The lesson
- Compare the same queries against themselves, then split the drop by page; a site-wide story is often one URL
The short answer
When Search Console impressions fall, there are three common causes, and only one of them is bad news:
- You lost rankings. Real, and worth acting on.
- Google stopped showing you for queries you ranked badly for. Impressions fall, but those were impressions on page five that nobody clicked.
- The data is incomplete. The newest days are not final yet (see the Search Console data delay), or the dimension you grouped by hides some rows.
The site total cannot tell these apart. A site that lost its worst queries and a site that genuinely improved both show "average position improved". You need two checks: compare the same queries against themselves, then split the drop by page.
How this caught us
In early September 2026 this site's weekly impressions fell hard enough that our first write-up called it a collapse. Here is what the two checks found, measured on 7 September 2026.
Check one: the paired comparison. We took the queries that appeared in both weeks and compared only those. There were 66 of them. Their impressions went from 102.8 to 110.0 per day, up 7%. Their median position improved by 2.1 places, with 38 queries up and 18 down. The queries we still ranked for were doing better, not worse.
Check two: split named from unnamed. Grouping by date counts every
impression; grouping by query counts only the rows Google will name. The
difference between the two is a third series. Named impressions went from
143.4 to 124.4 per day (down 13%), while the unnamed remainder went from
106.7 to 12.6 (down 88%). Most of the scary drop was in impressions we
could never see a query for.
Then the page split, which overturned our own explanation. We had decided the whole thing was long-tail noise falling away. Testing that on the page that fell hardest proved it wrong. Our Interaction to Next Paint explainer had been earning named queries at positions 20 to 54, not junk. All 22 of them were variations on one topic, and all 22 went to zero. That one URL was 17.6 of the 19.0 named impressions per day we lost: 93% of the visible drop. The rest of the site was flat.
How to run the checks yourself
- Pick complete windows. End both on a date the data actually contains, and compare per-day rates.
- Pull both windows grouped by
query, with a high row limit. Keep only the queries present in both. Compare their impressions and positions. This is the only view where "position improved" means something. - Pull both windows grouped by
dateand subtract the query totals. The remainder is what the query report cannot name. Report it as a range, not a cause. - Pull both windows grouped by
pageand sort by the loss. If one or two URLs account for most of it, you have a page problem, not a site problem.
Crawl Cove's Search page does step 4 for you: its Declining pages list compares two adjacent 28-day windows and names the pages whose clicks fell by a quarter or more. How the Search page surfaces declining pages explains the rule.
Tip
If one page lost a whole cluster of queries, look at that page before looking at the algorithm. Check it is still indexable, still the canonical, and still the best answer on the site for that topic. The quickest way to settle the first two is to ask Google live from the URL inspector, which returns the coverage state and the canonical Google chose.
The lesson
A page-level demotion and a site-wide change add up to the same total and call for opposite responses. Break the total down before you name a cause, and when the breakdown contradicts your story, believe the breakdown. It is the same reason a technical SEO audit works page by page, and the reason Crawl Cove reports issues per URL rather than as one site score.
Frequently asked questions
- Why did my impressions drop but my position improve?
- Usually because Google stopped showing the site for queries where it ranked poorly. Those low-ranking impressions pulled your average position down, so losing them lifts the average even though nothing got better. Compare the queries present in both periods to see whether rankings really moved.
- How do I know if an impressions drop is a real problem?
- Take the queries that appear in both periods and compare only those. If their impressions and positions held or improved, the drop is in queries you were never going to win. Then break the loss down by page, because it is often concentrated in one URL.
- Is an impressions drop always an algorithm update?
- No. A page losing one topical cluster of queries and a domain-wide change look identical in the site total. Only a per-page breakdown tells them apart, and they call for opposite responses.
- Can the Search Console data delay cause a fake drop?
- Yes. The newest days are not finalised for about two days, and the API leaves them out rather than returning zeros, so a window ending today is shorter than it looks.