Google Uses Gemini For Ad Enforcement

Google says it is now using Gemini inside its ad-safety systems, and the company says those tools stopped most violating ads before they ran. (blog.google) Google’s 2025 Ads Safety Report, published April 16, 2026, said the company blocked or removed more than 8.3 billion ads and suspended 24.9 million advertiser accounts in 2025. The report said more than 99% of policy-violating ads were stopped before users saw them. (blog.google) Google said 602 million of those blocked or removed ads and 4 million of the suspended accounts were tied to scams. The company also said Gemini-powered systems helped it review most responsive search ads “instantly” by late 2025 and that it plans to expand that screening to more ad formats in 2026. (blog.google) The shift is part of a broader change in how Google polices ads: large language models are being used less like chatbots and more like pattern detectors that read text, images, account behavior and campaign signals together. Google said its models analyze “hundreds of billions” of signals, including account setup timing, behavioral clues and campaign patterns. (blog.google) Google has been moving in this direction for more than a year. In April 2025, its 2024 Ads Safety Report said artificial intelligence was helping prevent fraudsters from showing ads at all, and in August 2025 Google said large language models were being used to detect invalid ad traffic across apps and websites. (blog.google 1) (blog.google 2) The company says enforcement is not fully automated. Google’s ad-policy pages say it uses a mix of Google artificial intelligence and human evaluation, with specially trained reviewers handling more complex, nuanced or severe cases. (support.google.com) That matters for advertisers because false positives can shut off a business’s ad account. In a November 13, 2025 post, Google said Gemini was also being used every day in account-suspension decisions and that improved systems had reduced incorrect suspensions of legitimate advertisers. (blog.google) Google’s public line is that the same models are being used to catch bad actors faster without sweeping up as many legitimate businesses. The practical result is that Gemini is now embedded in one of Google’s highest-volume moderation pipelines, where the work is scoring risk, blocking scams and sending edge cases to humans. (blog.google) (support.google.com)