Ask
22
@hollis_w ·

Google Ads billed 56 app installs and 13 were real players

A small Android puzzle app ran a Google Ads app campaign at CA$40 a day, goal set to installs, and published the numbers after two weeks and about $220: 56 installs billed, 13 people. Those 13 finished 92 games between them, so the app was not the problem.

Seven of the rest came from countries the campaign never targeted. The other 33 shared one pattern. Each phone reported Google Play as the installer, yet ran a build that Play had stopped serving days earlier, opened the app once, spent zero seconds on any screen and never returned. The developer says 29 of the first 30 involved no click at all: Google credited them to a view of the shortest video in the ad group.

That is the loop. A view followed by an install counts as a conversion, so every fake install tells the bidding that the placement works and sends it more budget. Removing the $1.50 target cost per install is what opened the tap: spend doubled to CA$80 that day, Google reported 21 installs and the app's own panel counted 1.

The same week, two other small developers hit Google Ads from the other side. A signed, notarized macOS terminal app had its ad account suspended for malicious software after $500 of spend, lost every appeal, and was reinstated without explanation once the write-up spread. And a banner imitating an iPhone Storage is Full alert was reported twice and ruled within policy both times.

The puzzle app's changes: goal switched from install to won a puzzle, and an invalid traffic claim is pending.

12 answers Share
Report

Answering anonymously, a moderator will review it first.

  • @jpl_88 · 2d ago · 3 replies

    Before touching the campaign, pull first opens grouped by app version. A copied APK can claim Play as its installer, but it cannot be a build Play no longer serves. That one column said more than any fraud report would have.

    24
    Share
    Reply

    Answering anonymously, a moderator will review it first.

    Report
    • @mkovac · 2d ago · 2 replies

      Only works if you release often. I went four months between builds last year, every fake install would have been on the current version.

      14
      Share
      Reply

      Answering anonymously, a moderator will review it first.

      Report
      • @jpl_88 · 3d ago

        Then first session length. Zero seconds on every screen is not a person. Real people fumble through onboarding, even the ones who uninstall a minute later.

        11
        Share
        Reply

        Answering anonymously, a moderator will review it first.

        Report
  • @beatrizq · 3d ago · 3 replies

    Moving the goal to an in-app event is right, but at 13 real players a fortnight the bidding gets almost nothing to learn from. Expect spend to drop close to zero for a while. That is still cheaper than funding the farm.

    19
    Share
    Reply

    Answering anonymously, a moderator will review it first.

    Report
    • @greta_v · 2d ago · 2 replies

      Pick something between install and winning. Finishing the tutorial, or opening the app again the next day. A script can clear one puzzle easily enough.

      12
      Share
      Reply

      Answering anonymously, a moderator will review it first.

      Report
      • @beatrizq · 2d ago

        Next day open is the one I would trust least, honestly. Reopening an app on a schedule is trivial for a farm. Winning a Sudoku at least needs some logic.

        8
        Share
        Reply

        Answering anonymously, a moderator will review it first.

        Report
  • @mkovac · 2d ago · 2 replies

    Did Google refund anything?

    11
    Share
    Reply

    Answering anonymously, a moderator will review it first.

    Report
    • @hollis_w · 8h ago
      @hollis_w OP ·

      Not yet, the claim was still open when it was written up. The version and session numbers are also the only evidence you have for a claim like that, so export them before the campaign changes anything.

      1
      Share
      Reply

      Answering anonymously, a moderator will review it first.

      Report
  • @lowtide_42 · 2d ago · 2 replies

    Put the data centre IP ranges into the exclusion list and a lot of this goes away. Mine has a few hundred networks in it by now.

    16
    Share
    Reply

    Answering anonymously, a moderator will review it first.

    Report
    • @sunporch · 2d ago

      That caught plenty five years ago. These farms route through residential proxies now, the infected TV boxes and home routers people sell access to. A list of hosting ranges moves the number less every year.

      1
      Share
      Reply

      Answering anonymously, a moderator will review it first.

      Report
  • @greta_v · 4h ago

    The pairing is what gets me. A notarized terminal app gets flagged as malware and loses every appeal, while a fake iOS storage alert passes review twice. The strict part and the lenient part are clearly not the same team.

    1
    Share
    Reply

    Answering anonymously, a moderator will review it first.

    Report
  • @sunporch · 4h ago

    Thirty-three users who never complained, never churned and never asked for a refund. Best cohort that app will ever have.

    1
    Share
    Reply

    Answering anonymously, a moderator will review it first.

    Report