You have probably seen it pass by: a plugin that promises to make your site "AI-ready" with a single click. What you get is an llms.txt, a text file in the root of your domain containing a summary of your most important pages. The idea sounds logical. You neatly tell language models what is on your site, they pick it up, and you become more visible in ChatGPT, Gemini and Claude.

There is one problem. There is no evidence that a single major AI assistant fetches the file, let alone uses it.

What Google itself says about it

John Mueller of Google was unambiguous on 17 June 2025: at that moment, no AI system used llms.txt. That is not a vague disclaimer, that is a direct denial of the promise with which plugins such as Yoast offer the file.

Google Search Central confirms the line: you do not need special AI files to appear in generative search results. Mueller even drew a comparison that is painfully accurate: llms.txt resembles the meta keywords tag. Search engines have ignored that for more than ten years, for exactly one reason. The site owner fills it in themselves. A signal you determine entirely yourself is by definition not a reliable signal.

Gary Illyes went a step further at Search Central Live. Google does not support llms.txt and has no plans to. On 3 December 2025 the file did briefly appear in Google's own developer docs, which was immediately presented online as proof of a U-turn. It disappeared again and no policy change followed. A file in a documentation repo is not a crawl instruction.

"No AI system currently uses llms.txt." John Mueller, Google, June 2025

So who does use llms.txt?

Adoption exists, but it sits in a very specific corner: developer tools, AI companies, technical documentation and SaaS platforms. There, the file does have a certain logic. If you offer API documentation and developers manually feed that documentation into a coding assistant, then a clean, summarised text file is practical. That is a concrete use case with a human user who supplies the file themselves.

What you do not see is adoption among ordinary brands, webshops, service providers or media companies. Precisely the parties the plugins are now selling the file to. The gap between the niche where it works and the market it is being offered to is the core of the problem.

Crawl budget is not free currency

People often say: if it does not help, it does not hurt. That is too easy, because crawling costs something.

Cloudflare Radar publishes the crawl-to-refer ratio: how many pages a bot fetches per visitor it sends back. In July 2025, ClaudeBot crawled 38,065 pages per referred visitor. Six months earlier that ratio was still 286,930 to 1. Bingbot sat around 40 to 1 in the same period.

That means AI crawlers fetch an enormous amount and give little back. In that context it is unwise to spend crawl attention on a file that demonstrably nobody fetches and that you also have to keep up to date. An outdated llms.txt with dead links and old prices is worse than no llms.txt. It is maintenance without return, in other words expensive symbolism.

The traffic AI does deliver

The reason people reach for quick fixes is understandable: AI traffic converts remarkably well. In June 2025, Ahrefs saw that 0.5% of traffic from AI search engines accounted for 12.1% of signups. That is roughly 23 times higher conversion than average.

But benchmarks contradict each other, and that comes with the territory. Semrush arrived at 4.4 times higher conversion for AI traffic in 2026. Amsive found no statistically significant difference at all in its own research, with a p-value of 0.794. Three serious parties, three outcomes.

The conclusion is not that everyone is wrong. The conclusion is that your own figures weigh more heavily than someone else's benchmark. Certainly more heavily than a single-file trick that promises to solve everything in one click. How you measure that is set out in the measurement method.

Visibility in the answer matters more than the click

Semrush analysed billions of visits across 17 sectors between January and December 2025. AI traffic grew 66%, but remained small at a share of 0.14% of total traffic.

That is where the real lesson lies. If you only look at clicks, AI visibility is a rounding error. But the user who asks in ChatGPT which providers in the Netherlands are the best fit gets an answer with or without your name in it. That comparison happens in the answer, not on your site. If you are not mentioned, you drop out before there is anything to click.

That is also why many sites are technically perfect but invisible: every SEO checkbox is green, and the brand does not appear in AI answers.

What does have an effect

None of the following points is a one-file trick. They work because they improve the source from which models assemble their answers.

That last point is where llms.txt is definitively exposed. The file delivers no measurable outcome. There is no dashboard showing that it was fetched and no link to a rise in mentions. Demonstrated, not assumed, is the standard. llms.txt does not meet that standard.

So should you throw it out?

No, not necessarily. If the file is already there and it generates itself without maintenance, the damage is limited. See it as a bet on a standard that may one day be adopted.

What you should not do is treat it as a strategy. No budget, no project plan and no quarterly target that leans on a file which Google has publicly said twice is not being used.

If you want to know whether your brand is currently mentioned in ChatGPT, Gemini and Claude, start with a measurement instead of a plugin. Take the free AI check and then look at which steps really contribute to improving the AI visibility of your brand.

A text file in your root changes nothing about what models say about you. The content they draw from does.