read it. It is also a genuinely useful artifact — a curated, model-readable index of your best content — that may age well if adoption grows. Treat it like a business card, not a billboard: cheap, professional, occasionally picked up.

If you add one: keep it short and curated (your pillar pages and best guides, not every URL), write real one-line annotations, and update it when your key content changes. A stale llms.txt is exactly as useful as a stale sitemap.

What to do instead (or as well)

Everything that measurably drives AI citations happens in the content itself, not in a sidecar file: answer-first passages, question-shaped headings, fact density, FAQ schema, freshness, and entity signals beyond your own site. That is the actual playbook, and we cover it in Generative Engine Optimization (GEO). And since AI engines fetch your HTML directly, the boring fundamentals — crawlable pages, clean structure, solid platform SEO — remain the real prerequisite for being cited.

DraftRank sits on the same side of this line: the articles it generates and publishes are structured for extraction and citation, because the content is where the leverage is — with or without an llms.txt at the root.