Finding 01
Ranking better cost us the clicks
Three query clusters held page one for a full month and returned 0.17 percent between them. Improving their position made the problem worse, not better.
- Window
- 28 days, 2026-08-13 to 2026-09-09
- Sample
- 3 clusters over 36 pages, 1,393,285 impressions producing 2,374 clicks
- Method
- Google Search Console, page level, one pull. All three clusters are page groups from the same export and the same window, so their impressions can be summed without double counting. Position is impression-weighted across the pages in each cluster.
The assumption underneath most search work is that position and clicks move together. Across a large enough sample they usually do. On individual query families they can move in opposite directions, and when that happens the cause is almost never the page.
What we saw
Three clusters on one property held page one for a full twenty eight day window. Between them they drew 1,393,285 impressions and returned 2,374 clicks, which is 0.17 percent.
| Cluster | Pages | Impressions | CTR | Position |
|---|---|---|---|---|
| Bank branch codes | 29 | 833,763 | 0.215% | 7.94 |
| Festival and observance dates | 5 | 367,983 | 0.068% | 8.61 |
| Calorie lookups | 8 | 191,539 | 0.173% | 5.63 |
One pull, one window, one dimension. Positions are weighted by impressions across the pages in each cluster rather than averaged flat, so a large page cannot be outvoted by a small one.
The correction that produced this table
An earlier version reported these clusters as 662,664 impressions. That figure was wrong twice over. It summed clusters measured in two different windows, and one of the three rows was a search-appearance row rather than a page group, which meant it also counted impressions that were already inside the other two.
Both faults are gone here. Every row is a page group, from one export, over one window, so the total is a real total. The auto-translated surface that used to sit in this table now stands on its own as a separate finding, which is where it belongs.
The mechanism
A query that resolves to one string or one number is answered in the result itself. A bank branch code is a single value. A festival date is a single date. A calorie count is a single number. The searcher has what they came for before a click is possible, and the better the engine gets at surfacing it, the more completely the click disappears.
What this rules out
It rules out the usual remedies. Rewriting the title does not help when the title is not the problem. Nor does improving the page, adding internal links, or chasing position further. All three clusters were already winning on the only axis those tactics move.
It also rules out reading the aggregate. Site-wide over this window the property returned 0.468 percent across 2,055,560 impressions, which looks like a site with a modest click-through rate. It is not. It is a site where some families convert normally and others cannot convert at all, and the average conceals both.
How to check your own
Take Search Console, compare impressions against clicks by query family rather than in aggregate, and hold position constant. The signature is impressions flat or rising while clicks fall and average position does not move. Where you find it, the question to ask is not how to rank better. It is whether the query can produce a visit at all.