Lighthouse Audit Fails llms.txt Files Without Markdown Links

Image credit: Search Engine Journal

A new Lighthouse audit designed to improve AI agent discoverability is failing llms.txt files that do not use markdown link syntax, even if the links themselves are valid, according to recent tests.

The issue stems from the llms.txt discoverability audit, introduced in Lighthouse version 13.3.0, which treats these files as markdown documents and requires specific formatting for link recognition.

The llms.txt specification, detailed on llmstxt.org, explicitly defines the file format as a markdown document, mandating bulleted lists of markdown links. This requirement ensures mechanical parseability for AI agents.

A test conducted on nohacks.co demonstrated that a valid llms.txt file containing plain-text links failed the audit, returning an error message stating, ‘File does not appear to contain any links’. The audit score for discoverability was consequently 0.67.

To pass the audit, links within the llms.txt file must adhere to markdown link syntax, such as [text](url). When the plain-text links in the test file were converted to this format, the audit score improved to 1.0.

This change occurred despite the file continuing to be served with a text/plain content type, indicating the audit’s strict enforcement of markdown structure over the HTTP header.

The Agentic Browsing category, part of the Chrome Lighthouse tool, includes six audits aimed at enhancing how AI agents interact with and understand web content.

The primary function of the llms.txt discoverability audit is to measure the mechanical parseability of the file for AI agents, rather than assessing the quality or usefulness of the file’s content, according to Lighthouse documentation.

Web developers using platforms like WordPress or other content management systems will need to ensure their llms.txt files conform to markdown link syntax to achieve optimal Lighthouse scores.

Source: Search Engine Journal