URL migration: the redirect plan that avoids the collapse

Start from the real list, not the planned tree

The list of pages to handle does not come from the plan of the future site: it comes from what exists and receives traffic. Cross three sources — the current sitemap, the server logs and the search data — to avoid missing the pages nobody documented.

Server logs reveal what the other sources ignore: old campaigns, forgotten inbound links, addresses with parameters, PDF files indexed for years.

Four rules are enough

1. One permanent redirect per address, to the final destination — never a chain.
2. A destination equivalent in content; the home page is an admission of failure, not a fallback.
3. An explicit fate for what disappears: a useful redirect, or a deliberate 410.
4. No redirect towards a page that is itself redirected in another system.

Test before, not after

The mapping file is tested on staging, address by address, with an automated check: expected status code, expected destination, number of hops. That check takes an hour to write and saves weeks of reconstruction.

Watch the right window

The first signals arrive within days in the server logs — spikes of 404s — and within two to four weeks in the search data. Do not conclude too early: a dip in the first week is normal, a persistent dip after a month is not.

Keep the redirects for a long time. They cost almost nothing and keep recovering inbound links years later. Removing them “to tidy up” is one of the rare decisions whose price you pay without ever making the connection.

The particular case of multilingual sites

When addresses change in several languages, each version must redirect to its language equivalent, never to the default language. The alternate declarations must be updated in the same move, otherwise two competing versions of each page end up coexisting.

A redirect is not a technical operation: it is the promise that a link made five years ago still works today.