Most redesign case studies are written after the problem is solved, when the story is neat. This one is being written in the middle, with the data we have and the test we're running. We'll publish the result either way.
What happened
The site is a B2B service business with about 390 pages: service pages, location pages, a blog and project write-ups. Over six weeks it went through two big changes:
- September 30: the site moved from WordPress to a new custom build. Same URLs, new design, new navigation.
- October 21: 47 service pages were rewritten to put the answer first, 43 pages had their dates updated, and a new blog post went out every day from October 22.
What the data shows
We tracked the same 40 pages, US searches only, in Google Search Console. Average position, weighted by impressions:
| Period | Average position |
|---|---|
| September 8 to 29 (before the move) | 21.2 |
| October 8 to 21 (after the move) | 36.3 |
| October 23 to November 4 (after the rewrites) | 52.8 |
Some individual pages fell harder:
- A roofing marketing service page: about 11 to 35
- A property management SEO page: 22.8 to 57.6
- A "top companies" list post: 9.9 to 32.7
The roofing page shows the scale. From September 6 to 30 it earned 185 clicks from 12,007 impressions. From October 23 to November 5 it earned 27 clicks from 186 impressions.
What we ruled out
"The page needs more content." The roofing page grew from 720 words (the old WordPress version) to 1,485 at launch to 4,909 today. It kept falling. More words wasn't the fix.
"Google can't find or index it." Search Console says the page is indexed, the canonical is correct, and Google crawled it on November 5.
"A core update hit everyone." We couldn't find a confirmed Google update in that window. That isn't proof there wasn't one, but nothing points to it.
"Another page on the site took over." No other page on the site picked up the queries the roofing page lost.
The best lead: the menu
We compared the old navigation with the new one using four archived copies of the old site from June and July.
- The old menu had 87 links, on every page of the site.
- The new menu has 35.
- 72 pages lost their place in the site-wide navigation, including the roofing page and the main city page.
A link in the main menu appears on every page of the site, so it's one of the strongest internal signals a page can get. Then we grouped pages by what happened to their menu link, comparing September 8 to 29 against October 3 to 20 (pages with at least 30 US impressions):
| Group | Pages | How many fell | Weighted position |
|---|---|---|---|
| Lost their menu link | 15 | 13 | 21.6 to 43.3 |
| Never in the menu | 6 | 2 fell, 3 rose | 19.8 to 15.4 |
| Kept their menu link | 3 | (too few to read) | 23.1 to 33.8 |
Pages that lost their menu link fell hard. Pages that were never in the menu held steady or improved.
Why we're not calling it proven
Two honest problems with this evidence:
- The groups overlap with other changes. The pages that lost menu links were also the main service pages, and those were rewritten at launch. The pages that were never in the menu were mostly older blog posts that kept their words. So "lost a menu link" and "got rewritten" are partly the same group.
- The rewrites made it worse. The property management page kept its menu link and recovered to about position 25 by October 12. It then fell to 59 after the October 21 rewrites. And the "top companies" post was never in the menu, held positions 5 to 9 through October 8, then fell to 34. So the menu doesn't explain everything.
What we can say: two site-wide changes three weeks apart, and the position drops track both. Losing navigation links is the strongest single pattern we can find.
What we're doing now
We're avoiding a third site-wide change until we know more. Here's the plan:
- Stopped publishing. The daily posts are paused, so new pages aren't diluting the signal while we test.
- Menu test. The roofing page is back in the main menu and footer. The main city page, which also lost its menu link, is staying out for now as a comparison.
- Trimmed low-value pages from the index. 117 thin location and project pages were set to noindex (not deleted), cutting the sitemap from 390 to 273 URLs. They can come back if the data says so.
- Waiting. No more SEO content or page changes until a review in mid-February. Rankings take weeks to settle, and changing things every week makes it impossible to know what worked.
What other site owners can take from this
Even before the test finishes:
- Save your old navigation before a redesign. Export the old menu, footer and sidebar links. The Wayback Machine saved us here, but it's luck whether it captured your site.
- Count menu links, old vs new. A cleaner, shorter menu looks better and can quietly take dozens of pages out of the strongest internal link spot on the site.
- Space out big changes. A redesign and a site-wide rewrite three weeks apart means you can't tell which one hurt you.
- Track the same pages over time. Site-wide averages hide what's happening. Pick a fixed set of pages and follow them.
We'll publish the result of the menu test once there's enough data to read it, whatever it shows.
