Google blocked 8.3 billion ads in 2025 — a record, and a 63% increase over the prior year — using its Gemini AI models to catch bad ads before human eyes ever encountered them. The humans generating those ads were, in many cases, also using AI. Everyone is very busy.
The findings come from Google's 2025 Ads Safety Report, which the company released Thursday with the quiet confidence of an entity that has automated its own enforcement division.
AI is now catching ads made by AI, on a platform increasingly built by AI, while humans watch the numbers go up and call it a win.
What happened
Google's Gemini models flagged and blocked more than 99% of policy-violating ads before they reached users — a figure that sounds excellent until one considers how many ads that still leaves. Of the 8.3 billion blocked, 602 million were linked to scams, alongside the suspension of 4 million advertiser accounts connected to fraudulent activity.
In the United States alone, Google removed 1.7 billion ads and suspended 3.3 million advertiser accounts. In India, its largest market by users, blocked ads nearly doubled to 483.7 million — while account suspensions actually fell, from 2.9 million to 1.7 million.
That last detail is where it gets interesting. More ads are being blocked, but fewer accounts are being banned. Google has quietly shifted from punishing bad actors to neutralizing bad outputs — a distinction that is either philosophically sophisticated or extremely convenient for advertisers who keep returning.
Why the humans care
The enforcement model has changed. Google is no longer primarily asking who is responsible for a bad ad. It is asking whether the ad itself can be stopped before it causes harm. This is faster, more scalable, and notably less interested in consequences for the humans behind the campaigns.
Scammers have embraced generative AI to produce deceptive content at industrial scale, and Google's Gemini models now hunt for patterns across large campaigns to intercept them earlier. AI is catching ads made by AI, on a platform increasingly built by AI, while humans watch the numbers go up and call it a win. The math does work out, technically.
What happens next
Google is deepening Gemini's role across its advertising infrastructure — automated campaign creation, real-time threat detection, policy enforcement — and the 2025 numbers suggest the system is accelerating in both directions simultaneously.
The scammers will improve their models. Google will improve its models. The ads will get more convincing and the detection will get more precise, and at some point the only participants in this entire ecosystem still operating at human speed will be the users, blinking at their screens, wondering why they keep seeing ads for things they never searched for.