Point it at the page that replaced it with a permanent 301 redirect, once you've checked that page actually answers. That single change fixes the URL for the person who clicks an old campaign link or email, and for the AI crawler that fetches the same address and otherwise gets nothing to work with.
This is a narrow, concrete case of the access layer the what is GEO pillar puts first: before identity, before content, before trust, a page has to actually answer when something asks for it. A dead URL fails that first layer for everyone who still has the old address, whether that's a person, a crawler, or a link on someone else's site you don't control.
Because an AI crawler that requests a URL and gets an error behaves the same way any crawler does: it records the failure and moves on, it doesn't retry patiently. The page is simply absent from whatever that engine builds next, until, if ever, it tries the address again. An old campaign link, an email footer, or a page on someone else's site that still points at your retired URL is sending both readers and crawlers into that dead end, and every one of those visits is wasted the moment the server answers with an error.
Why it matters: the link itself still has value, whoever is clicking it wanted something from you. A redirect recovers that value; a 404 throws it away a second time.
Two places show it without guesswork: Search Console's Pages report, and a direct request to the URL itself. Search Console lists pages Google can't index, including ones returning "Not found (404)", and many of those still show impressions or clicks from old links it hasn't forgotten yet. A direct HTTP request shows the same thing for any URL you already suspect, AI crawler log or not.
Do this: fix the URLs with real incoming links or search impressions first. A 404 on a page nothing ever pointed to costs nothing by comparison.
To the current page that answers the same question the old URL answered, not to your homepage by default. A redirect to an unrelated page reads as broken to the person who clicked it, and gives an AI crawler a page that has nothing to do with whatever it was trying to fetch, which is barely better than the 404 it replaced.
| What you find | What's actually happening | The fix |
|---|---|---|
| 404 Not Found | The page is gone and nothing replaced it at that address | 301 redirect to the page that now covers the same topic |
| Redirect chain of several hops | The page moved more than once, each move adding a hop | Point the original URL straight at the final destination, one hop |
| Redirect loop | Two URLs redirect to each other, so neither ever resolves | Remove one side of the loop and keep a single, final target |
| 429 or 5xx on request | Rate limiting or a server error, not a missing page | Raise or exempt the limit for verified crawlers, or fix the server error directly |
Bottom line: one clear redirect, straight to a page that answers, beats a chain, a loop, or a guess at the homepage every time.
It tells whoever requests the old address, person or crawler, that the page has moved there for good, and sends them straight to the new one. Google's own documentation treats a permanent redirect, status 301 or 308, as the reliable, server-side way to make that move, and uses it as a signal about which URL is now the canonical one. A redirect served from the server resolves before a page even has to load, which is faster and more dependable than a client-side trick like a meta refresh.
Proof: Google Search Central — Redirects and Google Search.
Why it matters: this isn't a Google-only mechanic. ChatGPT's search mode, Perplexity and Claude's web search all fetch a page before they can quote it, so a redirect that resolves cleanly reaches every one of them, not just classic search.
Yes, once you approve it. Ask Reflexa Agent to fix a URL that answers 404 or points at an old page, and it names the exact redirect, from which URL to which page, checks that the target actually answers before proposing it, and applies the 301 on your click. On WordPress and on a coded site it lands the same way every other approved change does, a revision or a commit; through a connected GitHub repository the redirect lands as a commit named for it; on Shopify, where the platform's own API is the door, redirects are one of the fields that API reaches. Undo takes it off again on any of them.
What stays yours to decide: the agent names the redirect and checks the target, it doesn't choose a destination you haven't approved.
None of this requires buying anything, and each step takes a few minutes:
Bottom line: most of these are one-redirect fixes once you know which URLs still matter, the hard part is finding them, not fixing them.
Set a permanent 301 redirect from the old URL to the page that replaced it, and confirm that page actually answers before you ship the redirect. That single change fixes it for a person clicking an old link and for an AI crawler fetching the same address.
No. A well-formed redirect that resolves quickly to a 200 is a pass, not a failure. What fails a crawler is a 404, a 429, a 5xx, or a redirect chain so long it times out before reaching a real page.
The closest current page that answers the same question the old URL answered, not your homepage by default. A redirect to an unrelated page reads as broken to a reader and gives a crawler nothing relevant to cite.
Both. ChatGPT's search mode, Perplexity and Claude's web search all fetch pages before they can quote them, the same way Google's crawlers do, so a URL that errors out is unusable to any of them, not only to classic search.
Reflexa Agent can. Ask it to fix a URL that 404s or points at an old page, and it names the exact redirect, checks that the target answers, and applies it on your approval, as a revision on WordPress, a commit on a connected repository, or through Shopify's redirect field; undo takes it off again.
The free check reads your priority pages from outside, the way a crawler does, and flags every one that doesn't return a clean 200. 3 minutes, evidence included.
Done for you by Reflexa Agent: how the agent fixes this on your site · the 23 checks it runs · Reflexa Agent på svenska