How to build a redirect map that actually holds

Every post-migration recovery project I have seen starts the same way. Someone exports the top pages from Search Console, discovers that forty of them now 404, and asks how a redirect map that was signed off could have missed them.

It missed them because of how it was built.

The traffic export problem

The default method is to export the top pages by clicks, map those, and write pattern rules for everything else. It is fast, it feels comprehensive, and it systematically misses the URLs that matter most for the site’s overall authority.

Here is why. The pages that earn organic clicks and the pages that have accumulated external links are two different sets with partial overlap.

A conference talk write-up from 2019 might get thirty organic visits a year and have links from two hundred domains. It will not appear anywhere near a top-pages export. It is doing real work for your domain, and a pattern rule built around your current content taxonomy will not have anticipated it.

The fix is to build the source list from three inputs and take the union: the Search Console pages export, a full crawl, and a referring-domains export. Then rank twice — once by clicks, once by referring domains — and treat the top of each list as tier one.

Pattern rules first, so the exceptions become countable

The instinct is to start mapping URLs individually. Don’t. Start with the rules.

Write the pattern rules that handle the bulk — /blog/2019/slug to /notes/slug, /category/x to /shop/x. Then run them against the actual URL inventory and record two numbers: how many URLs matched, and how many did not.

That second number is the real scope of the project, and it is now a countable list rather than an anxiety. On a typical mid-sized site, rules handle 85–90% and the remaining 10–15% is where the whole engagement lives.

Critically: test the rules against the inventory, not against the pattern. Every rule has exceptions. The URLs created before the current template existed, the manual one-offs someone made in the CMS, the legacy paths from two platforms ago. They only appear when you run the rule against real data.

The exception rows are content decisions

What is left after the rules is not a technical problem. It is a set of business questions wearing technical clothing.

Where should a discontinued product go — the closest replacement, or the category? Which of two overlapping guides survives a consolidation? Does a campaign landing page from 2021 with forty referring domains get redirected to the current equivalent, or is there no current equivalent?

None of these have a correct answer that a crawler can determine. They need someone who knows the business, and they need a deadline.

This is why the map has an owner column. An exception row without a named owner is a row that will be resolved at 6pm on the Thursday before launch by whoever is still online, and it will be resolved with a catch-all.

Testing is per URL, not per rule

“The rule covers that directory” is not evidence. A rule can be correct and still produce the wrong result for a specific URL, because another rule runs first, or because the URL has a trailing slash the rule does not anticipate, or because the destination itself redirects.

Test tier-one URLs individually against the target environment and record two things per URL: the final status code and the hop count. Anything that is not a single hop to a 200 is a finding.

Then do it again against production immediately after cutover. Staging and production edge configurations differ more often than anyone expects, and the whole point of the exercise is to catch that before Google does.

The verified column

Add a date column recording when each row was last tested. It seems bureaucratic until week three post-launch, when someone asks whether a particular section was ever checked and the honest answer is that nobody remembers.

It also surfaces a real risk: releases after launch introduce redirect chains. A rule added in week three often runs before the migration rules and quietly adds a hop to everything. Re-testing tier one at the 30-day mark catches it. The date column is what tells you it is time.

---

The full sequence this fits into is on the website migration SEO services page, and the phase gates are set out in the migration checklist. If you want to check your own coverage before doing anything else, the diagnostic on the homepage is the short version of this method.