A site migration is any change that alters the URLs a site serves, the platform that serves them, or the templates and content model behind them. That includes replatforming, moving domain, restructuring URLs, redesigning templates, consolidating two sites into one, and moving from a page-based CMS to a component-based one. Search engines experience all of these as the same event: the thing they had indexed is no longer where or what it was.
This is the largest controllable SEO risk most companies take. Not the largest risk (an algorithm update is larger and nobody schedules it). It’s the largest one you choose, put in a roadmap, and give a launch date.
It is also the one where the incentives are worst, because a migration that goes perfectly often produces no evidence that anyone did anything.
What counts as a migration
The definition matters because projects get de-risked according to what they are called, and the dangerous ones are rarely called migrations.
- Platform. Magento to Shopify Plus, a bespoke stack to Salesforce Commerce Cloud, WordPress to a headless Next.js front end. URL patterns, rendering behaviour and internal linking all change at once.
- Domain. A rebrand, an acquisition consolidating two brands, or a move from a subdomain to a subfolder. Conceptually the simplest and operationally the least forgiving.
- URL structure. Flattening a deep hierarchy, adding or removing a language directory, changing product URL patterns. No visible change to users, a total change to every indexed address.
- Template or design. The one that gets skipped. A redesign that removes a sidebar removes thousands of internal links, and internal linking is how importance is distributed across a large site.
- Content model. Moving from long pages to modular components, or from an editorial CMS to a structured product catalogue. Same URLs, materially different content depth per page.
If a project changes any of those five, it needs migration discipline, whatever the Jira epic is named.
The asymmetry
A migration executed flawlessly holds your organic performance flat and delivers whatever business benefit motivated it: a faster stack, a cheaper licence, a brand that finally matches the company, etc. Organic search does not improve because of a migration; it improves afterwards, because the new platform lets you do things the old one blocked.
A migration executed badly takes 20 to 50 per cent of organic revenue away and you spend two or three quarters getting it back, assuming you diagnose it correctly. Some of it does not come back at all, because competitors occupied the positions while you were repairing redirects.
The best possible outcome of a migration is that nothing happens. That is the entire case for taking it seriously, and it is a hard case to make to a steering committee.
I have watched this asymmetry produce a predictable failure: SEO is consulted two weeks before launch, when the URL structure is already built and the redirect map is a line item in QA. At that point the honest advice is to delay the launch, which is the one recommendation nobody has authority to accept.
What done well looks like
A well-run migration is mostly inventory and evidence gathered before anyone writes a redirect rule.
- Build the URL inventory from at least four sources. A full crawl finds what is linked. The Search Console bulk export finds what earns impressions, including pages your crawl missed. Server logs find what Googlebot fetches. Your backlink index finds URLs that are dead internally but still receiving external links. A redirect map built from a crawl alone will miss the orphaned page that has 400 referring domains.
- Map to the closest genuine equivalent, one to one. Not to the homepage, not to a category hub. If there is no equivalent, decide deliberately whether the page should be recreated or retired, and record the decision.
- Redirect with 301s, in a single hop. Chains accumulate across successive migrations until a URL redirects four times; consolidate them before launch rather than after.
- Test on staging with a full crawl. Password-protect it, crawl it, and diff the rendered output against production template by template. Titles, headings, canonicals,
hreflang, structured data, word count, internal link counts per page. - Preserve internal linking density. Compare inbound internal links per URL before and after. This is architectural SEO work, not a content task. A redesign that halves them will suppress performance on pages that redirect perfectly.
- Have a rollback plan and a monitoring window. Daily crawl-and-index checks for the first two weeks, then weekly. Know before launch who can call a rollback and on what evidence.
Concede the trade-off honestly with your stakeholders: this adds weeks to a project timeline, and those weeks are your insurance premium. On a site where organic drives a small share of revenue, paying it in full may not be rational. On a site where organic drives half of new business, skipping it exposes you to significant risk.
When a drop is not a failure
Not every post-migration decline is a mistake. If the migration deliberately retired 60,000 thin pages, impressions should fall and the pages that remain should perform at least as well. If a rebrand moves you off a domain with 20 years of history, some loss of brand-query volume is the price of the rebrand and was agreed by someone above the search team.
The distinction to hold onto is between planned and unplanned change. Agree before launch which metrics may move, in which direction, and within what band. Without that, every wobble in week two becomes a crisis meeting and the genuine problem gets lost in the noise.
What to do first
If a migration is on your roadmap for this year, the first move is not an audit. It is getting a named search owner into the project at the point where URL structure is still a decision rather than a fact. This is usually three to six months before anyone thinks to ask.
If you have already launched and something is wrong, start with the redirect map against the Search Console export from before the change, then the internal link comparison. Those two explain most post-migration losses, and both are still fixable months later.

