Missing llms.txt in an AI audit: what the warning means
Check the audit’s exact URL, HTTP response and purpose before treating a missing llms.txt file as an AI-search problem.
Imagine an AI audit gives your website a red card: “Missing llms.txt.” Before adding a file to clear the warning, open the finding and identify which URL and check produced it. A missing optional index, an HTML error page returned as a success, and a server failure call for different decisions. None of those observations, by itself, tells you whether an AI answer will cite your site.
Find out what the check measures
Save the tool name and version, date, exact warning and tested URL. Read that check’s documentation. At our source check on November 9, 2026, Lighthouse’s llms.txt documentation said a server error triggers a flag and a missing file returning 404 is Not Applicable. It described the file as optional. Another tool may simply report whether it found a file. Its red label does not change what that request actually returned.
For Google AI Overviews and AI Mode, Google says no new AI text file is required. That answers a Google Search eligibility question. It does not establish how every other assistant handles an index. Keep the audit criterion and the platform requirement separate when deciding what to fix.
Read the response behind the result
Open the exact URL the audit checked. To inspect the request in Chrome, open the Network panel, reload the URL and select its request. Look at the request method, final URL after redirects, status, headers and Response body. A HEAD request or a Content-Type header alone cannot show you what a GET returned.
Consider a fictional documentation site, docs.example. Its /llms.txt returns 404, but /docs/llms.txt returns a Markdown index. The current llms.txt proposal allows files at a root or a subpath, covering pages below that path. A root-only audit has not inspected the documentation index. That is a difference in scope, not proof that the file must be duplicated at the root. Check where the intended client looks before changing the placement.
Now suppose /llms.txt returns HTTP 200, but its body is a branded “Page not found” HTML screen. A check based only on the status might call this a success. Record the actual result: the URL returned an HTML error page, not the intended Markdown index. Ask whoever manages the routing to investigate that response. If you instead get a homepage, a login page or an access challenge, record that result; the text/html header alone cannot distinguish them.
If the request returns 500, record a server error. It does not establish that the file is missing. If it returns readable Markdown, inspect the content and links too. The proposal requires a project or site heading, but a heading alone does not show that the index describes the current site or points to useful pages.
Decide who the file is for
A maintained documentation index can give a person or an agent a short route into relevant material. OpenAI publishes one for its developer documentation. That shows a published index exists. It does not establish a requirement for other websites, or show that ChatGPT Search fetches their files.
If you have a specific client or documentation workflow that will use the file, identify the pages it should find. Keep the summary accurate, follow the links and check that their destinations are current and readable. Include the index in your update process when those pages move. A small site can have that use case too.
If your aim is simply to make Google’s AI features find an otherwise useful page, the missing-file card has not established a new requirement. Check the page’s real access and content issues using the content and technical basics guide. Adding an index cannot establish that the linked page is accessible or that its claims are clear.
Close the fix with evidence you can observe
For the fictional site, suppose you correct a link in /docs/llms.txt that pointed to a deleted setup guide. Reopen the index, follow the corrected link and confirm the expected guide loads. Rerun the original audit with the same tool version, record its time and confirm you are viewing the new run’s result. If it still checks only /llms.txt, its result may remain unchanged even though the documentation link is fixed.
Keep the completion record precise: “The documentation index links to the current setup guide; the root-file check still reports its own result.” A later logged request shows only what that record captures, not that a model used the content. A citation needs an observed answer. Whether this edit changes citations, visits or conversions remains unverified until you have evidence for those outcomes.