My Search Console impressions fell 57% overnight. My rankings didn't move.

On September 10, Google showed my site in search results 527 times. On September 11 it was 293. A week later it was 115, and it has sat around 150 a day ever since.

A chart like that makes you reach for the word "penalty." I run TameTheBot, a free library of copy-paste AI prompts sorted by job, as a side project, and this is my third post here about watching a small site through Google Search Console. The first was about Google indexing 4 of my 122 pages. The second was about that number climbing to 115 while I changed almost nothing.

This one is about a drop that, as far as I can tell, was mostly not real.

What I ruled out first

I started with Google's Search Status Dashboard. No core update started in September 2026. There was a spam update, but it began on September 24, thirteen days after my drop, and the one before that started on August 18 and finished within three days.

Then my own deploy log. That week I shipped four small content edits: one new prompt page and three internal links. No template change, nothing touching robots.txt or canonicals.

Then the index. Indexed pages went from 147 to 157 over the following weeks. Google wasn't throwing pages out.

So nothing on either side explained it.

Comparing ten days before with ten days after

Search Console's Performance report can compare two date ranges. I set September 1 to 10 against September 12 to 21 and skipped the 11th, the day it broke. Impressions went from 4,010 to 1,707. Clicks went from 63 to 28.

Splitting by device changed the picture:

Device	Impressions, Sep 1 to 10	Impressions, Sep 12 to 21	Avg position before	Avg position after
Desktop	3,106	1,217	18.9	17.7
Mobile	893	484	10.7	10.3

Desktop lost 61% of its impressions, and its average position got better.

Average position in Search Console is weighted by impressions. If a page shows up at position 8 for one query and at position 80 for another, both feed into the average. Lose the position-80 impressions and the average improves, even though nothing you rank for actually moved. The country split said the same thing: US impressions fell 68% (2,061 to 668) while the US average position barely changed, 24.0 to 24.4.

The queries that vanished

Sorting queries by how many impressions they lost made it concrete. Of the 20 biggest losers, 16 already ranked below position 20 before the drop. That's page three or deeper. A few of them:

Query	Impressions before → after	Position before
sales outreach ai prompts	17 → 6	83
nurse chatbot	15 → 0	73
hr prompt library	13 → 0	61
ai prompts for territory planning	12 → 2	59

Very few people scroll to position 83. Some of the queries don't read like a person typed them either. "Evaluate the prompt expansion company bluedot on real estate activities management" earned 11 impressions in the first window and 1 in the second.

My guess is that something which used to load deep result pages (rank trackers, scrapers, AI agents) stopped being counted around the 11th. I can't prove that. What I can say is that the shape matches September 2025, when Google changed a results parameter that SEO tools relied on and plenty of sites saw desktop impressions fall while average position rose. It's the same month and the same shape a year apart, but I'm not claiming it's the same cause.

The part that was real

Clicks fell too, from 63 to 28, and clicks are people. Two pages account for 13 of the 35 lost clicks:

Page	Avg position before → after	Clicks before → after
/for/real-estate-agents	13.2 → 18.7	8 → 1
/for/teachers	11.3 → 15.4	8 → 2

Those two actually slid. The query "ai prompts for real estate" went from position 19 to 72. Neither page changed that week, so I don't have a clean cause.

At this size a lot of it is noise. My page for college professors kept the same impressions (186, then 185) at about the same position, and its clicks still went from 10 to 3.

What I'd check next time, in order

1. Google's status dashboard and your own deploy log for that day.
2. Impressions next to average position. If impressions fall and position holds or improves, you probably lost deep impressions, not rankings.
3. The device split. Impressions you'll never get a click from tend to pile up on desktop.
4. Lost queries sorted by their old position. Losing impressions at position 50 and below rarely costs you a visitor.
5. Clicks per page, not the sitewide total, to find the pages that really moved.

For three months I'd been reading that impressions line as a measure of progress. A good part of it was counting searches nobody looked at. The 28 clicks were the honest number, and they were always small.

Data: Google Search Console Performance and Page Indexing exports for tamethebot.com, pulled 5 November 2026. Indexing data runs through 23 October.