For almost two years, Google has been quietly experimenting with machine learning and AI tools to tackle invalid traffic – ad interactions caused by bots, accidental clicks, or click fraud rather than genuine user interest – according to statements given to Adweek. Since late 2023. these efforts have significantly reduced IVT.
The company said Google’s ad traffic quality team is collaborating with Google Research and Google DeepMind to use multi-modal large language models that simulate human interaction with websites and mobile apps. These models are able to determine all functions of the app, and interact with it as well – whether it is through navigation tools, screenshotting or observing ads. A very important function they have is identifying deceptive or disruptive ads, as well as identifying when ad buttons should be visible but aren’t.
With the machine learning tools along with traditional backend data, the Gemini powered system flags this disruptive content which essentially forces users to interact with them against their will. Once these disruptions are identified, they are sent for a human review in order to ensure proper policy enforcement.
In a pilot program conducted from December 2023 to October 2024, a 40% drop in mobile invalid traffic linked to ads that violate placement rules was registered. Although the system is not yet fully automated, Google reports that the pilot not only reduced such activity but also enhanced the speed, precision, and consistency of enforcement, paving the way for broader implementation in the future.
Addressing IVT is critical to safeguarding advertisers, publishers, and end users, Per Bjorke, Google’s director of product management for ad traffic quality, told Adweek. In the first quarter of 2025 global IVT traffic rate on web was 18% and on mobile apps was 31%, remaining a significant challenge for the digital advertising industry.
Google says it’s confident that its AI investments will continue to play a key role in reducing invalid traffic and ad fraud. Bjorke added “We make advancements, the bad actors adapt. We make more advancements, and they adapt again.” According to Bjorke, AI tools offer a critical advantage in this cycle by responding and adapting faster than traditional methods, allowing them to catch malicious activity sooner.
