Search Console has a report that makes growing sites look like they are failing. Impressions jump, clicks rise, and the click-through rate collapses. Somebody sees the red arrow on CTR, blames the titles, and changes twenty of them. This note is the arithmetic behind that arrow, on a real site, and the split that shows what happened.
What did the numbers do?
Answer. Clicks rose 31% and impressions rose 338% in the same 28 days, so the ratio between them fell by two thirds.
Nothing on the site got worse. CTR is clicks divided by impressions, and the site started appearing for four times as many searches while collecting a third more clicks. The new impressions were queries the site had never ranked for, mostly at positions 8 to 20, where almost nobody clicks anything.
Why does the site-wide average lie?
Answer. Because it mixes two populations with very different click rates: people searching for your name, and people searching for what you sell.
Brand queries (someone typing "omniroom") click through at 40% or more; the searcher already wants you. Non-brand queries at position 11 click through at 1% or 2%. A site that was almost all brand traffic has a high average CTR by definition. When it starts ranking for non-brand queries, the average must fall, even if every individual query keeps its normal rate.
- clicks
- 77, was 91
- impressions
- roughly stable
- click-through
- high, as always
- clicks
- 11, was 1
- impressions
- 895, was 191
- click-through
- low, as always at positions 8 to 20
Read separately, the story is clear: brand demand softened slightly (77 from 91 clicks), and non-brand visibility went from nothing to something (895 impressions, 11 clicks). The 18% to 6% average is those two facts added together and divided, which is to say, noise.
Where does this go wrong in practice?
Answer. Someone treats the average as a title problem and rewrites titles that were working.
We watched a version of this on another site: a CTR "drop" was diagnosed as weak titles, and a batch of title changes went out. The pages whose titles changed ranked higher and were seen more; clicks did not follow, because the queries that had been added were informational ones with an AI Overview on top. The title was never the constraint. The measurement window for those changes was then contaminated for a month.
Two habits prevent it. Split brand from non-brand before reading CTR at all. And when CTR moves, look at the impression count first; if impressions moved more than clicks, the ratio is telling you about the denominator.
What should you watch instead?
Answer. Non-brand clicks, non-brand impressions, and CTR per query for the queries you actually care about.
Non-brand clicks are the number that says whether new people are finding you. Non-brand impressions say whether you are being shown for their questions at all. CTR is useful only per query, compared with what is normal for that position; a query at position 3 with a 1% click rate is a real signal (something above you is absorbing the click), while a site-wide 6% is not a signal of anything.
Moonleap reports the split by default when Search Console is connected, using your project name and domain as the brand terms, and flags per-query anomalies rather than the average.
FAQ
How do I define brand queries in Search Console?
Use a query filter with a regular expression that matches your name and its misspellings, then invert it for non-brand. Moonleap builds this from the project name and the host label automatically.
Should I ever worry about site-wide CTR falling?
When impressions are flat and clicks fall, yes; that is a real loss. When impressions rise faster than clicks, no.
Is a rising average position always good?
Mostly. It can also fall when you start ranking for many new queries at low positions, for the same denominator reason. Read it alongside impressions.
Sources
- Google Search Console, property omniroom.ai, 28 days to 4 September 2026 versus the 28 days before (our own export)
- Google, "Performance report (Search results)", metric definitions https://support.google.com/webmasters/answer/7576553