The short answer
Each page should say the same thing to a search engine after the switch as it did before, on the same URL. So capture each page before you change its route, switch the route, and compare the two on the public domain. Move the next batch only when the current one has passed.
This does not guarantee rankings. Google says to expect some temporary movement during a move, and nobody can promise where things land. Parity removes the problems you caused yourself.
The Site Move Playbook has a parity check for a one-off launch in its rebuild chapter. This chapter is what changes when you run that check again and again, with two systems live.
Why stages change the check
- The URL stays, so nothing warns you. In a normal move a redirect report shows what moved. Here the page simply starts coming from somewhere else. If a title goes missing, nothing flags it.
- You check many times. The routine has to be quick and repeatable, or it will be skipped by batch three.
- Old pages can change too. Links to moved pages and shared menus live on the old system, so old pages need checking as well.
- A leak can come from either side. Either system can expose an origin hostname.
Capture before you switch
Take this snapshot from the live public URL, just before the batch. Store it next to the route table, one row per URL.
| Capture | Why it matters here |
|---|---|
| Status code and final URL | A new redirect is the first sign of a slash or path mismatch |
| Title, meta description, H1 | The easiest to lose, and the easiest to compare by script |
| Canonical and robots tags | These tell you what the page is asking for |
| Schema as a block of text | Compare types and URLs, because they are often hard-coded |
| Internal links in the body | Shows if the new page links out to the same places |
| The heading list and word count | A fast way to spot content that was cut |
| Social tags | og:url and og:image often carry the origin hostname |
| Clicks and impressions | From Search Console, for comparing later |
Save the rendered HTML as well as the fields. When someone asks later what the old page said, it will be the only copy you have once the old site is gone.
On one revamp, the pages kept the same URLs, titles, meta, schema and links. That is the list that matters.
The batch check, step by step
Run this on the public domain right after you switch the rules for a batch. Repeat steps 2 to 6 the next day, in case a cache was hiding something.
| # | Check | Pass means |
|---|---|---|
| 1 | Confirm which system answered each URL | All on the system the route table says |
| 2 | Request each URL with redirects turned off | A 200, with no hop |
| 3 | Compare fields with the snapshot | No differences you did not plan |
| 4 | Search page source for origin hostnames | Zero matches |
| 5 | Read the canonical on every page | Absolute, public URL, and points to itself |
| 6 | Crawl the whole public domain | No new broken links or redirect chains |
| 7 | Check both sitemaps | Each URL in one sitemap, on the public domain |
| 8 | Inspect a few URLs in Search Console | Google can fetch the live page, and sees the right canonical |
Canonicals: agree in every place
Google's canonical guide calls redirects and canonical tags strong signals, and sitemap listing a weak one. Signals can stack, but they should not disagree. Some details matter more once you have two systems:
- Use absolute URLs. Google recommends them, and warns that relative ones can cause trouble when a testing site gets crawled. An origin hostname is exactly that case.
- Every page names itself. Google recommends a canonical on the canonical page too.
- Do not use robots.txt to hide an origin. Google says not to use it for canonicalisation, because a blocked URL can still be indexed without its content. A canonical to the public URL is the fix Webflow's checklist points to.
- Do not use noindex to choose between versions of a page. Google advises against it, because it blocks the page from search entirely.
From time to time, run a site: query for your origin hostnames. It is a quick check that nothing has leaked.
Sitemaps during a staged move
Keep one sitemap per system, and list both in an index or in robots.txt. Google's sitemap guide allows several sitemaps and sitemap index files, and says submitting them separately lets you track each one in Search Console. That gives you a simple way to watch moved and unmoved pages separately.
- Full, absolute URLs on the public domain only. Google will try to crawl them exactly as listed.
- Serve them from the public domain. Unless you submit through Search Console, a sitemap affects only what sits below its own folder, so the site root is the safest place.
- In Webflow, the checklist says to use your global canonical URL as the base, or to enable a custom sitemap. It also says to replace the default sitemap address Webflow adds to robots.txt.
- Move each URL from one sitemap to the other in the same batch as its route.
- Keep
lastmodhonest. Google uses it only if it proves accurate, so a moved page with unchanged content should not look freshly rewritten.
Two things that often get missed
Links from unmoved pages. Old pages that link to moved pages should keep working, because the URL is the same. Check for old redirects, though. The old system may have a redirect rule that no longer fires, and that rule may need to exist on the new side. Our redirect map template works well for keeping track of where each rule lives.
Speed on the moved pages. A proxy adds a stop on every request. Compare Core Web Vitals and first response time for moved pages before and after, and look at your cache if they have slowed down.
When to start the next batch
Google suggests watching the Sitemaps and index reports, the search queries for your pages, and your server logs for crawler activity and errors during a move. For a staged move, compare the moved URLs against the group that has not moved. If both move together, the cause is probably not your batch.
Decision rule Start the next batch when the check is clean, and Google has crawled the most important pages in the batch and accepted their canonicals. If the batch shows a real drop that the unmoved group does not, stop and find out why. Never move a new batch to cover a problem with the last one.


