A website migration that kept every ranking the old site had earned

Business context

The client is an industrial equipment manufacturer with about 55 staff. It sells pumps and filtration units through distributors and a direct inquiry form. This was a website migration SEO project from the start: the old site ranked well for specific product searches, and those rankings were the asset to protect.

The site itself, though, was the problem. It ran on an old theme, took six seconds to load on a phone, and needed a developer to change a paragraph. The manufacturer had tried a redesign two years earlier and lost half its organic traffic for a quarter, and that history made everyone cautious about touching it again.

The problem

The old site had around 400 pages, many of them ranking. Any move risked breaking the URLs that search engines had indexed. The previous redesign had changed URLs without redirects, so old links returned errors and rankings dropped.

The team also had no inventory of what ranked. They knew a few important product pages, but not the long tail of technical guides and application notes that quietly brought in traffic. Without that list, a rebuild would guess at what to keep.

What we built

We rebuilt the site on Next.js and treated the migration itself as the SEO project. The design work was the easy half; the URL map was the hard half.

The URL inventory

We crawled the old site and combined the results with Google Search Console and Semrush data. Every page got a row: current URL, clicks, impressions, top query, and whether it still had value. That produced a ranked list of pages we could not afford to lose.

The redirect map

For every URL that changed, we wrote a one-to-one redirect to its closest new page. Pages with no new equivalent kept their content and their URL. We deliberately did not send everything to the home page, which is what the previous redesign had done. Where two old pages overlapped in intent, we merged them into one stronger page instead of redirecting both to a thin one. That reduced the page count from around 400 to 340 without losing the traffic those pages earned.

Structured data and speed

Each product page got Product and Organization structured data. Images were compressed, unused scripts removed, and the new pages loaded in under two seconds on a phone.

Rollout

• Weeks 1 to 2: Crawl the old site and build the URL inventory.
• Weeks 3 to 6: Rebuild pages and carry content across, improving the thin ones.
• Weeks 7 to 8: Build and test the redirect map on a staging copy.
• Week 9: Launch, with the old site kept readable for two weeks.
• Week 10: Daily Search Console checks for crawl errors and lost rankings.

Before and after

Item	Before	After
Mobile load time	About 6 seconds	Under 2 seconds
Ranking pages	No inventory	412 mapped by traffic value
Changed URLs	No redirects in the earlier redesign	One-to-one map for every change
Monitoring	None	Daily checks for 30 days

Results

Ninety days after launch, organic clicks were within 3% of the same period before the move. No ranking page was lost. Google indexed the new URLs within two weeks, and 404 errors from the redirect map were close to zero.

The pages that gained most were the technical guides. With the content rewritten and internal links added, they picked up 26% more impressions quarter on quarter. Page speed on mobile went from six seconds to 1.8, which helped the product pages most.

The old site stayed online for a fortnight. We would keep it longer next time.

The discipline is the same on a smaller scale. Our technical SEO work for a local plumbing company also started by protecting what already ranked before changing anything.

Lessons learned

• The redirect map took longer than the design, and it was the part that protected the traffic. We spent three of ten weeks on it. Underestimating that would have cost the whole project.
• Search Console data alone missed pages. Several guides ranked through image search and showed almost no clicks, but they supported product pages through internal links. The crawl found them; the console did not.
• We launched on a Thursday, which left only one working day to react before the weekend. Nothing broke, but a Monday launch would have been the safer choice.