301 Redirects: What Google Does With Them and Where They Fail Silently

The redirect table was ready. Every old address had its new destination. We tested all 57 pairs, one by one, and every one of them reached a live page.

Then we checked against another list, Google’s own. Five old pages turned up returning a 404 error. Together they had 437 impressions. None of them was in our table. ¹

This happened on my own site. Between August 31 and September 4, 2026, seogenome.com, our company site in Brazil, moved off WordPress. The new site is built in Astro and hosted on Vercel. ¹ The migration had a table, tests, and a final check. Even so, the redirects failed in five places the table test could not catch.

This piece explains what Google does with each type of redirect. Then it opens up the five failures, with dates and numbers. And it ends with a test you can run on your own migration.

What a 301 redirect is

Think about moving house. You file a change of address with the post office, and every letter for the old place goes to the new one. Mark it permanent, and the old address comes off the carrier’s list. Mark it temporary, for a month away, and the old one stays your real address.

On the web, that notice is a server response. Whoever requests the old address gets a code and the new address, and the browser follows on its own. This is called a redirect. The 301 code says the move is permanent. ²

Google sums up the difference by what shows up in search. A permanent redirect shows the target in the results. A temporary one shows the source. ³ With a permanent redirect, indexing uses the redirect as a signal that the target is the version that represents the page. This is called the canonical URL. ³

The choice comes down to one question: will you undo this? Google recommends the permanent kind when you are sure it will not be reverted. ³ A service that is temporarily down calls for the temporary kind, and the original address stays in search. ³

301, 302, and 308: what changes for Google

There is an idea going around the industry that a 301 passes “almost all” of the authority and a 302 passes none. Google’s documentation describes something else. On 301s and the other permanent redirects, the migration guide is direct: they do not cause a loss in PageRank. ⁴

The difference between the two families lies in which address becomes canonical. With a temporary redirect, the crawler follows it, but indexing does not use it as a canonical signal. The target can still be indexed, if other signals point to it. ³

The two families, according to Google’s table: ^{3 5}

1. Permanent. 301, 308, zero-second meta refresh, and JavaScript redirects.
2. Temporary. 302, 303, 307, and meta refresh with a delay longer than zero.

The 308 confuses a lot of people, because it is the youngest of the bunch. For Google’s crawlers, a 308 is equivalent to a 301, and a 307 is equivalent to a 302. ⁵ The difference matters to other software. Under the HTTP specification, a client that receives a 301 may turn a form submission into a plain page request. The 308 does not allow that switch. ² Google asks for the right code in each case, because other clients read the difference. ⁵

On our site, the two live side by side. Vercel responds with a 308 when a rule is marked as permanent. ⁶ We declared 301 for our address changes, because that was the code WordPress already returned. ¹ For Google, it makes no difference. ⁵

Methods outside the server call for care. An instant meta refresh, written inside the page, counts as permanent. A delayed one counts as temporary. ³ JavaScript comes last, for a mechanical reason. Google only sees that redirect after rendering the page, and rendering can fail. ³

All permanent methods have the same effect on search. What changes is how long search engines take to notice each method. ⁷ So whenever you can, keep the redirect on the server.

The five places our redirects failed silently

All five share the same shape. The rule was right for the address someone wrote in the table. It failed on the address that a visitor, the crawler, or the platform itself actually produced.

The trailing slash the platform adds before the rule

Picture an apartment building with a list of former tenants posted at the front desk. Before checking the list, the clerk rewrites what you say in the building’s format. You say “twelve,” and the clerk writes down “Apartment 12.” The list said “twelve.” The clerk finds nobody and sends you away.

The new site uses addresses that end in a slash. Vercel enforces this with a setting that returns a 308 to the slash version. ⁶ This is called URL normalization. The order is something we found by measuring: the platform normalizes the slash before it checks the redirects. ¹

On September 2, with the domain already pointed, old addresses in the /br/ folder with no pair returned a 404. The rule meant to catch all of them did not match paths ending in a slash. The address the clerk corrected slipped past the list. ¹ The fix was a second rule, written in the slash form from the start. ¹

I measured again on September 14. An old address without a slash goes through two hops. First a 308, which only adds the slash. Then the 301, which leads to the new page. ⁸ It works, and that is why every rule in our table is now written with the slash from birth.

The www that served the whole site

On September 3, a production audit found a second site live. The address www.seogenome.com redirected the homepage. Every other page responded with a 200, with the full content. ¹

It was the same slash defect. The www rule covered paths without a slash, and the whole site uses the slash. During that time, the only defense was the canonical tag, pointing to the right address. ¹

Google treats redirects and canonical tags as strong signals, and says they stack. ⁷ During those days, our site was relying on only one of them. Today the www returns a 308 at every level. ^{1 8} How much the crawler spends on each host shows up in the crawl stats report.

