Should you publish llms.txt
or ignore the hype?

What is the practical answer on llms.txt?

The practical answer is simple. If your team can generate and maintain llms.txt without drama, publish it. If doing that would distract you from fixing crawl access, server rendered content, internal linking, source clarity, and entity consistency, ignore the hype and do the foundational work first.

This is one of those topics where the internet rewards certainty more than accuracy. A lot of GEO commentary treats llms.txt as a direct line into AI assistants. The evidence does not support that leap. Google states llms.txt is not used by Search. A large study across about 300,000 domains found 10.13% adoption, 0% among the top 1,000 sites, and no citation lift after controls.

That does not make the file useless. It makes it limited. A limited file can still be worth shipping if the cost is tiny and the maintenance burden stays tiny.

What does llms.txt actually do?

At best, llms.txt is a structured hint. It can point a reader or a tool toward pages you consider important, canonical, and readable. Think of it as a shortlist, not a control panel.

• It can help you state which pages matter most
• It can reduce ambiguity for humans inspecting your site setup
• It can serve as internal discipline, because you must decide what your source documents really are
• It does not force crawlers to fetch those pages
• It does not make weak pages more quotable
• It does not repair rendering, access, or trust problems

This distinction matters. Teams often reach for a file because a file feels shippable. But AI visibility is mostly a content access and extractability problem. If the model pipeline cannot reliably fetch, parse, and quote your material, another text file at the root changes very little.

Why has llms.txt become so hyped?

Because it fits a pattern people like. It is new, named for LLMs, easy to explain in one sentence, and cheap to implement. That makes it perfect social media bait. It also lets vendors imply a fast win without talking about the boring work, like content architecture, source ownership, and crawl diagnostics.

There is also a second reason. Many teams are uncomfortable with uncertainty in AI search. When causality is messy, people overvalue anything that looks explicit. A root file feels explicit. Unfortunately, explicit is not the same as influential.

You also see bad statistics recycled here. The familiar GEO multiplier claims around tables, FAQ schema, and recency are often repeated without solid sourcing. Those are exactly the kinds of stats I would not use to make a technical decision. If the evidence chain is weak, treat the recommendation as a hypothesis, not a playbook.

When is publishing llms.txt still worth doing?

It is worth doing when it is a byproduct of good housekeeping. If you already know your canonical documents, have a clear documentation structure, and can regenerate the file as pages change, the upside is modest but real enough. You reduce confusion. You create a clean reference point. You spend very little.

That is how I would frame it to an operator. Not as a growth lever. As a hygiene task.

• Publish it if it takes little effort
• Keep it short and focused on source pages
• Update it when your key pages change
• Do not let it replace sitemap, schema, internal linking, or rendering work
• Do not report it as an AI visibility win unless you can prove that with before and after evidence

When should you ignore llms.txt for now?

Ignore it for now if your house is messy. That includes pages rendered client side, important text hidden in tabs that are not present in the initial HTML, weak title and heading signals, duplicate source pages, messy canonicals, or a site where the best information lives in PDFs and slide decks nobody can parse cleanly.

Also ignore it if your team will publish the file once and never maintain it. A stale llms.txt is worse than no file, because it creates false confidence internally. People assume the AI visibility box is ticked. It is not.

This advice also fails for teams that need political cover more than technical leverage. Sometimes an internal stakeholder wants to see a visible action. In that case, llms.txt can be a harmless concession, but be honest in the memo. Say clearly that it is not expected to change search performance or citation rates on its own.

What should you fix before llms.txt?

Fix the things that decide whether an assistant can access and quote you in the first place. Server logs and crawl tests matter more than trend chasing. We have good reason to care about rendering here, because AI crawlers do not execute JavaScript. They fetch JavaScript files and never run them. If your key content depends on execution, you are creating your own invisibility.

If this sounds basic, good. Basic is where most of the lift still sits. Foundational fixes are less glamorous than a new file standard, but they are more likely to change what gets crawled, understood, and cited.

How should you publish llms.txt if you decide to do it?

Keep it boring. Boring is good. Include the few pages you would want a new analyst, partner, or machine process to read first. Prefer stable URLs with clear ownership. Use plain language titles. Do not dump every blog post into it just because you can.

• Point to canonical product, documentation, research, and policy pages
• Exclude thin pages, campaign pages, and near duplicates
• Review it whenever nav, docs, or core messaging changes
• Make sure every linked page is directly readable in HTML
• Treat it as a maintained asset, not a launch artifact

I would also avoid turning llms.txt into a theology debate. The better question is operational. Does this file help us present our source material more clearly at near zero cost. If yes, ship it. If no, skip it and move on.

Who should not follow the publish it anyway advice?

Teams with limited engineering time should be especially strict here. If your roadmap forces trade offs, choose the work that improves access, clarity, and quotability directly. llms.txt is not where I would spend scarce cycles.

Very small sites may not need it at all. If you have a handful of clean, server rendered, well linked pages with obvious canonicals, the file is unlikely to change anything material. Complex documentation sites, publisher archives, and large product catalogs may get more organizational benefit, but even there I would not oversell it.

So the honest founder answer is this. Publish llms.txt when it is cheap, clean, and maintainable. Ignore the hype when people present it as a shortcut to AI citations. There is no shortcut around accessible pages, clear sources, and consistent facts.