Google’s install optimization got gamed by bots
A bot farm ate most of this $220 install campaign — the fix was making the ad goal harder to fake
Google's own conversion loop rewarded fake installs with more ad spend, until the developer changed what counted as a win
When an ad campaign optimizes for 'install', a bot farm can cheaply fabricate installs and Google's algorithm will reward the farm with more impressions — turning the optimization loop against the advertiser. The only real defense found here was changing the conversion goal to something bots can't cheaply fake ('won a puzzle'), not disputing traffic after the fact.
Two weeks of billed installs, by what they actually were
Source totals 56 billed installs but only breaks down 53 explicitly; the gap between 56 and 53 is unaccounted for in the source.
How the panel and the raw logs disagreed
Google Ads reported 21 installs in a single day; the app's own admin panel logged just 1. The gap wasn't a bug in reporting delay — it was 20 phones running an old Android build that the Play Store had already stopped serving. You can't get that version from Play, yet every device listed Google Play as the installer. Each opened the app once, spent zero seconds on any screen, and never returned — spread across 28 phone models and 19 states, an oddly wide footprint for identical behavior.
The loop that kept getting worse
- 01Campaign goal set to 'installs', target CPI removed
Spend jumped to CA$80/day, more than double the CA$40 budget
- 02Farm watches the shortest ad video without clicking
Cheaper and avoids Play Store detection
- 03Farm installs the app from a saved file, not the store
Faster than a real Play install
- 04Google counts view→install as a valid conversion
No distinction from a genuine install
- 05Algorithm serves more ads to the same farm
It looks like the highest-converting audience
What the 13 real installs actually did
Why this reads as a bot farm, not noise
20 devices in one day running a Play-unavailable app version, all crediting Google Play as installer
Zero screen time, zero return visits, across 28 phone models and 19 states
7 more installs from countries the campaign wasn't even targeting
Ad group's shortest video was watched but never clicked — consistent with farms minimizing detectable interaction
Still open
- 01
Outcome of the invalid-traffic refund claim filed with Google
- 02
Whether the new 'won a puzzle' conversion goal actually reduces farmed installs going forward