Google has been using its Gemini AI technology to identify and curb ad fraud and invalid traffic (IVT) across its platforms. The tech giant has been quietly testing the system for over a year and a half, pairing AI tools with machine learning to detect and remove deceptive or disruptive advertising activity.
As per media reports, Google’s ad traffic quality division has worked with Google Research and Google DeepMind to deploy multimodal large language models capable of navigating websites and apps like a human user, observing ads, taking screenshots, and detecting hidden or misleading elements. This AI-driven approach is supplemented by machine learning and traditional backend analysis to flag accidental clicks, invisible ads, and pop-ups that force user interaction.
A pilot program run between December 2023 and October 2024 resulted in a 40% reduction in mobile IVT stemming from “deceptive or disruptive” ads, Google reportedly said. While the process is not yet fully automated, the pilot also improved enforcement speed and accuracy.
Per Bjorke, Google’s director of product management for ad traffic quality, was quoted in media reports saying: “If you [as an advertiser] spend $100,000, and $10,000 goes to invalid traffic, that’s $10,000 wasted that you could have spent on actually acquiring users. It’s also a problem for good publishers because any dollar paid to bad actors is a dollar that should have gone to a good publisher. It puts the whole advertising ecosystem at risk.”
Bjorke also pointed out the impact on end users, citing an incident where a pop-up ad interrupted a user’s attempt to dial 911. He added that certain ad fraud schemes involve malware installations on devices, creating further risks.
Google has already dismantled multiple large-scale ad fraud operations this year, including removing 352 Android apps linked to a scheme generating 1.5 billion daily bid requests, and nearly 200 apps from a separate botnet operation in February. In July, the company filed a lawsuit against a group of China-based hackers accused of infecting over 10 million Android devices with malware.
Commenting on the AI’s evolving role, Bjorke said, as per media reports: “This is always an adversarial kind of ‘tit for tat’ type of game. We make advancements. The bad actors change their behavior. We are making more advancements. They are changing their behavior.”
He further noted that while the system is not “foolproof,” AI can react more quickly than traditional data analysis, making it “more robust in these types of use cases.”
Last week, Google’s ad safety team reportedly claimed in a blog post that it had achieved a 10,000-fold reduction in the AI training data required to detect policy-violating ads, signalling growing efficiency in tackling ad fraud.
