Map every old address to a new one before the switch, not after. Amlwch Aggregates, a builders' merchant on Anglesey, moved 1,200 pages from an old shop platform to a new one on 4 May 2026. Traffic fell by about a fifth and recovered in seven weeks, which is a good outcome and was decided almost entirely by work done in April.
A dip is normal and its size is decided by the mapping. Something always falls: addresses change, templates change, and the engine has to re-establish what is where. The question is whether it falls by a fifth for seven weeks or by three quarters for a year, and the difference is a spreadsheet.
What is done before the date
Freeze the old site, crawl it, and keep the crawl. Amlwch stopped publishing on 20 April, crawled the site, and saved the export: 1,200 addresses with their titles, their status codes and their inbound link counts. That file is the only record of what existed, and after the switch there is no way to make another one.
Pre-launch
Four steps completed before the switch date
Every pre-switch step builds the baseline record and map before anything moves, so the migration starts weeks ahead of launch.
- 01Freeze the old siteStop publishing on 20 April so the crawl record stays accurate until switchover.
- 02Crawl and save the exportCapture 1,200 addresses with their titles, status codes, and inbound link counts.
- 03Map destinations and deliberate 404sAssign every URL a target, leaving 61 discontinued products to return 404 intentionally.
- 04Change only the platformKeep templates and copy identical so any post-launch movement has a clear cause.
Changing platform, design, and copy on the same weekend leaves no way to isolate which change caused a traffic drop.
Then the map. Every address gets a destination, and 1,200 addresses mapped, 61 of them deliberately left to 404, which is the row most migrations lack: pages for products discontinued years ago, with nothing sensible to point at, and a deliberate 404 is a better answer than a redirect to a category that does not answer the question.
Change one thing at a time, and a platform and a design are two things. Amlwch kept its templates and its content identical and moved the platform, which is unglamorous and is the reason the recovery can be attributed to anything. A firm that changes platform, design and copy on one weekend has no way to know which of the three caused a fall.
On the day
Check the redirects on the day, from the crawl you kept. Feed the 1,200 old addresses back through a crawler and read the final status of each: anything that is not a 200 after one hop is a fault, and the day of the move is the cheapest time to find it. Amlwch found 34, all from a rule written with the wrong prefix, and fixed them in an hour.
Launch day
Checking all 1,200 old URLs on the day of the switch
Running the saved crawl list back through a crawler catches faulty rules while they are cheapest to fix.
- CutoverSwitch platformActivate the new site with the pre-built redirect map in place.
- TestRecrawl the 1,200 old URLsConfirm every old address reaches a 200 status after a single hop.
- RepairFix prefix faults immediatelyAmlwch caught 34 faults from one wrong prefix and repaired them within an hour.
- RecrawlSubmit the old sitemapPrompt crawlers to fetch and follow every redirect, then remove the file a month later.
Submitting the old sitemap of redirecting URLs speeds up discovery of the new destinations.
Keep the old sitemap and submit it after the move. This is counterintuitive and it works: submitting a file full of addresses that now redirect is the fastest way to get every one of those redirects fetched and followed, and the file can be removed a month later.
Where this stalls: watching the wrong number for a fortnight
Watch impressions rather than traffic for the first fortnight. Traffic after a migration is noisy for reasons that have nothing to do with search, and impressions move first and more honestly. Amlwch's impressions were back to their April level by 22 May while traffic was still down, which is what recovery looks like from the inside and is invisible if only sessions are being watched.
Monitoring
Why impressions recover before sessions do
Judging a migration by noisy day-four session counts leads to panic rollbacks that double the disruption.
- Day 4The rollback temptationTraffic looks low and noisy, tempting teams to revert into a second migration.
- First fortnightWatch impressions, not sessionsImpressions move earlier and show search visibility returning cleanly.
- 22 MayImpressions reach April baselineAmlwch's impressions recovered while session numbers were still lagging behind.
- Week 6Pre-agreed wait thresholdHold steady until the six-week window agreed in advance before touching architecture.
Reverting early turns a seven-week dip into a six-month recovery across two sets of URLs.
The stall is the meeting on day four. Traffic is down, somebody proposes reverting, and reverting is the one action that turns a seven week dip into a six month one, because it means two migrations instead of one and a second set of addresses to re-establish. The number to have agreed in advance is how long the firm will wait before acting, and at Amlwch it was six weeks.
What is checked afterwards
Search Console holds both halves: the page indexing report shows the new addresses being stored, and the performance report shows the impressions returning. Both are free, and it replaced counting pages by hand and believing the total. The line to watch is crawled but not indexed on the new addresses, which should shrink week by week and did.
Verification
What confirms a migration has settled
Search Console tracks storage and visibility while sectioned crawls verify the 1,200 mapped addresses.
- Page indexing reportWatch new URLs get stored and crawled-but-not-indexed shrink week by week.
- Performance reportConfirm impressions return steadily without counting pages by hand.
- Sectioned Screaming Frog checksAudit the 1,200 URLs in smaller batches within the free crawl allowance.
- Sitemap and redirect alignmentEnsure submitted files and redirect rules agree on every live destination.
1,200 mapped URLs, readable paths, and live sitemaps
A migration exercises redirects, URL structure, and sitemaps together on the same weekend.
Screaming Frog SEO Spider does the crawl before and the check after, free up to a few hundred addresses at a time, and it replaced clicking through a site with a notepad. On 1,200 pages that means running it in sections, which is the honest limitation of the free version at this size.
Every rule in the map is moving a page without losing what it earned, the addresses being moved to should have been decided as readable, stable and shallow, and the file submitted afterwards is listing what you want indexed. A migration is the one occasion when all of the layer that decides whether any of the rest counts is exercised on the same weekend.