404 errors do not hurt your rankings.

Why 404s are normal

The web deletes things. Products are discontinued, jobs are filled, campaigns end, someone mistypes a URL in a link on a forum in 2019 and a crawler is still trying it. A site that has existed for a few years and returns no 404s at all is more suspicious than one that returns plenty, because it usually means something is answering 200 to everything, which is a genuine problem.

Google has been consistent on this: 404s are a normal part of the web and do not cause a ranking penalty for the site. The status code is doing exactly what it is for, telling a client the resource is not there. The page drops out of the index, which is correct, because it is gone.

The four cases that genuinely matter

A 404 on a URL that has external links. This is the only one with a direct cost. Someone linked to that page, the link still exists, and the signal now arrives at nothing. Redirect it to the genuine equivalent if one exists. This is worth checking specifically rather than treating all 404s alike, because it is where the money is.

Internal links pointing at 404s. A 404 someone reached from your own navigation is your fault twice: a visitor hit a dead end, and a crawler spent a request discovering that you link to something that is not there. External 404s are the web being the web. Internal ones are a bug.

A sudden spike. The count is not a signal but a change in it usually is. A jump after a release almost always means a template dropped a path segment, a migration missed a rule, or a bulk edit broke a URL pattern. The 404s are the symptom you are lucky to be shown.

Soft 404s, which are the actual problem. A page returning 200 while telling the user nothing is here is genuinely harmful, because it invites Google to crawl and assess pages with no content. That includes empty search results, discontinued product stubs and custom error templates returning 200. A real 404 is healthier than a fake 200 every time.

404 or 410

If something is deliberately and permanently gone, 410 says so explicitly, and Google tends to drop those slightly faster. It is a small optimisation and not worth rebuilding anything for, but for a bulk removal of content you know is never coming back, it is the more honest code. Where you are unsure, 404 is fine, and the difference between them is not something to spend a sprint on.

What to actually do with the report

Sort by whether the URL matters rather than by count. Check which 404ing URLs have external links, and redirect those to real equivalents. Find and fix internal links pointing at dead URLs. Watch for spikes as a release alarm. Then ignore the long tail, because the long tail is mostly crawler curiosity, ancient mistyped links and URLs that have not existed since a redesign three years ago, and none of that is costing you a position.

The instinct to drive the number to zero is the same one that produces 200-page audits: it treats every line in a report as equally actionable, when the work is deciding which handful of lines are worth anything at all.