Google Blocked 602 Million Scam Ads With Gemini as AI Turns Ad Safety Into a Real-Time Cyber Fight
Google says it is using its Gemini models around the clock to identify and block scam ads in real time, marking one of the clearest examples yet of how large platforms are turning generative AI into a frontline security tool. In its newly released 2025 Ads Safety Report, the company said Gemini-powered systems helped block or remove more than 8.3 billion ads and suspend 24.9 million advertiser accounts last year, including 602 million scam-related ads and 4 million accounts tied to scams. Google also said its systems now catch more than 99% of bad ads before they are shown to users.
Those are enormous numbers, but the bigger story is not just scale. It is speed. Google says Gemini helps analyze hundreds of billions of signals, including account age, payment behavior, policy history, creative patterns, and campaign relationships, to decide whether an advertiser is acting like a legitimate business or like a fraud operation. That shift matters because scam advertising has become faster, cheaper, and more personalized in the age of generative AI. The platforms defending against it are now being forced to operate at machine speed too.
What Google Actually Reported
Google’s official report says Gemini has improved the company’s ability to understand advertiser intent rather than relying only on static content checks or keyword rules. That is an important change. Scam ads often no longer look obviously malicious on the surface. They may imitate real businesses, use polished language, rotate domains quickly, or exploit trending events and public figures. According to Google, Gemini helps its systems distinguish between normal commercial behavior and coordinated deception more effectively than older models alone.
The topline numbers are striking. Google says that in 2025 it blocked or removed over 8.3 billion ads, suspended 24.9 million accounts, and removed ads from billions of pages across the web. Within that, scam enforcement stood out as one of the most important categories, with 602 million scam-related ads blocked or removed and more than 4 million scam-linked advertiser accounts suspended. That suggests scam prevention is no longer a narrow policy bucket. It is one of the main jobs of the system.
Why Gemini Matters Here
Google’s description of Gemini’s role is less about flashy generation and more about structured judgment. The company says the model helps identify patterns across huge volumes of data, reason about intent, and review suspicious activity more quickly when users report deceptive ads. Google says these improvements helped quadruple actionable reports from users and also reduced incorrect advertiser suspensions by 80%, which is significant because safety systems have to balance aggressive enforcement with fairness to legitimate businesses.
That balance is harder than it sounds. A platform the size of Google has to stop fraud at global scale without breaking the economics of legitimate digital advertising. If enforcement is too weak, scammers thrive. If it is too aggressive and inaccurate, real advertisers get caught in the blast radius. Google is effectively arguing that Gemini helps solve both sides of that problem by improving precision as well as speed.
The Scam Ad Problem Is Getting More Sophisticated
This report lands at a time when AI-generated fraud is becoming more polished across the internet. The Associated Press noted that Google is trying to counter a wave of increasingly convincing scams, including fake celebrity endorsements, health scams, and other deceptive ad formats made easier to produce with generative tools. That is a critical point. The old internet scam often looked sloppy. The new one can look professional, localized, emotionally tuned, and highly scalable.
That evolution changes the nature of ad security. It is no longer enough to rely on static signatures, obvious typos, or known bad domains. Platforms now need systems that can infer motive, behavioral similarity, and campaign-level abuse patterns even when individual ads look clean in isolation. Google’s emphasis on intent analysis suggests that is exactly where the battle has moved.
Why This Is Really a Cybersecurity Story
At first glance, ad moderation can sound like a trust-and-safety or policy issue rather than a security one. That framing is now too narrow. Scam ads increasingly serve as the entry point into financial fraud, credential theft, fake investment schemes, malware delivery, and identity compromise. A deceptive ad is often not the end of the attack. It is the first lure in a broader intrusion or fraud chain.
That is why Google’s numbers matter beyond advertising. Blocking hundreds of millions of scam ads means interrupting enormous volumes of potential downstream harm before a user clicks, calls a fake support line, transfers funds, or enters credentials into a spoofed site. In practical terms, ad safety has become part of the internet’s wider defensive perimeter.
The AI-vs-AI Dynamic Is Now Out in the Open
One of the clearest themes in the report is that Google now sees abuse prevention as a contest between automated offense and automated defense. Attackers are using AI to generate more variations, test more narratives, and scale deception more cheaply. Google is using Gemini to compress analysis time, correlate signals faster, and stop violations before they are served. That is the new shape of the fight.
This matters because it changes expectations for every major platform. Once one company demonstrates that large-scale AI can materially improve abuse detection, the bar rises for everyone else. Security teams inside search, social, commerce, marketplaces, and adtech ecosystems are all being pushed toward the same question: can your defenses reason fast enough to keep up with machine-generated abuse.
What Security Teams Should Take From This
For defenders outside Google, the report reinforces a wider lesson. AI is becoming most useful in security not when it replaces analysts entirely, but when it helps sort massive volumes of weak signals into clearer judgments about risk, intent, and priority. That applies to scam ads, but also to phishing, fraud, account abuse, insider anomalies, and cloud misconfigurations. The pattern is the same: too much data, too little time, and too many adversaries learning to hide inside normal-looking behavior.
There is also a process lesson here. Google’s claim that user reports became more actionable and faster to process shows that AI works best when it strengthens the loop between machine triage and human feedback. The future of security operations is probably not purely autonomous. It is a tighter cycle where humans surface signals, models add context and prioritization, and reviewers move more quickly on what actually matters.
The Limits of the Victory Lap
Google’s figures are impressive, but they should not be mistaken for proof that the scam problem is solved. The same report and outside coverage both suggest the fraud ecosystem is still expanding and adapting. A platform blocking billions of bad ads is evidence of strong defenses, but it is also evidence of how much malicious volume is trying to get through in the first place.
That is why the most honest reading of the report is not that Gemini has ended scam advertising. It is that the scale of abuse has become so large and so dynamic that without AI-assisted defenses, major platforms would likely struggle to cope. The workload is simply too vast for manual review or older rule-based systems alone.
The Bottom Line
Google’s 2025 Ads Safety Report shows just how industrialized the online scam ecosystem has become and how aggressively the company is leaning on Gemini to contain it. Blocking or removing 8.3 billion ads, suspending 24.9 million accounts, and taking action against 602 million scam ads is not just a moderation update. It is a picture of a digital battleground where fraud campaigns and platform defenses are both increasingly automated.
The real takeaway is simple. Scam ads are no longer fringe internet clutter. They are part of a broader cybercrime pipeline, and Google is treating them that way. Gemini’s role suggests the future of this fight will depend less on static rules and more on fast, intent-aware systems that can spot abuse before humans ever see it.
References
