CTR is the trap
Click-through rate on remnant display measures how clicky the inventory was. On our flights that was not the same as whether a person arrived.
Published 2026-09-09 · Search job: why CTR is misleading
CTR is misleading on remnant display because leftover inventory is good at minting billed clicks. The rate is clicks over impressions. It does not ask whether a person arrived, whether the click was invalid, or whether seven hosts produced the whole pile. The June 2026 learning flight was read on clicks and CTR. That readout later had to be thrown out. The 2026-09-02 Supply Bench pull made the same point with grain.
This is the Lab protocol. The essay is CTR is a trap.
What the later pull did to a "healthy" rate
Inside the $316.64 observed slice of campaigns 4791 through 4810:
- Flight average about 0.28 percent CTR.
- Seven domains produced 41.3 percent of clicks from 4.7 percent of impressions.
people.com: 144 impressions, 23 clicks, 15.97 percent CTR.msn.com: 785 impressions, 42 clicks, 5.35 percent CTR. July 2026 had already flagged the same host at about 3.7 percent against a 0.15 percent average.- Reseller 0.423 percent versus Direct 0.184 percent.
- Same creative: OPENX 0.111 percent (44 percent of spend), SMARTADSERVER 3.074 percent, THIRTY_THREE_ACROSS 18.18 percent.
- 1,073 placements, 93.4 percent, had zero clicks and took 64 percent of observed spend.
A blended CTR can sit on 0.28 percent and look ordinary while the click mass is the seven-domain table. Seven domains produced two fifths of the clicks.
What we refuse to compute from CTR
We will not rank SSPs, domains, or ads.txt paths as quality from this rate. A high CTR can be contamination. A low CTR can be ignored inventory. We had no IVT add-on. invalid_impressions of zero is a measurement gap, not a clean bill.
We will not treat CTR as a substitute for click-to-page-view. Anonymized main-seat brand-name display flights sat at 15.85 percent click-to-page-view when usable and 3.75 to 6.68 percent when not. Those are visit ratios, not billed-click ratios. Click-to-page-view benchmarks we trust.
How we read a flight instead
How to judge display traffic is the scoring page: strict IVT once a vendor tag is on, click-to-page-view once the join is honest, concentration in the meantime. CTR can stay on the table as a diagnostic. It cannot be the score we optimize.
If a later flight contradicts a rate on this page, we will edit this file and date the change. We will not freeze 0.28 percent as a category benchmark.