The uppercase the old server forgave

WordPress served the short /go/ links in any letter case. Vercel compares the address letter by letter. ¹ Its documentation says the rule is case-sensitive, precisely so you do not create duplicate content. ⁶

Google Analytics recorded real clicks on /GO/DNA-GPT/, /GO/GOOGLE-MUDOU/, and /GO/SEO-LOCAL-GMB-E-PLEPER/. ¹ It looks like someone typing on a phone with caps lock on. At the switchover, those clicks started landing on a 404. On September 2, every /go/ link got an uppercase variant, with the same destination. ¹

There is a limit here, and I measured it on September 14. The mixed-case spelling, /Go/Dna-Gpt/, still returns a 404. ⁸ The fix covers the two forms that showed up in the data. Every other combination is still left out.

The pages only Google’s list knew about

This is the one from the opening. On September 2, we audited the redirects against two lists. The first was our table, with 57 pairs, and every one of them reached a live page. ¹

The second was the list of addresses Search Console knows, over a 16-month window. It had 37. ¹ It found five pages returning a 404, with 437 impressions between them. All of them sat outside the folder the general rule covered. ¹

There were four author pages, in two spellings and two prefixes, plus the sample post that comes installed with WordPress. The author pages went to the author’s page, because someone looking for the author wants that page. The sample post went to the homepage. ¹

Two days later, the crawl report brought up another kind of leftover. Google had climbed one level in an article series and hit the parent folder, which never existed. That folder got a redirect to the main piece of the series. ¹ Four requests for a file Apple devices check, to see whether a site has an app, stayed at 404 on purpose. For anyone asking for a file that does not exist, a 404 is the correct answer. ¹

The links nobody published

The last finding is still open. Over two years, Analytics saw 99 different /go/ paths. Of those, 94 ended in a 404, for a total of 181 views. ¹

None of them came from our link shortener. Judging by the pattern, these are people typing by hand what they heard in a talk or saw in a bio. Most of them belong to the DNA of a Prompt series. ¹

One case we already handled shows the mechanism. A space pasted into a bio sent 14 views to an error page in 90 days. Both spellings with the space now lead to the right article. ¹ The 94 paths are still waiting for a decision on where they should go. ¹

How to set up your redirects

The playbook for a migration, with the URL map and the owner of each step, is in SEO for companies. Here you get the rules for the redirect itself, and the places where our own site still departs from them.

Straight to the final destination. Imagine a letter that passes through four post offices before it arrives. Each handoff is a hop, and the sequence is called a redirect chain. Google follows up to 10 hops. ⁵ The migration guide asks you to go straight to the final destination. When you cannot, keep the chain short: ideally no more than 3 hops, and fewer than 5. ⁴ A long chain slows the visitor down, and not every browser can handle it. ⁴

I measured our worst case. Anyone who types an old address with http, with www, and without a slash goes through 4 hops before reaching the page. ⁸ That is under 5 and above the ideal. The first three only fix the shape of the address: the protocol, the slash, and the host.

Each address with its own pair. Sending many old URLs to the homepage confuses visitors, and Google may treat it as a soft 404. ⁴ The guide gives the exception: content from several pages merged into one can receive all of them. ⁴ A deleted page with no equivalent should return a 404 or a 410. ⁴ Over time, Google stops using a URL that has started returning 404 or 410. ⁵

Our house rule for the link shortener points the same way. Every /go/ link has to work and point where it used to point. An article that moved goes to its new address. ¹

I am declaring one place where we depart from the guide. The general rule for the /br/ folder sends everything left without a pair to the homepage. It sits last in the table, after every pair with a destination of its own. ¹ Even so, it is the design the guide advises against. I am putting it on record so you can weigh it in your own migration.

On the server, whenever you can. Google ranks the methods by how likely it is to interpret each one correctly, and the server-side redirect comes first. ³ Platforms like Blogger and Shopify may already come with a built-in solution. ³

The test that catches what the table misses

Testing the table proves the table works. You still have to test what people and the crawler actually request. There are two fronts.

The spellings. Pick ten important old addresses. Run each one in five forms: with a slash, without a slash, in uppercase, with www, and with http. The result has to be a 301 or 308 leading to a page that returns a 200. The migration guide suggests command-line tools and mentions Screaming Frog for crawling your own site. ⁴

The lists. Your table is the first one. The second comes from Search Console: the pages that had impressions over the last 16 months. The third comes from Analytics: the visits that ended on an error page. The second list is the one that found our 437 impressions. And the third is the one that found the 94 paths. ¹

In the Search Console Page indexing report, two labels call for different readings. “Page with redirect” is expected. The old URL redirects, and so it stays out of the index. ⁹ “Redirect error” calls for action. It shows up with a chain that is too long or a loop. It also shows up with a URL over the maximum length, or a bad or empty URL along the way. ⁹

The migration guide records a mistake Google sees often. It is a redirect pointing to a URL that does not exist on the new site. ⁴ The five-form test catches that one too.

The question you ask from now on

From now on, when someone tells you the redirects were tested, you will ask against which list. And in which spellings.

The checked table shows the promises that were kept. The failures live in the addresses nobody wrote into it.

The first step fits into today. Open Search Console, export the pages from the last 16 months, and run the ten most important ones in the five forms. If a 404 shows up, you have found your first leftover. That test belongs in any SEO audit, along with the other fronts of the technical SEO section.