What Is llms.txt and Do You Need It?

Probably yes, you should add an llms.txt file, but for reasons most articles get wrong. It is a 30-minute insurance policy, not a growth lever. Confusing the two is why founders waste a month optimizing a file that no major AI engine reads.

We run an AI-visibility engine across 10+ brands. We ship llms.txt on most of them. We have also watched, in our own analytics, that it moved citations by roughly nothing. Both things are true at once, and that is the part the SEO-tool blogs bury. This is the honest version.

What is an llms.txt file, in plain English?

An llms.txt file is a plain Markdown document at yoursite.com/llms.txt that gives a large language model a clean, summarized map of your site. It was proposed by Jeremy Howard of Answer.AI and published in September 2024, with the spec living at llmstxt.org.

The structure is simple: a required H1 with your project or company name, a blockquote summary, optional free-text sections, and H2 headings that list links to your most important pages. The stated purpose is inference-time context — helping a model that is already answering a question pull the right pages — explicitly not training and not SEO. The spec's own reasoning is that AI context windows are too small to swallow a whole website, so a curated summary helps.

Do ChatGPT, Claude, Perplexity, or Google actually read llms.txt?

No. As of early 2026, no major AI company has committed to reading llms.txt in production. That includes OpenAI, Google, Anthropic, Meta, and Mistral. Their crawlers — GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, Google-Extended — overwhelmingly fetch your HTML directly and skip the file.

Google has been the bluntest. Gary Illyes said at a mid-2025 Search Central event that Google does not support llms.txt and has no plans to. John Mueller compared it to the old keywords meta tag: site-controlled, easy to game, and ignored for a decade. Some in the SEO community pushed back that the comparison is imperfect, which is fair — but the headline holds: Google is not reading it.

This is the single fact every "you need llms.txt now" post soft-pedals. If your goal is to get pulled into an AI answer today, the file is not the mechanism. The mechanism is your actual content being crawlable, extractable, and corroborated across the web. We cover that in depth in why AI crawlers can't see your website.

Does llms.txt help SEO or get me cited in AI answers?

Not measurably, today. SEO-citation studies show no detectable ranking or AI-citation lift from adding the file. It does not influence Google rankings, and it does not move whether ChatGPT or Perplexity names your brand.

Here is the hard truth we tell every founder: roughly 80% of getting recommended by machines is classic SEO done right — crawlable pages, real authority, content that answers the question. The new 20% is extractability, cross-web consistency, citations on trusted third-party sources, and measuring whether your brand shows up inside the answers. llms.txt sits in neither bucket cleanly. It is table-stakes hygiene, not the engine.

If you want the model that actually drives recommendations, read how ChatGPT decides which brands to recommend. The levers there — being mentioned on sources the model already trusts, structured data it can parse, a consistent story across the web — are where the energy belongs. A Markdown summary file you control is, to an AI engine, just another thing you said about yourself.

Is llms.txt the same as robots.txt or a sitemap?

No. They solve different problems, and the names being similar causes most of the confusion.

• robots.txt tells crawlers what they are allowed to fetch. It is about permission.
• sitemap.xml lists every URL you want indexed. It is about discovery and completeness.
• llms.txt is a curated, human-readable summary of your most important content, formatted for an LLM's limited context. It is about prioritization and context.

A sitemap says "here is everything." An llms.txt says "if you only read a few things, read these, and here is what they mean." The catch: robots.txt and sitemaps are universally respected by crawlers. llms.txt is respected by almost none of them yet. So the analogy that flatters llms.txt is exactly the one that overstates its impact.

How do I create an llms.txt file?

Create a plain text file named llms.txt, write it in Markdown, and serve it at your domain root so it resolves at yoursite.com/llms.txt. The minimum viable version takes well under an hour.

1. Start with an H1: your company or project name.
2. Add a blockquote one-paragraph summary of what you do and who you serve.
3. Write a short free-text section with the context a model would need — your positioning, your core offers, anything easily misread.
4. Add H2 sections (for example "Docs," "Products," "Key Pages") with bulleted Markdown links, each link followed by a short description.
5. Optionally add an "Optional" H2 for lower-priority links a model can skip when space is tight.

Keep it honest and current. A stale llms.txt is worse than none, because the one place it does get read — coding agents — will faithfully repeat whatever you wrote. If you publish docs, point the file at your cleanest, most extractable pages. While you are in there, make sure those pages carry proper structured data too; see schema markup, the language AI actually reads.

Who's actually using llms.txt, and why?

Developer-tool companies are the real adopters, because the file genuinely works in one place: IDE and coding agents. Tools like Claude Code, Cursor, and Continue load llms.txt to get project context when a developer is working against your docs or SDK.

That is why the adopter list skews technical: Anthropic, Stripe, Cursor, Cloudflare, Vercel, Mintlify, Supabase, LangGraph. These are companies whose customers are developers pointing AI tools at their documentation. The file pays off because the consuming tool actually reads it.

This is the use-case split nobody draws clearly. There are two worlds. One is dev-doc context for coding agents — real, working, valuable today. The other is consumer-brand visibility in AI answers — where the file does basically nothing. When a founder reads "Stripe and Anthropic use llms.txt" and assumes it will get their coffee brand cited in ChatGPT, they have crossed from world one into world two. The technique does not travel between them.

Why does Google say I don't need it but Chrome audits for it?

Because they are different teams with different jobs, and 2026 made the split visible. This is the freshest, least-covered angle in the whole topic.

In 2026, Google Search Central stated that site owners should "not need to create machine-readable files, AI text files, markup, or Markdown" to be understood by Search. Around the same window, Chrome's Lighthouse shipped an llms.txt audit that flags server errors when retrieving the file.

So Google Search is telling you not to bother, while Google's own browser tooling added a check for the file. There is no contradiction once you see the lanes: Search is talking about ranking and indexing, where llms.txt is irrelevant. Lighthouse is browser-agent tooling, anticipating a world where on-device and browser-resident agents might consume the file. Same parent company, opposite signals, and a lot of founder confusion in between. The practical read: ship the file so the Lighthouse audit passes and agents can use it, and ignore any implication that it helps you rank.

So should your business add one in 2026?

Yes, for most sites — as cheap optionality, not as strategy. The math is simple: a basic file costs a few hours once, the downside is near zero, and there are two upsides. First, coding agents already use it. Second, if browser-agent adoption grows the way Lighthouse seems to bet, you are pre-positioned. That is a reasonable low-cost, low-yield bet.

What you should not do is treat it as your AI-visibility plan. We have never seen a brand get cited because of its llms.txt, and we are not going to promise you will be the first. If you want to know where llms.txt sits relative to the work that actually compounds, our AEO maturity ladder places it where it belongs: a hygiene rung, not a growth one. Ship the file in an afternoon, then go spend your real month on extractable content, structured data, and earning mentions on sources the models already trust.

Italo Campilii

Founder, Acromatico · runs AI visibility & brand systems across 10+ live brands