Remember when the sketchy stuff stayed on Display and YouTube? Limited ad serving, that throttle Google slaps on accounts it doesn’t fully trust yet, has lived over there for years. New accounts, unverified accounts, anyone who hadn’t earned the benefit of the doubt. Search was supposed to be the clean room. The place where the auction was tight, the review was tighter, and you paid a premium for the privilege.
That framing is getting harder to defend. In 2026 Google expanded its ad review process so that user complaints can trigger throttling, not just policy violations detected on Google’s end. Which is a quiet admission worth sitting with: if complaints are now a signal strong enough to slow an advertiser down, the automated review that runs first was missing things people could see with their own eyes.
Why the machine misses
I review automated tools for a living, and this pattern is familiar to the point of boredom. Any classifier ships with a tuning decision baked in. You can catch more bad actors and burn more legitimate ones, or you can protect the legitimate ones and let more bad actors through. There is no third option where the model is simply correct. Every vendor picks a side of that tradeoff and then describes their choice as accuracy.
Google’s own explanation for dodgy ads getting through comes down to two things: gaps in human oversight and reliance on automated systems that miss some violations. That’s honest as far as it goes. But as an ex-QA person, “the automation missed it” is only half an answer. The other half is who was supposed to catch what automation missed, and how many of them there are relative to the volume flowing through.
The atomic14 writeup on this puts it well by borrowing Hanlon’s razor — never attribute to malice that which is adequately explained by stupidity. Then it does the uncharitable thing anyway and asks whether it’s actually in Google’s interest to remove ads that generate revenue. I don’t think there’s a room where someone decides to let scams run. I do think there’s a spreadsheet where false positives cost measurable money today and false negatives cost reputation slowly, in a column nobody owns. Systems drift toward whichever error is cheaper. That’s not a conspiracy, it’s incentive gravity, and it explains the observed behavior better than incompetence does.
The complaint mechanic shifts who does the work
Here’s what bugs me about complaints-as-a-trigger, and it’s the same thing that bugs me about half the AI tools I test. It looks like an improvement to enforcement. Functionally, it moves review labor downstream onto users. You saw the bad ad. You reported it. Your report is now training signal and enforcement input. You are an unpaid reviewer for the largest ad system on earth, and the compensation is that maybe the ad stops running.
For advertisers, the same change reads differently. Complaints are now an attack surface. If throttling can be triggered by user reports rather than a policy finding you can actually contest, your serving volume becomes partly a function of how people feel about your creative. Legitimate businesses in categories that annoy people — debt, insurance, weight anything — inherit a risk they cannot fully control.
Automation is also making the calls you used to make
This is where the toolkit angle matters more than the ad-quality argument. The July 2026 Google Ads terms of service changes authorize automation to do things advertisers used to control. Final URL Expansion is the one to check first. If it’s enabled, Google may send traffic to landing pages you never designated for that campaign, and the new terms authorize exactly that.
Read that alongside the AI Overview eligibility rules and a pattern shows up. By Google’s own rule, exact-match and phrase-match keywords can still trigger ads above or below an AI Overview, but only broad match and keywordless targeting are eligible for placement inside it. Precision targeting gets you the old real estate. The new real estate requires handing the targeting decision to the system.
So advertiser control is being traded for reach, while ad review remains automated with acknowledged gaps. Both sides of the marketplace are being asked to trust the same layer of automation, and that layer is demonstrably imperfect enough that Google added user complaints as a backstop.
What I’d actually do
- Audit Final URL Expansion on every active campaign. If you didn’t turn it on deliberately, turn it off and see what happens to your numbers.
- Treat your match-type mix as a strategic choice about how much control you’re trading, not just a targeting knob.
- Assume complaints affect delivery now, and monitor serving volume as a health metric rather than an afterthought.
- Keep reporting bad ads. The system is using your reports whether or not that feels fair.
The tool works. It just doesn’t work the way the marketing says, and the gap between those two things is where your budget lives.
🕒 Published: