Most SEO advice is written about other people's sites. This is about ours. We took one of our own sites and deleted 85% of it — from 38,745 pages down to 5,751 — in a single change. Here is why we did it, what Google did next, and what we would do differently. It is an honest case study, including the part that went wrong.
Why we cut it
The site had a crawl-budget problem. We had published over 38,000 URLs, and Cloudflare's edge data showed Google was crawling only about 20 new URLs per day. At that rate it would take roughly five years just to discover the whole site. Meanwhile, of the salary pages that had been shown in Search Console, most ranked well — so the pages were not bad, there were simply far too many of them for the domain's authority. The inventory was the bottleneck.
How we did it
We did not cut randomly. We measured demand for every page (search volume per topic), kept the pages with genuine demand plus any that already had proven Search Console impressions, and removed the rest. The kept set: 5,751 pages. The removed pages were taken out of the sitemaps and served a 410 Gone status — which tells Google to drop them, as opposed to a 404 which invites re-crawling.
What Google did next
Two things happened, one expected and one not.
Expected: Google slowly digested the removal. The 30,000-plus removed URLs moved into "Discovered – currently not indexed" and then dropped out over the following weeks. Crawl attention concentrated on the remaining pages.
Not expected: site-wide impressions collapsed shortly after the cut — falling by around 99% for a period. Here is the subtle part: the surviving pages stayed indexed (URL Inspection confirmed it, positions on the residual impressions were normal) but stopped being served. That distinction — indexed but not served — is the crux. This was not de-indexing; it looked like an algorithmic suppression, and the timing pointed at the scale of the cut itself.
The pages stayed in the index. They just stopped showing up. "Indexed" and "served" are not the same thing.
What we learned
- Prune in steps, not all at once. An 85% cut in one move is a shock to how Google evaluates a site. Cutting 10–20% at a time, watching the response between steps, is far safer.
- noindex before 410. Signalling removal gradually (noindex first, then remove) is gentler than a hard mass-410.
- Indexed is not served. Watching only your indexed count would have missed the real problem entirely. You have to measure serving — impressions — separately.
- Pruning fixes crawl budget, not authority. Fewer pages helped Google reach the good ones, but the underlying ceiling was the domain's authority. Pruning alone does not build that.
Why we are telling you this
Because we measured all of it — crawl rate at the edge, index status per URL, impressions day by day — we could see exactly what happened and why. A site owner watching only the indexed count would have concluded everything was fine while their traffic quietly vanished. That is the whole argument for measuring properly instead of guessing. If you want that kind of measurement on your own site, that is what we do.
