Citably.

Notebook ·  Ed. 03  ·  August 8, 2026

Nothing is reading your llms.txt yet.
Add it anyway, but know why.

The question keeps coming up from founders: everyone says to drop an llms.txt on your site, so does anything actually read it? Honest answer, as of August 2026: no major AI engine has committed to consuming it in production, and Google has said outright that it does not support the file. It is cheap hygiene worth doing. It is not the visibility lever it gets sold as.

Get my free citability check See the full audit format

Finding 01

0

major AI engines confirmed to fetch /llms.txt in production.

Does any AI engine actually read llms.txt right now?

No. As of August 2026, none of the major AI providers have committed to reading llms.txt in their production systems. OpenAI, Anthropic, Perplexity, Meta, and Mistral have made no public commitment to ingest it on a schedule, and Google has stated plainly that it does not support the file. Check your own server logs and you will most likely see what we see in the logs we have access to: GPTBot, ClaudeBot, and PerplexityBot request robots.txt and your pages, and they do not request /llms.txt.

That is the gap between how the file is marketed and what it does. It gets pitched as the way to feed AI a clean version of your site. But a file only helps if something asks for it, and right now the engines that would matter are not asking. The page sits at your root, correctly formatted, and goes unread by the systems it was written for.

“No major AI engine has committed to consuming llms.txt in production as of August 2026, and Google has publicly said it does not support the file.”

Finding 02

suggests,
never forces

A curation file has no teeth. robots.txt does.

Is llms.txt the same thing as robots.txt?

No, and the confusion is worth clearing up. robots.txt is an access-control file: it tells crawlers what they may and may not fetch, and the major bots honor it. llms.txt is a curation file: a Markdown page that points to your most important content, like a hand-picked table of contents for a machine. It cannot block a crawler, cannot force a citation, and cannot change what a model retrieves. It only suggests where your best material sits, to any system that decides to look.

So the two files are not siblings. One is a lever with something on the other end, honored by the crawlers that matter. The other is a note left on the door in case a reader shows up who cares to read it. Treating llms.txt like a robots.txt for AI is the mistake that makes people expect a result it was never built to deliver.

Finding 03

~20

minutes to add it. That is the whole case for doing it now.

Then why does everyone tell me to add it?

Because it is cheap option value. A useful llms.txt takes roughly twenty minutes to write and almost nothing to keep current, and if adoption does arrive, the file is already in place. That is the honest case for it: not that it works today, but that the cost of being early is close to zero. You are buying a lottery ticket that happens to be free.

Where the reasoning breaks is when a twenty-minute file gets treated like a strategy. We have watched founders put llms.txt at the top of their AI-visibility to-do list and leave the things that actually move the needle further down. The order is backwards. Add the file, then forget about it, and spend the real hours on the work below.

The move

Add the file in an afternoon.
Then go win where it counts.

So should I add it, and what do I do instead?

Add llms.txt once, then put your effort into the three signals AI engines actually use today: original data only you can publish, clean schema markup so machines can parse what your page asserts, and passages written to be lifted verbatim with your name attached. Those are what get a page pulled into an answer right now. llms.txt is the twenty-minute box you tick on the way to that work, not the work itself.

Concretely: write the file so it lists your highest-value pages in plain Markdown with one honest line of description each, drop it at your root, and move on. Then go publish a number nobody else has, mark up your pages so their claims are legible, and structure your best answers so a model can quote them without breaking them. That is the sequence. The file is hygiene. The data, the schema, and the citable passages are the strategy.

The honest catch

“Nobody reads it” is a snapshot.
Not a permanent verdict.

What could make this wrong?

A negative is hard to prove across every system. We can read public statements and our own logs, but we cannot see inside every retrieval pipeline, and some smaller agents, dev tools, and experimental crawlers do already fetch llms.txt. So the honest claim is narrower than “nothing reads it, ever”: no major consumer AI engine has committed to using it in production, by their own word and by what shows up at the root of the sites we watch.

And this can flip. If one big engine announces it consumes llms.txt next quarter, the calculus changes overnight, and the sites that spent twenty minutes early will be glad they did. That is exactly why the recommendation is add it and forget it, rather than skip it or obsess over it. Cheap insurance against a future that has not arrived is still worth buying. Just do not mistake the insurance for the house.

Where this comes from

Public word,
plus the logs.

Two sources sit under this read. First, the public position of the major providers as of August 2026: none have committed to systematic llms.txt ingestion, and Google has stated it does not support the file. Second, crawler behavior in server access logs, where the named AI bots request robots.txt and content pages but not /llms.txt.

Treat the shape as real and the timing as provisional: no major engine consumes llms.txt today, the file is cheap enough to add regardless, and this is an August 2026 snapshot of a standard still being decided.

Want to know what AI actually says about you?

Free citability check.
24-hour turnaround.

We check the signals that move AI visibility today, not the ones that might matter later. Scoped to 3 findings, same method as the audits we sell.

Get my free check See a full audit

Filed by Jake Pereira · Founder, Citably · August 8, 2026 · Atlanta GA · ET

We re-check this one every quarter. The day a major engine says it consumes llms.txt, this post gets an update and the recommendation changes with it.