AI-referred 404s: recover the right destination without misleading redirects

Establish the evidence

Use authorised server logs, a reproducible referring link or a user-supplied response capture. Record the requested path, timestamp, response status and referrer if available. A missing referrer means the source is unknown; a suspicious path alone does not establish ChatGPT involvement. Avoid storing private conversation text or access tokens in a general audit sheet.

Match the intended task

Look for a former URL, a legitimate spelling variation or an existing page that genuinely answers the request. If a permanent replacement is clear, document a targeted redirect. If the path invents an unavailable service, do not send visitors to a vaguely similar sales page. A helpful 404 can explain that the resource is unavailable and offer search and relevant navigation while preserving the 404 status.

Validate the whole recovery

Test the old path, any redirect chain, final status, canonical and destination content in both languages. Preserve query parameters only when appropriate and safe. Update internal links that still point to an obsolete path. Record the observation period and subsequent request counts without calling a reduction in 404s a measured revenue gain.

An illustrative recovery decision

A fictional link requests an outdated AEO worksheet filename. If the current worksheet solves the same task, a narrow redirect may be appropriate. A made-up customer case should instead remain unavailable: inventing the case to satisfy the URL would create a factual problem. Keep an owner and review date for each mapping rather than redirecting every unknown path to the homepage.

Continue with the related decision

• AEO learning paths
• Agent-friendly websites: understandable information and authorised actions
• Discuss the scope of AEO and AI-search work