Build1 publisher3 min readPublished
Chrome ships an llms.txt audit that a spec-valid file can fail
Version 2 of the llms.txt proposal landed in August, Lighthouse has audited the file since Chrome 13.3, and three of the four big WordPress SEO suites generate one. Google Search still says nothing fetches it.
The Engineer · Build desk
What happened
- Chrome's Lighthouse has carried an Agentic Browsing category since version 13.3, and one of the four audits in it is named "llms.txt does not follow recommendations."
- Chrome's documentation says a 404 marks that audit Not Applicable, and the category has no 0 to 100 score because, it says, the standards for the agentic web are still emerging.
- Three of the four biggest WordPress SEO plugins now generate the file in their free builds, and the default output of the five plugins compared varies in length by a factor of 200.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- decision The file costs a few kilobytes, so the choice that carries weight is which plugin writes it. The free builds disagree on how much to emit, and on whether to emit anything at all until an owner finds the switch.
- constraint An unscored audit that is skipped when the file is absent cannot support a ranking argument, so anyone selling llms.txt as an SEO win is selling participation in Chrome's data gathering.
- contradiction A site owner justifying the work by pointing at Chrome's audit is citing one Google product against another. Documentation use is the only claim with named evidence behind it.
- exposure While Issue 17082 is open, a pass or fail in PageSpeed Insights tells you about the Chrome build running the test.
The spec and the Lighthouse audit disagree about what counts as a file. The proposal at llmstxt.org requires one thing, "An H1 with the name of the project or site. This is the only required section." [2] A file whose entire contents are the line `# Example Shop` is valid. [3] The Lighthouse audit fails a file with no H1, a file that is too short, and a file without links, and passes everything else. [9] The smallest valid file fails two of those three conditions. [1]
v2 changed where a file applies, not what goes inside it. [4] The changes page states that "v1 permitted llms.txt files in subpaths without saying what that meant," and that "v2 defines it: a file covers the pages under its path, and the most specific file applies." [5] A page can now point at its file with `<link rel="describedby" href="/llms.txt">` in the head, or with an HTTP `Link:` header set at the CDN. A page's markdown twin may sit at `page.md` as well as the longer `page.html.md` that v1 required. [6] The `## Optional` section survives as a convention for links an agent can skip, and tools no longer read it as a mechanical signal. [7] DebugBear's writeup of the Agentic Browsing category is dated May 10, 2026, three months before v2 was published. The recommendations the audit checks were written before the current text existed. [8][2]
Whether anything reads the file is where Google's own products split. John Mueller, Search Advocate at Google, wrote on Bluesky on June 17, 2025: "FWIW no AI system currently uses llms.txt." [12] He pointed at server logs: "It's super-obvious if you look at your server logs. The consumer LLMs / chatbots (the ones that SEOs want traffic from) will fetch your pages - for training and grounding, but none of them fetch the llms.txt file." [13] Search Engine Journal ran the disagreement as a headline, "Google's llms.txt guidance depends on which product you ask." [14] The proposal narrows its own claim to one domain. It says the file is used "most heavily for software documentation, where coding agents follow them to find API references and tutorials." [15] A WooCommerce catalogue is not API documentation. No vendor has published a statement that its shopping or search crawler reads the file. [16]
Publishing is nonetheless widespread. On September 24, 2026, developers.openai.com/llms.txt, docs.claude.com/llms.txt and wordpress.org/llms.txt all answered 200. The WordPress file runs 75 lines, and it opens by telling agents that WordPress "powers more than 43% of all websites on the internet." [21] Howard writes that "thousands of sites now publish an llms.txt file." [22] A plugin directory search for llms.txt returned 521 results that same day, and the largest single-purpose plugin, Website LLMs.txt, lists 40,000 active installs. [18]
The defaults are where a site owner actually has to choose. The dev.to comparison of five free builds was written by the developer of Prime SEO, one of the five. It puts the difference in default output length at a factor of 200, and reports Yoast SEO 28.5 shipping the feature switched off. [19][24][23] Checking your own output in PageSpeed Insights is unreliable while Issue 17082 is open. A spec-compliant file returning 200 produced "Fetch of llms.txt failed" there, on an older Chrome build than the documentation calls for. [11] Publishing costs a few kilobytes and, with a plugin, no time. Crawler access to the HTML pages is still decided by robots.txt, and llms.txt does not replace it. [17]
What to watch
- Whether Lighthouse Issue 17082 closes and PageSpeed Insights begins fetching compliant files on the Chrome build the documentation specifies.
- Whether any vendor states that its shopping or search crawler reads llms.txt, which would extend the file's usefulness past software documentation.
- Whether Yoast turns the feature on by default in a release after 28.5, and whether the fourth big SEO suite ships a generator in its free build.