Lighthouse Fails Your Llms.txt Without Markdown Links

Chrome’s caller Lighthouse Agentic Browsing audit treats your .txt record arsenic a markdown document. If your llms.txt does not usage markdown nexus syntax, you neglect the audit, moreover erstwhile each nexus successful the record is meticulous and works. I ran the audit connected nohacks.co. Two of six audits passed. Three came backmost not applicable. One failed: the llms.txt audit, pinch the verbatim correction “File does not look to incorporate immoderate links.” The hole was 5 characters per link. The record is still served arsenic plain text. Only the audit consequence changed.

Lighthouse 13.3.0 shipped the Agentic Browsing class alongside Performance, Accessibility, SEO, and Best Practices. Six audits successful the default set: accessibility character well-formedness (agent-accessibility-tree), cumulative layout displacement (cumulative-layout-shift), llms.txt discoverability (llms-txt), and 3 WebMCP checks (webmcp-registered-tools, webmcp-form-coverage, webmcp-schema-validity). The class returns a fractional walk ratio alternatively of a 0-to-100 score, because the standards for the agentic web are still successful motion.

1 Of 6 Audits Failed On Nohacks.co

I ran the audit via the Lighthouse CLI: npx lighthouse@latest https://nohacks.co --only-categories=agentic-browsing. Six audits returned. Three came backmost not-applicable, each WebMCP: webmcp-registered-tools, webmcp-form-coverage, and webmcp-schema-validity. Lighthouse gives nary logic for a not-applicable result, it conscionable marks the audit and moves on. nohacks.co does expose WebMCP, but only done the experimental imperative navigator.modelContext API (two glossary tools, 2 for an agentic-browser directory), pinch nary declarative shape annotations. The scan ran successful a default headless Chrome 150 pinch nary WebMCP flag, truthful the not-applicable verdict could mean the website exposes thing these audits recognize, aliases that the scan situation had nary WebMCP API progressive astatine the time. Lighthouse does not opportunity which. Two audits passed cleanly: agent-accessibility-tree reported “All audits passed,” confirming the semantic HTML and ARIA building is well-formed capable for agents to navigate, and cumulative-layout-shift came backmost astatine zero.

One audit failed: llms-txt. The verbatim correction connection from Lighthouse was:

File does not look to incorporate immoderate links.

The class people was 0.67. That was the first surprise. The record astatine nohacks.co/llms.txt has galore links. Navigation paths to articles, episodes, guests, the glossary. RSS provender URLs. Audio record URL patterns. The record is complete 5 kilobytes of system content. So why was Lighthouse reporting zero links?

Lighthouse Parses .txt As Markdown And Rejects Plain-Text Links

The record hold is .txt, but Lighthouse parses the contents arsenic markdown, and demands markdown nexus syntax for immoderate matter to count arsenic a link. The record is named llms.txt. The HTTP server returns it pinch a text/plain MIME type. Open it successful a browser, and you spot plain text. But the llms.txt specification astatine llmstxt.org defines the format arsenic a markdown document. The spec is explicit: “Each conception contains a markdown slug database of links. Each database point has a nexus followed by optional notes astir the link, separated from the nexus by a colon.” Lighthouse’s parser enforces that strictly. Every nexus must beryllium encoded arsenic markdown nexus syntax, [text](url), pinch quadrate brackets astir the nexus matter and parentheses astir the URL.

My record had been utilizing a much earthy plain-text format:

Same destinations. Same descriptions. Same information. Lighthouse’s parser does not registry those lines arsenic links. Across the full file, it registered precisely zero. Audit fails.

A record pinch a .txt extension, served pinch a text/plain MIME type, that fails an audit unless it is formatted arsenic markdown. That is simply a mismatch the audit furniture is going to person to beryllium much honorable about. The hold says 1 thing. The MIME type says 1 thing. The parser is the root of truth, and the parser demands markdown.

The Fix Is 5 Characters Per Link

Wrap each nexus target successful markdown bracket-paren syntax, [text](url), and switch the - separator earlier each explanation pinch : . Five characters per link. Mechanical conversion, repeated crossed the file.

I made the edit. Re-ran the audit. Score went from 0.67 to 1.0. The audit title flipped from “llms.txt does not travel recommendations” to “llms.txt follows recommendations.” No item items successful the after-report. Clean pass.

The record is still served arsenic text/plain. The record hold is still .txt. The record contented is still the aforesaid content. Only the nexus encoding changed.

Lighthouse Measures Parseable Link Syntax, Not File Quality

The audit checks whether your record is mechanically parseable. It does not cheque whether the record describes your website usefully. Both sounds are existent astatine the aforesaid time.

The first read: The audit is measuring thing real. Markdown nexus syntax is mechanically parseable. Plain-text descriptive lines are not. If an AI supplier (or the Lighthouse parser opinionated successful for an agent) needs to extract links from the record programmatically, the markdown format is required. The audit is correct that the record earlier my hole could not beryllium parsed for links by the modular tooling. The conversion to markdown nexus syntax fixes a existent interoperability gap.

The 2nd read: format compliance is not the aforesaid arsenic record quality. A thoughtfully-written, accurate, broad llms.txt that uses plain-text descriptions fails this audit. A thin, auto-generated llms.txt pinch markdown nexus syntax passes. The audit cannot show the quality betwixt the two. The WordPress plugin AIOSEO, utilized by complete 3 cardinal websites per its WordPress.org listing, generates llms.txt files pinch markdown nexus syntax by default, a default-on behaviour Glenn Gabe surfaced, and the plugin’s ain archiving confirms. Those auto-generated files usage markdown nexus syntax because that is what the generator emits. Most of them astir apt walk this audit. Most hand-curated, owner-aware llms.txt files astir apt neglect it.

That spread is worthy reasoning astir earlier treating the audit’s pass/fail arsenic a measurement of really agent-ready your website really is. The audit is checking whether your record is parseable. It is not checking whether your record is useful.

Should You Care About Lighthouse Agentic Browsing’s Llms.txt Check?

Yes, but narrowly. Lighthouse tin show you whether your llms.txt is parseable arsenic markdown. It cannot show you whether the record describes your website honestly. That cheque is yours. Open Chrome DevTools, click the Lighthouse tab, verify the Agentic Browsing class is checked, and tally Analyze connected your URL. The audit takes nether a minute. If it fails connected the no-links error, the hole is 5 characters per nexus and 5 minutes of editing. If it passes, the harder mobility is the 1 Lighthouse cannot ask. Was the record auto-generated by a plugin you did not configure, aliases did you constitute it yourself, and either way, does it picture what your website really is?

The Machine-First Architecture Structure pillar sits beneath each of this: information models earlier page layouts, rendering independence, contented that does not dangle connected client-side JavaScript aliases human-display defaults to beryllium machine-readable. The llms.txt audit is simply a constrictive cheque astatine that layer. The bigger structural question, whether your machine-readable aboveground describes your website accurately, is yours to run.

More Resources:

• • Google Confirms llms.txt Has No Current Implementation
  • Google’s llms.txt Guidance Depends On Which Product You Ask
  • What The Latest Web Almanac Report Reveals About Bots, CMS Influence, And llms.txt

This station was primitively published connected No Hacks.

Featured Image: Darko 1981/Shutterstock