How to Fix 404 Errors (and When to Leave Them Alone)

A 404 means the page is not there. That is not a malfunction, it is an answer, and quite often it is the correct one. Pages get deleted. Products get discontinued. A URL somebody typed wrong never existed in the first place. In all three cases the honest response is the one your server already gives.

The finding in front of you says some of your URLs return an error. What it cannot tell you is which of those URLs deserved to live. That judgment is yours, and getting it wrong in the enthusiastic direction, by redirecting everything somewhere, does more damage than leaving the errors alone.

Why this matters for AI traffic

Google is explicit about what a 4xx does to your presence: “Google doesn’t index URLs that return a 4xx status code, and URLs that are already indexed and return a 4xx status code are removed from the index.” Crawl frequency for those URLs then decreases over time. So a page that dies quietly leaves the index quietly.

For a retrieval engine, that is usually fine. The cost lands somewhere else. Every dead URL that something still points at, an internal link, a sitemap entry, a link from another site, is a path that goes nowhere. A crawler spends its budget discovering the same absence repeatedly, and any authority pointed at that URL from outside arrives at nothing. The page is gone either way. The question is whether you are still advertising it.

The expensive version of this is the one Google warns about directly. Serving something other than an error, your homepage, a friendly “sorry, not found” page that returns 200, produces a soft 404: a page that says error but claims success. Those get crawled as though they were real content, so you have replaced one dead URL with an unbounded supply of duplicate live ones.

What a dead page should return

Google’s guidance is unusually blunt for a company that hedges: “Don’t create fake content, redirect to your homepage, or use robots.txt to block 404s, all of these things make it harder for us to recognize your site’s structure and process it properly. Instead, if your page is no longer available, and has no clear replacement, it should return a 404 (not found) or 410 (Gone) response code.”

Read the condition in that sentence, because it is the whole decision: has no clear replacement. It sorts every dead URL you have into two piles.

A clear replacement exists. The product moved, the post was rewritten at a new URL, the service page was renamed. Redirect it, permanently. Google’s redirect guidance is that a permanent redirect shows the new target in search results, while a temporary one keeps showing the source, so permanent is what you want when the move is real.
No clear replacement exists. The thing is gone and nothing else answers the same question. Return 404 or 410 and let it go. Sending that visitor to your homepage helps nobody: they came for one specific thing, and the homepage is a page that does not have it.

A 500 or a timeout is a different animal entirely and does not belong in this decision. Those are not answers, they are failures, and they say your server broke rather than your page moved. Fix the server. The SEO question does not start until the page reliably returns something.

When is it worth fixing, and when should you leave it?

Fix it when:

Something still points at the URL: an internal link, your sitemap, or a link from another site you would rather not waste.
The URL had traffic or rankings, and a page that answers the same question exists now.
The finding is a 500 or a timeout. That is not a judgment call, that is a broken server.

Leave it alone when:

The page is genuinely gone and nothing replaces it. The 404 is correct. Chasing it to zero is chasing the wrong number.
The URL was never real: a typo in someone else’s link, an old scanner probing for admin paths, a query string nobody generated on purpose.
The only replacement you can think of is the homepage. That is the tell that there is no replacement.

The honest tradeoff: a pile of 404s in a report looks like a problem, and cleaning it to zero feels productive, but most of that work moves nothing. What pays is the small subset with links or traffic pointed at them, where a redirect recovers something real. I would fix those, fix every 500, and let the rest 404 in peace.

Once you know which dead URL has a real replacement, the redirect itself is a one-line change. Get AI Traffic applies it with set_redirect, which points one URL at one target and leaves the rest of your dead URLs returning the error they should return. Note what that means: the tool takes the target from you. It has no way to know whether a replacement is genuinely a replacement or just the nearest page, and that distinction is the entire subject of this article.

Frequently asked questions

Do 404 errors hurt my rankings? Not as a penalty. Google removes 404 URLs from the index and crawls them less over time, which is the system working correctly. What costs you is a dead URL that internal links, your sitemap, or other sites still point at, because that traffic and authority now arrive nowhere.

Should I redirect my 404s to the homepage? No, and Google says so directly. A removed page with no clear replacement should return 404 or 410. A homepage redirect either becomes a soft 404 or dumps someone who wanted a specific thing onto a page that does not have it.

What is a soft 404? A page that reports success while showing an error. The server returns 200, but the content is an error message or a fallback page. Google detects the mismatch and reports it in Search Console, and the URL keeps getting crawled as though it were real content.

What is the difference between 404 and 410? 404 says not found, 410 says gone and not coming back. Both tell browsers and search engines the page does not exist. Use 410 when you know the removal is permanent, and 404 when you are not sure or do not care to be precise.

A 500 error and a 404 are both in my report. Are they the same problem? No. A 404 is your site answering correctly about a page that does not exist. A 500 or a timeout is your site failing to answer at all, on a page that probably should. Treat the 500 as an outage and the 404 as an editorial decision.

Not Every 404 Is a Bug, and Redirecting Them All Is Worse

Why this matters for AI traffic

What a dead page should return

When is it worth fixing, and when should you leave it?

Frequently asked questions

Sources