Google App Ads Bot Installs: $220 Spent, 60% Robots
Nick Abe spent CA$220 on a Google app-ads campaign for his puzzle app Dayzle and did what almost nobody does: he opened the raw analytics export and counted. The September 11 writeup is the cleanest first-party ad-fraud dataset I've seen this year. Over two weeks Google billed 56 installs. 33 match a bot signature — side-loaded from a saved APK copy (old versions the Play Store had stopped serving days earlier), one open, zero seconds on any screen, never seen again, across 28 phone models in 19 states. 7 more came from countries the campaign didn't target. That leaves 13 humans, who finished 92 games between them.
The number that matters
Do the division the ad platform won't do for you: CA$220 ÷ 13 real users ≈ $17 per actual human, against a $1.50 cost-per-install target — an 11x miss invisible in Google's own dashboard, which happily reported "21 installs" on the single worst day. The perverse part is the feedback loop. The bots watch the shortest ad video, skip the click, and install from a cached APK because it's faster. Google counts view-then-install as a conversion — so the bot farm's efficiency improved the campaign score, and the algorithm funneled more spend into the fraud. The optimizer did its job perfectly. The signal it optimized on was fake.
Why it matters
The thread under the post has a decade-long small advertiser describing the bot share "steadily increasing" with no reliable way to report it — this isn't one bad campaign, it's the measurement layer. Two days ago the same pattern surfaced on the organic side, with Google rewriting result links to an undecodable /goto redirect to raise scraping costs. Google is spending real engineering hardening its surfaces against bots while the billing surface pays them. When the platform is also the auditor, and the platform's revenue scales with reported conversions, the honest number only ever comes from your own server logs. Abe's fix is the template: optimize on an in-app event (a finished game) no bot farm can fake cheaply. His refund request is still pending.
The verdict: an 11x gap between billed and real cost is not a targeting problem — it's the business model working as designed. If you run install campaigns, your analytics export is the only ad audit that doesn't answer to the ad network.