Google's Gemini AI Is Taking On Ad Fraud

Google's Gemini AI has moved from pilot to the centre of its ad defences. Its 2025 Ads Safety Report says Gemini now stops more than 99% of policy-violating ads before they serve. That is real progress, but Gemini is built for Google's own policy enforcement, not your campaign optimisation. It is largely reactive on the buy side, confined to Google's platforms, and gives advertisers no control or independent visibility. A layered defence, platform AI plus independent, real-time verification, is what actually protects performance.

When the largest ad platform in the world turns its most advanced AI on ad fraud, it warrants attention. What began as a quiet pilot is now central to Google's defences: in its 2025 Ads Safety Report, Google credits Gemini, a system of multimodal large language models, with stopping the majority of deceptive ads before they ever serve. It is a genuine step forward for Gemini AI ad fraud detection. On its own, it is also not enough.

This piece builds on our guide to how machine learning detects ad fraud in real time, looking at what platform-level AI does and does not cover. The takeaway is simple: Google's advances are welcome, and independent click fraud protection remains necessary for reasons that have nothing to do with how capable Gemini is.

What Gemini AI Has Achieved So Far

The origin was a pilot running from December 2023 to October 2024, which delivered a 40% reduction in mobile IVT from "deceptive or disruptive" ads, according to Google Ads reporting. Gemini mimics human behaviour, navigating apps and sites, capturing screenshots, observing ad placements, and flagging when an ad button is invisible yet still registers impressions, or when a pop-up forces unwanted interaction. Those detections are routed through human review for accuracy.

Two years on, the scale is far larger. In its 2025 Ads Safety Report, published in April 2026, Google says Gemini-powered systems stopped more than 99% of policy-violating ads before they served, blocked or removed over 8.3 billion ads, and suspended 24.9 million advertiser accounts, including 602 million ads and 4 million accounts tied to scams. By the end of 2025, the majority of Responsive Search Ads were reviewed instantly and blocked at submission.

That progress is real, but it sits alongside a problem that is not shrinking. Pixalate's Q3 2025 benchmarks, drawn from more than 106 billion programmatic impressions, put global invalid traffic at 21% on web and 33% on mobile apps. Google's gains protect its ecosystem and the people who see ads. They do not, on their own, protect an individual advertiser's budget from the bots and fake clicks that quietly inflate spend.

Where Gemini Helps, and Where It Falls Short

Its strengths are clear. Gemini identifies deceptive formats such as hidden ads, pop-ups and accidental clicks, blocks malicious creative before it serves, and improves policy enforcement and traffic quality inside Google's own ecosystem.

Its limits are just as clear, and they are structural rather than something a better model would fix. The pre-emptive blocking Google reports applies to policy-violating ad creative, protecting users and platform integrity, not the validity of the traffic hitting your campaigns. Incrementality and repeat clicks are still not treated as IVT, so they inflate cost per acquisition without being flagged. Buy-side enforcement remains largely reactive, so ad spend may already be gone by the time action is taken. Coverage stops at Google's platforms, leaving Meta, mobile networks and affiliate channels out of scope. And there is no advertiser-level control or visibility, so you cannot customise thresholds or independently verify what is filtered.

Why Independent Verification Still Matters

Even as Google advances, Gemini is designed for policy enforcement, not advertiser optimisation. Google itself frames this as an arms race, noting that bad actors now use generative AI to produce deceptive ads at scale. The announcement reinforces a truth the industry has long understood: what click fraud actually is keeps changing, and detection is a moving target. As Per Bjorke, Google's director of product management for ad traffic quality, put it: "We make advancements. The bad actors change their behavior. We are making more advancements. They are changing their behavior."

That gap is widening as fraudsters deploy bots that mimic human behaviour well enough to slip past basic filters. Much of what Gemini overlooks falls under sophisticated invalid traffic, which needs full-funnel monitoring to catch.

Independent protection fills the gaps that platform tools leave: real-time prevention that blocks IVT before budget is wasted; cross-channel coverage across Google Search, Meta, programmatic and affiliate networks; rules you control, such as click-frequency caps and device-level blocking; full transparency from click to conversion; and incrementality monitoring that stops repeat activity inflating CAC. This independent layer complements Google's, the separation of church and state.

A Layered Defence Wins

Gemini AI is a meaningful move against ad fraud, not a complete solution. It does not stop repeat clicks, it does not reach beyond Google, and it gives advertisers no control or independent validation. So how you cover the gap? A layered approach: platform-level protection plus real-time, cross-channel ad fraud prevention you control and an audit independently. To see how much invalid traffic Google isn’t blocking on your account,, book a TrafficGuard demo.

Frequently Asked Questions

Does Google's Gemini AI protect my ads on every platform?‍

No. Gemini's protections operate within Google's own ecosystem, such as Google Ads and Google Ad Manager. Ads on Meta, third-party mobile networks or affiliate channels fall outside its scope, which is why cross-platform protection is needed to cover every channel.

Is Gemini AI detecting ad fraud in real time?

Partly. By the end of 2025 Google was reviewing most Responsive Search Ads at submission and blocking harmful creative before it served. That is creative moderation, though. Invalid-traffic enforcement that protects your budget, the bots and fake clicks hitting your campaigns, still happens largely after the ad has served, so spend can be lost before it is credited. Independent tools like TrafficGuard detect and block invalid traffic in real time, before the click or impression is charged.

What kinds of invalid traffic can Gemini AI detect?‍

In its pilot, Gemini proved effective at spotting hidden or hard-to-see ads that still register impressions, disruptive pop-ups and manipulative placements, and accidental clicks. It is less suited to bots, malware-driven traffic, click farms and non-incremental clicks that span multiple platforms.

How does Gemini compare to dedicated click fraud protection software?‍

Gemini improves Google's own policy enforcement; it is not an advertiser-first prevention tool. Dedicated software adds cross-channel coverage, full click-to-conversion transparency, custom rules and thresholds, and real-time prevention rather than post-click enforcement.

What do Google's latest ad-safety numbers actually mean for me?‍

In its 2025 Ads Safety Report, Google says Gemini stopped over 99% of policy-violating ads before they served, and it blocked or removed more than 8.3 billion ads. Those figures measure harmful and deceptive ad creative removed from Google's ecosystem, not invalid traffic eliminated from your specific campaigns. The original 40% figure referred to a mobile IVT reduction in Google's 2023 to 2024 pilot, again within Google's own platforms.

If Google is using AI against fraud, do I still need a third-party tool?‍

Yes, for reasons unrelated to Gemini's quality. Platform tools protect the platform's interests and its ecosystem, not your individual ROI, and they offer no independent verification. An independent layer gives you control, transparency and cross-channel coverage.

Does Gemini catch repeat clicks and non-incremental traffic?‍

No. Google does not treat incrementality and repeat clicks as IVT, so they can inflate CAC without being flagged. Catching them requires incrementality monitoring at the advertiser level, which independent tools provide.

Is Gemini the same as the invalid-click protection already built into Google Ads?‍

No. Google Ads has long filtered some invalid clicks and issued credits after the fact, and Gemini is a separate, broader system aimed mainly at blocking deceptive and policy-violating ad creative before it serves. Neither is advertiser-controlled. The invalid-click filtering is a retrospective credit with little visibility into what was caught, and Gemini protects Google's ecosystem rather than your budget. For real-time control and transparency across every channel, you still need independent verification.