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

On Wednesday, Google suspended 39.2 million ad accounts on its platform in 2024, more than triple the number of times its previous year, with the latest crackdown on advertising fraud.

By leveraging the large-scale language model (LLM) and using signals such as business spoofing and illegal payment details, search giants said they can suspend “majority” advertising accounts before serving their ads.

Last year, Google launched over 50 LLM enhancements to enhance safety enforcement mechanisms across all platforms.

“These AI models are extremely important to us and have brought about a series of impressive improvements, but humans are involved in the entire process,” said Alex Rodriguez, general manager of ad safety at Google, in a virtual media roundtable.

The executive told reporters that a team of more than 100 experts gathered at Google, including the ADS Safety team, The Trust and Safety Division and researchers from the DeepMind. They analyzed deepfark ad fraud, including spoofing public figures, and developed countermeasures.

The company introduced technical measures last year and over 30 advertisements and publisher policy updates. The moves helped to suspend more than 700,000 problematic ad accounts, and the company claims it has dropped 90% in deepfark ad reports.

In the US alone, Google suspended 39.2 million advertiser accounts last year, cutting 1.8 billion ads last year, and said there are significant violations that it is linked to ad network abuse, trademark misuse, healthcare claims, personalized ads and false dependency.

India, the world’s most populous country and the second-largest internet market after China, saw 2.9 million account suspensions last year, Google said, with the US removing 247.4 million ads in India as well as five policy violations related to financial services, trademark misuse, ad network abuse and personalized ads and games.

Among all advertiser account suspensions, Google said it has suspended 5 million accounts for fraud-related violations.

Overall, the company has removed almost 5 billion ads related to fraud.

Google also saw more than 8,900 new election advertisers in 2024, with half of the world’s population taking part in the polls and removing 10.7 million election ads. However, Rodriguez noted that the amount of election ads compared to Google’s overall ad count is relatively small and has no significant impact on this year’s safety metrics.

In total, Google said it blocked 5.1 billion ads last year and removed 1.3 billion pages. In comparison, It’s blocked It took action against 2.1 billion publisher pages in 2023, exceeding 5.5 billion ads.

Google told TechCrunch that the decline in numbers showed an improvement in its prevention efforts. By improving early detection and suspension of malicious accounts, there will be fewer ads, or fewer ads by reaching the platform, the company said.

The company also said it had limited its 9.1 billion ads last year.

Importantly, large outages can raise concerns about how businesses will apply the rules. Google said it would provide an appeal process that includes human reviews to ensure “appropriate behavior.”

“In many cases, some of our messages were a little more confusing to advertisers, not as clear and transparent about details, about the rationale, or reasoning, and we ended up updating a bunch of policies related to it, Rodriguez said.