llms.txt in 2026: Is Anyone Actually Reading It?

You've been told to publish llms.txt or get left behind in AI search. You've also seen the take that it's a marketing stunt no crawler respects. So which is it — nice-to-have or existential?

Neither. It's a thirty-minute chore worth doing, and a terrible thing to build a strategy around. Let's walk through the numbers and what to ship instead.

What llms.txt actually is

llms.txt is a proposed convention: a Markdown file at the root of your domain that gives large language models a curated, machine-readable index of your most important content — short summary at the top, links to canonical pages below. Think "sitemap.xml, but for meaning, written for a reader that speaks Markdown." It is not a standard. It is not endorsed by the IETF, W3C, or any of the labs building the models — it is a good idea proposed by one developer that a handful of AI-native companies started shipping to signal intent.

Who publishes it now

Adoption is the friendly half of the story. A broad 300k-domain sweep found llms.txt on roughly 10% of sites; an Ahrefs sample weighted toward tech and SaaS put it closer to 28%. Both trend up year over year.

The list of publishers you'd recognize is short but telling: Anthropic, Cursor, Vercel, and a big chunk of dev-tooling documentation. That's not accidental — those are the companies whose customers are the AI agents. Publishing llms.txt is how they say, out loud, "we take machine consumption seriously." Everyone else has mostly not bothered, and that's a rational read on the usage numbers.

Who's actually reading it

Here's the part the "you must publish llms.txt" content mill glosses over. Adoption is climbing. Usage is essentially zero.

Server-log studies through mid-2026 found only about 3% of published llms.txt files receive any bot requests at all. GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, and Google-Extended overwhelmingly skip the file and crawl your HTML the way Googlebot has for two decades. When they hit /llms.txt, it's usually a one-off probe, not a production pipeline.

Even more damning: no major AI vendor has publicly committed to reading llms.txt in production. OpenAI hasn't. Anthropic hasn't — and Anthropic publishes one. Google's May 2026 guidance said explicitly that no special markup or llms.txt file gets you into AI features. Meta and Mistral have said nothing at all.

llms.txt has more consultants selling it than crawlers reading it. Ship the file. Don't build the strategy around it.

Why it still matters — but not as an SEO lever

The real value of llms.txt is as a Business-to-Agent (B2A) publishing surface. As agentic browsers and shopping agents proliferate — the story we tell in our post on agentic browsers and your website — sites need a machine-readable way to say "here is what I offer, here is how to route to it." llms.txt is a plausible starting point for that surface, alongside structured data, sitemaps, and clean semantic HTML. It's also a cheap forward-compatibility hedge: if a major crawler flips the switch in 2027, you're already there.

What it is not: a ranking factor, an AI-citation lever, or a substitute for the work that earns citations. That work is the same as it was — answer-first content, structured data, real E-E-A-T signals, and a page a crawler can actually reach. Our modern SEO strategy for 2027 post walks through every layer.

llms.txt: what it is and what it isn't

What DOES get you cited in AI answers

Publishing a Markdown index does not earn citations. What does:

• Answer-first content. The first 100 words answer the question the title implies.
• Real crawlability. Server-rendered where it matters. What's behind a scroll handler or client fetch is invisible to the summarizer.
• Real E-E-A-T signals. Named authors, real bios, real reviews on the page.
• Corroboration and freshness. Models cite sources they encounter often. Show up where your industry talks.

The working playbook is in get cited in AI Overviews: a 2026 playbook. Read that instead of a fifth "why you need llms.txt" post.

The 30-minute llms.txt starter kit

You're convinced it's worth publishing but not worth agonizing over. Here's the file that earns its keep. Put it at /llms.txt, plain Markdown:

• H1 with your company name.
• Blockquote with a one-sentence description of what you do.
• H2 sections for the parts of your site that matter (Services, Docs, Pricing, Blog), each with a Markdown list of links and short human-readable descriptions.
• An "Optional" section for pages that give context but aren't core (about, careers, press).

Ship it. Move on.

Where the standard is headed in 2027

Two things could happen. One: a major crawler starts honoring the file, everyone else follows, and the sites that shipped one in 2026 look prescient. Two: a competing spec emerges — W3C-adjacent, with a working group, handling licensing and agent authentication — and llms.txt becomes the Betamax of B2A publishing. Both are plausible, which is why the right posture is publish it, don't marry it.

Where people go wrong (and when to call a pro)

The expensive mistakes with llms.txt aren't technical. They're strategic.

Our services page covers what we handle. Short version: we ship the boring, durable work. The llms.txt file is a rounding error.

Frequently asked questions