What this guide covers
This guide covers the URL and indexing side of a replatform: building the inventory, mapping old paths to new ones, choosing status codes, and verifying the result. It does not cover choosing a platform or migrating payment configuration.
Assumed knowledge
- Access to the current site and its hosting
- Some familiarity with HTTP status codes
- Search Console access, or the ability to request it
Replatforming is the single most common way an established online shop loses revenue overnight. The build is usually fine. What fails is the handover between the old set of addresses and the new one.
This guide is the sequence that avoids that.
Build the URL inventory first
You cannot map what you have not listed. Before any design work, assemble every URL the current site has ever served. Pull from as many sources as you can get:
- The current sitemap, if there is one, and any historical sitemap files still on the server
- A full crawl of the live site
- Search Console performance and page indexing exports
- Server access logs, which reveal pages nothing links to any more
- Analytics landing page reports going back at least twelve months
- A backlink export, so you know which pages other people point at
The output is one row per URL with its current status code, what the page is for, whatever traffic data you have, and a decision about where it should go.
Decide a disposition for every row
Each URL gets exactly one of four outcomes.
| Disposition | When to use it | Status code |
|---|---|---|
| Keep | The page exists on the new platform at the same address | 200 |
| Redirect | The page has a genuine equivalent at a new address | 301 |
| Gone | The page is deliberately removed for good, for example a discontinued line | 410 |
| Not found | The page is removed and you are not certain the removal is permanent | 404 |
The temptation is to redirect everything to the homepage so that nothing appears broken. Resist it. A redirect to an irrelevant page is treated as a soft error, and it is worse than an honest 404 because it also wastes the visitor's time.
Map redirects to the closest real equivalent
A redirect should take someone to the page that answers the need the old page answered.
- A discontinued product goes to its replacement, or failing that its category
- A merged category goes to the surviving category
- A retired blog post goes to the post that supersedes it, if one exists
Where there is no equivalent, do not invent one. Return 410 if the removal is permanent and 404 if you are unsure.
Keep the chain to one hop
If /old-product already redirects to /interim-product, and you now move that to /new-product, update the first redirect to point straight at the final address. Chains waste crawl budget and every additional hop is another chance for something to break.
Check for loops explicitly before launch. A redirect that targets itself, or two that point at each other, will take a page down.
Get the canonical and sitemap behaviour right
Three rules cover most of it.
- Every indexable page declares an absolute canonical URL pointing at itself, on the host you have chosen as canonical.
- The sitemap lists only canonical URLs that return 200 and are indexable. No redirects, no filtered parameter URLs, no removed pages.
- The lastmod value reflects a real change to the page. Touching every lastmod on deployment day tells a search engine nothing and trains it to ignore the signal.
Rehearse the whole thing
A migration plan that has not been executed is a hypothesis.
- Run the full redirect map against the staging environment and record the status code and final destination for every row
- Confirm the canonical host redirects are in place from raw requests, not from a browser that may have cached an earlier redirect
- Restore the backup into an isolated environment and check it actually works
- Agree who makes the decision to roll back, and what the signal is
After the cutover
Expect some movement in rankings for a few weeks. That is normal. What is not normal, and what you should watch for, is a sustained fall in indexed pages or a rise in crawl errors.
Check in the first forty eight hours:
- Canonical host and certificate behaviour from a raw request
- A sample of redirects from each disposition group
- Sitemap retrieval and the reported indexed count
- Server error rates and any spike in 404 responses
Then review weekly for the first month. If something has gone wrong, the evidence is usually in the logs well before it shows up in a ranking report.
Key takeaways
- Build the URL inventory before any design work, from sitemaps, crawls, logs, analytics and backlinks
- Every URL gets one of four outcomes: keep, redirect, gone or not found
- Never redirect everything to the homepage; it is treated as a soft error
- Flatten chains so an old URL reaches its destination in one hop
- Rehearse the full redirect map on staging and record the status code for every row
- Watch indexed page counts and crawl errors for the first month, not rankings
Frequently asked
Questions readers ask most
Will I lose rankings when I replatform?
You can, and it is the most common way an established shop loses revenue overnight. It is also largely avoidable with a signed URL map and tested redirects. Expect some movement for a few weeks; a sustained fall in indexed pages is a different matter and needs investigating.
How long before traffic recovers after a migration?
A few weeks of movement is normal while search engines recrawl and consolidate signals. If coverage has not stabilised after about a month, something in the redirect map or the canonical setup is usually wrong rather than it being a waiting problem.
Should I redirect old pages to the homepage?
No. A redirect to an irrelevant page is treated as a soft error and it wastes the visitor time as well. Redirect to the closest genuine equivalent, and where none exists return 410 if the removal is permanent or 404 if you are not certain.
What if I have no record of my old URLs?
Pull what you can from server logs, analytics landing pages, backlink exports and any archived sitemap. Where records genuinely do not exist, record that as an accepted risk and get it signed off rather than assuming the pages did not matter.



