Google Targets Bad Ads, Not Bad Actors, to Clean Up Online Marketplace

Google’s 2025 Ads Safety Report shows that the surge in blocked ads is largely a byproduct of the company’s deeper integration of its Gemini family of AI models into the ad‑delivery pipeline, according to the report released Thursday. By deploying Gemini to scan creative assets, landing‑page copy and campaign metadata before they reach users, Google says it intercepted more than 99 percent of policy‑violating ads “earlier and with greater precision” than in prior years. The result was a record 8.3 billion blocked ads worldwide—up from 5.1 billion in 2024—while the number of advertiser‑account suspensions fell dramatically. The shift signals a strategic pivot from blunt, account‑level enforcement toward granular, content‑level policing, a move the company believes will curb false positives and preserve legitimate spend.

The data also reveal a geographic split that underscores the model’s scalability. In the United States, Google removed 1.7 billion ads and suspended 3.3 million advertiser accounts, with the most common violations tied to ad‑network abuse, misrepresentation and sexual content, the TechCrunch report notes. By contrast, in India—Google’s largest user market—the firm blocked 483.7 million ads, nearly double the prior year, while account suspensions dropped from 2.9 million to 1.7 million. The decline in suspensions, Google’s VP of ads privacy and safety Keerat Sharma told reporters, reflects “layered defenses” such as mandatory advertiser verification, which weeds out bad actors before they can create campaigns. Sharma added that the AI‑driven, creative‑level approach has cut incorrect suspensions by roughly 80 percent year over year, a metric that suggests the technology is improving both efficiency and accuracy.

A deeper look at the composition of the blocked inventory highlights the growing threat posed by generative‑AI‑powered scams. Of the 8.3 billion ads stopped, 602 million were linked to scam operations, and 4 million advertiser accounts were identified as fraudulent, the report states. Google attributes the rise in scam‑related ads to the ease with which malicious actors can now generate deceptive content at scale, a capability that Gemini’s pattern‑recognition algorithms are specifically designed to counter. By analyzing large‑campaign signals—such as repeated phrasing, image manipulation signatures and anomalous targeting patterns—Gemini can flag and block suspect ads before they appear on search or YouTube, a capability that traditional rule‑based filters struggled to achieve.

The broader strategic implication is that Google is using Gemini not only as a defensive tool but also as a catalyst for automation across its ad ecosystem. The company is embedding the models into campaign creation workflows, real‑time policy monitoring and threat‑response loops, according to the TechCrunch briefing. This integration allows the platform to react to emerging violations within minutes, rather than relying on post‑hoc human review. Analysts have long warned that the ad‑tech arms race will increasingly hinge on AI’s ability to parse intent and context at scale; Google’s numbers suggest it is ahead of many competitors in operationalizing that capability.

However, the report also raises questions about the long‑term efficacy of a content‑only enforcement regime. While blocking individual ads reduces collateral damage to legitimate advertisers, it does not eliminate the underlying infrastructure that enables fraudsters to spin up new accounts. Google’s reliance on verification and AI‑driven detection may lower the incidence of false suspensions, but the persistence of 4 million fraudulent accounts in 2025 indicates that a portion of the threat surface remains unaddressed. As the advertising market continues to expand—projected to exceed $800 billion globally by 2026—the balance between precision policing and systemic deterrence will be a key metric for investors evaluating Google’s ability to safeguard its revenue stream without stifling growth.