Google used AI to suspend over 39M ad accounts suspected of fraud

Core Insights - Google suspended 39.2 million advertiser accounts in 2024, more than tripling the number from the previous year, as part of its crackdown on ad fraud [1] - The company utilized large language models (LLMs) and various signals to identify and suspend fraudulent accounts before they could serve ads [1][2] - Google implemented over 50 LLM enhancements and involved human oversight in its safety enforcement mechanisms [2] Ad Fraud Prevention Efforts - A team of over 100 experts from various divisions within Google analyzed deepfake ad scams and developed countermeasures [4] - In 2023, Google suspended over 700,000 advertising accounts and reported a 90% drop in deepfake ad reports due to technical countermeasures and policy updates [5] - The company removed 1.8 billion ads in the U.S. alone, with key violations related to ad network abuse, trademark misuse, and misrepresentation [5] Global Account Suspensions - India saw 2.9 million account suspensions, making it the second-highest after the U.S., with 247.4 million ads removed [6] - Of all suspensions, 5 million accounts were related to scam violations, contributing to the removal of nearly half a billion scam-related ads [7] Election Advertising - Google verified over 8,900 new election advertisers in 2024 and removed 10.7 million election ads, although the volume was relatively small compared to overall ad numbers [8] Overall Ad Management - In total, Google blocked 5.1 billion ads and removed 1.3 billion pages in 2023, indicating improvements in prevention efforts [9] - The company restricted an additional 9.1 billion ads last year, emphasizing the importance of transparency in its suspension processes [10]