SEO consultant Slobodan "Sani" Manic found that Chrome's new Lighthouse Agentic Browsing audit failed his llms.txt file with the error "File does not appear to contain any links," even though the file listed working links, because Lighthouse requires them in Markdown text syntax rather than plain text. Per Manic's July 3, 2026 report for Search Engine Journal, converting the links to Markdown format raised his site's Agentic Browsing score from 0.67 to a perfect 1.0, with no other changes made. Chrome's own developer documentation and DebugBear confirm the llms.txt audit separately checks for an H1 header and a 50-character minimum length. Google has said elsewhere that llms.txt is not required for AI search visibility.
The gap this story exposes matters more than the specific fix: Lighthouse's new Agentic Browsing category treats llms.txt as a Markdown document and fails files with plain-text links even when every URL resolves correctly, so a well-written, accurate llms.txt can fail a supposed agent-readiness check while a thin, plugin-generated file with Markdown links passes. For any team publishing machine-readable site summaries, passing the audit is a formatting exercise, not a proxy for whether the file is actually useful to an AI agent.
What happened
Slobodan Manic, an SEO consultant and host of the No Hacks podcast, ran Lighthouse's Agentic Browsing category (npx lighthouse@latest https://nohacks.co --only-categories=agentic-browsing) against his own site and reported the results in a July 3 Search Engine Journal article. Two of six audits passed (accessibility tree structure and cumulative layout shift), three WebMCP checks came back not-applicable, and the llms-txt audit failed with the verbatim error "File does not appear to contain any links," despite the file containing multiple working navigation and RSS links written as plain-text dashes rather than Markdown syntax. The category score was 0.67. After Manic rewrote each line using Markdown link syntax, text, changing roughly five characters per link while keeping the file served as plain text, the score rose to 1.0 and the audit flipped to "llms.txt follows recommendations."
Technical context
Chrome's own Lighthouse documentation confirms the llms-txt audit is part of an experimental Agentic Browsing category, introduced in Lighthouse 13.3.0, that scores sites on a fractional pass ratio rather than 0-100 because the underlying standards are still evolving. DebugBear's technical breakdown of the same audit specifies three separate checks: the file must contain an H1 header, must be longer than 50 characters, and must contain at least one Markdown-formatted link, text, failing any one fails the audit. Manic's testing indicates the parser does not credit plain-text URLs or dash-separated link lists as links at all, regardless of whether they are functional.
For practitioners
Teams that already publish an llms.txt file should re-run it through Lighthouse's Agentic Browsing audit (or DebugBear's equivalent scan), specifically checking for Markdown link syntax rather than just working URLs, since automatically generated files, such as those produced by default by the WordPress plugin AIOSEO, are more likely to already use Markdown links than hand-written ones. Passing the audit does not confirm the file's content is accurate or useful, so validation pipelines shouldn't treat a clean Lighthouse score as a substitute for reviewing what the file actually describes.
What to watch
Google has said llms.txt is not required for AI search visibility, a distinction Search Engine Land reports separates Lighthouse's agent-readiness checks from search-ranking guidance, so treat a failing score as a tooling-compatibility issue rather than an SEO penalty. Also watch whether Lighthouse's parser is updated to recognize plain-text links, since the current implementation rewards auto-generated, Markdown-formatted files over thoughtfully hand-curated ones that don't follow that syntax.
Key Points #
- 1Chrome's Lighthouse Agentic Browsing audit failed a working llms.txt file because its links were written as plain text instead of Markdown syntax.
- 2DebugBear documentation confirms the audit separately requires an H1 header, a 50-character minimum length, and at least one Markdown-formatted link to pass.
- 3Converting five characters per link raised the site's score from 0.67 to 1.0, showing format compliance, not file usefulness, drives the result.
Scoring Rationale #
Solid practitioner-relevant technical story with a hands-on, independently reproducible finding (verified via the original author's article plus Chrome's own docs and DebugBear), but it concerns one narrow, experimental Lighthouse audit rather than a broad platform or model change. Lowered slightly from the prior 6.1 to reflect that Google itself says llms.txt isn't required for AI visibility, capping the story's practical stakes.
Sources #
Public references used for this report. Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.