Is llms.txt worth it? An honest take for SaaS teams

For AI search visibility, no. llms.txt is a markdown file that points AI tools to your most useful pages, but Google says you don't need it for its AI features, and server logs show AI search bots almost never request it. It is worth an afternoon if you sell to developers or have API docs, because coding agents do read it.

Why is everyone asking about llms.txt?

Picture a SaaS founder opening an automated site audit. Page one, in red: "Missing llms.txt." Page two is a proposal to fix it, priced like a small website project. Her next move is to type the title of this post into ChatGPT. (She's made up, but that audit page is a very real genre.)

I get why it's tempting. robots.txt tells crawlers where they can go. sitemap.xml tells them what exists. A third little file that tells AI what matters sounds like the obvious next piece, and it's the rare AEO task you can finish before lunch and point to. (If you want the whole picture of what actually gets software named in AI answers, that lives in my guide on how to get your SaaS recommended by ChatGPT.)

The trouble is that "you can do it" and "it does something" are two different claims. So let's look at what the file is, who actually opens it, and when I'd bother.

What is llms.txt, exactly?

It's a proposal from Jeremy Howard, first published in September 2024 on llmstxt.org. You put a markdown file at /llms.txt with a specific shape:

• An H1 with the name of the site or project.
• A short blockquote summary.
• H2 sections that list links to your key pages, each with an optional note.
• An "Optional" section for links an agent can skip when it's short on room.

The proposal also suggests offering clean markdown copies of pages (the same URL with .md on the end), so a model doesn't have to dig your content out of navigation, cookie banners and JavaScript.

The best way I can describe it: it's the welcome binder on the kitchen counter of a vacation rental. Wi-Fi password, how the coffee maker works, which drawer has the corkscrew. Genuinely helpful. But nobody has ever booked a cabin because of the binder. They booked because of the listing, the photos and the reviews. (The binder helps the guest who's already inside.)

Keep that picture. It explains almost everything that follows.

Does ChatGPT or Google actually use llms.txt?

Here's the part the audit tools leave out.

Google has said no, more than once. In June 2025, Google's John Mueller wrote "FWIW no AI system currently uses llms.txt," adding that AI crawlers fetch pages but you won't see them request the file in your logs (Search Engine Roundtable). Google's own documentation for AI Overviews and AI Mode now says it plainly: "You don't need to create new machine readable files, AI text files, or markup to appear in these features" (Google Search Central).

Mueller also explained why it can't work for discovery. On Google's Search Off the Record podcast, he argued that an LLM can't trust a file a site writes about itself as a way to choose between sites. He did leave one door open: once someone, or some automated system, is already on your site, the file might help it get something done (Search Engine Journal). That's the binder again. It helps the guest inside; it doesn't book the cabin.

The server logs agree. Ahrefs studied 137,210 domains in May 2026 (Ahrefs). About 28% published a valid llms.txt, and 97% of those files got zero requests that month. Of the requests that did happen, AI retrieval bots, the ones that fetch pages to answer a question in real time, made up 1.1%. SEO audit tools made up 21.7%. Ahrefs' own conclusion was that, for AI search visibility, the file is "largely decoration."

I'll admit the irony: the biggest readers of llms.txt are the tools that tell you you're missing one.

So who does read it?

Coding agents. In that same Ahrefs data, AI agents and agent tooling were the largest slice of AI bot traffic, and Claude-Code was the second most active named AI tool, ahead of every AI retrieval bot and assistant except GPTBot (Ahrefs).

That fits the binder perfectly. A developer asks their coding assistant to wire up your API. The agent is already "inside the house," reading your docs on purpose, and a clean index of pages plus markdown copies saves it from guessing. That's why companies with big developer docs publish one. Anthropic's developer docs, for example, have an llms.txt that indexes the documentation and points to a full-text version.

Notice what that example is and isn't. It's a docs site helping agents use the docs. It isn't proof that anyone's AI search reads your marketing site's file to decide who to recommend.

The three-reader test: should your SaaS have one?

When a client asks me about llms.txt, I ask who the file would actually be written for. There are three possible readers, and only one of them shows up in the logs.

Then the decision is short:

• Make one if you sell a developer tool, have public API docs, or expect buyers to point an AI agent at your help center. Pair it with markdown versions of your docs pages, which is where the real value lives.
• Let your platform make one if your CMS or docs tool generates it for free. Free and harmless is fine. Just keep it accurate.
• Skip it if you're a typical B2B SaaS marketing site whose goal is getting named in AI answers. Spend that afternoon on the work that actually moves recommendations.
• Never pay for it as a project. If someone quotes you real money to "optimize your llms.txt," that's a binder with a mortgage.

What it depends on

"It depends" only counts if I tell you on what. Here's what flips my answer:

1. Who your user is. Developers increasingly read docs through an agent. Ops managers comparing tools mostly don't. The more technical your buyer, the more the file is worth.
2. How messy your pages are for machines. If your docs live behind heavy JavaScript or tabs that hide content, a markdown copy is a real improvement for an agent. If your pages are clean HTML, the gain shrinks.
3. Whether you'll keep it current. A stale llms.txt pointing at retired pages is worse than none. Ahrefs also flags that agents are built to trust the file, so treat it like code: review changes, and don't let anyone edit it casually (Ahrefs).
4. Whether a standard emerges. Proposals for agent-readable sites are still moving. If a major AI provider announces it uses llms.txt for search, I'll change my answer. That hasn't happened as of this writing.

What to do instead if you want AI to recommend you

The listing, the photos and the reviews. That's what gets the cabin booked, and it's what gets software named.

In practice, that means pages that answer buyer questions directly, comparison pages for the matchups people actually ask about, and a reputation on the sites AI engines read. If you're still getting your bearings on the basics, start with what AEO means for B2B software.

On the technical side, make sure AI crawlers aren't blocked in robots.txt and that your key pages render without JavaScript. Tidy schema markup helps engines understand who you are, though it's labeling, not a ranking lever. And measure before and after: set up AI referral tracking in GA4, then check your own server logs for requests to /llms.txt. If nobody's knocking, you have your answer for your site specifically.

If you'd rather talk through where your AI visibility actually leaks, grab 30 minutes with me. I'll tell you honestly whether the binder is the problem. It almost never is.

Here's to spending your afternoons on the work that books the cabin. 🍻