Google blocked or removed 8.3 billion ads in 2025 and suspended 24.9 million advertiser accounts, with 602 million of those ads tied directly to scams.
Those numbers show that the volume of fraudulent material attempting to reach users has grown large enough to require an AI system operating at an industrial scale to contain it.
Gemini now analyzes hundreds of billions of signals in real time, such as account age, behavioral cues, and campaign patterns, catching over 99% of policy-violating ads before they run.
The fraction that cleared that filter still reached users across one of the world’s largest ad networks.
Generative AI has made fake ads, fake users, fake clicks, and fake devices cheaper to produce and harder to distinguish from legitimate activity.
Traditional solutions have proved inadequate as AI-driven fraud evolves faster than detection methods. Google’s answer of using more AI deployed faster commits both sides to continuous escalation.
A separate group of companies is building verification systems that record who saw an ad and make that record permanent.
| Metric | Figure | What it shows |
|---|---|---|
| Ads blocked or removed by Google in 2025 | 8.3B | Fraudulent or policy-violating ad volume is massive. |
| Advertiser accounts suspended | 24.9M | Bad actors are operating at account-farm scale. |
| Scam-related ads removed | 602M | Scams are a major category inside the broader fraud problem. |
| Policy-violating ads caught before serving | 99%+ | AI defense is working, but only by processing enormous signal volume. |
| Signals analyzed by Gemini | Hundreds of billions | Ad safety is becoming an AI-vs-AI infrastructure fight. |
The verified attention model
Hakuhodo, the Japanese advertising giant, partnered with Tools for Humanity and LG Electronics to test a “Human-Verified Ad Network” that served ads exclusively to human-verified users, with every impression logged to LG’s blockchain infrastructure.
The pilot ran in Japan from July through August 2025, involving more than 3,500 participants and ten advertisers across electronics, travel, food, cosmetics, and education.
Hakuhodo integrated its “boba” mini-app with World ID verification and LG’s blockchain ledger, creating a closed loop where only human-verified users received ads and every impression was recorded on-chain.
World ID lets users prove they are unique humans without revealing personal information. Under that architecture, advertisers pay for impressions that carry a verification receipt tied to a confirmed human identity.
According to figures reported by the companies involved, the pilot produced a 50% increase in click-through rates and a 15-point improvement in bounce rates.
A mainstream electronics company and Japan’s second-largest advertising agency ran a blockchain verification test on a live campaign and published the results, separating this move from white paper proposals.
The verified conversion model
In January 2025, Coinbase acquired Spindl, an on-chain ads and attribution platform rebuilding the ad-tech stack on-chain, to address what Coinbase called the “on-chain discovery problem” for blockchain app builders.
Spindl was founded by Antonio García Martínez, an early member of the Facebook ads team who shipped Facebook’s first version of keyword targeting, audience targeting, and Facebook’s programmatic ad exchange FBX.
Spindl focuses on proving that an ad drove real action, such as a wallet interaction, an app install, a token purchase, or a staking event.
Traditional attribution systems infer causality from cookies, click paths, and probabilistic matching. Spindl traces a user journey from a web click to an on-chain action, providing advertisers with a ledger entry and a verifiable chain of custody.
Spindl operates on Base, Coinbase’s Ethereum layer-2 network, and maintains open standards for publishers and advertisers.
The two models address different parts of the same problem: Hakuhodo and LG verify that a human saw the ad, and Spindl verifies that the ad resulted in a real action.
| Model | Example | What it verifies | How blockchain is used | What advertisers get |
|---|---|---|---|---|
| Verified attention | Hakuhodo + LG + World ID | A real human received the ad | Impression history is recorded on-chain after proof-of-human verification | A receipt that the ad reached a verified human user |
| Verified conversion | Coinbase + Spindl | An ad led to a real action | User journey is traced from click to wallet or app event | Attribution from campaign spend to on-chain outcome |
| Conditional payout layer | Future extension | Whether a verified event occurred | Smart contracts or rules-based systems release payment after proof | Pay-for-outcome ad settlement |
| Wallet-based targeting | Crypto apps, gaming, commerce | Audience relevance based on on-chain behavior | Wallet activity helps define segments or campaign eligibility | Targeting without relying only on cookies or device IDs |
Why this matters beyond crypto
Dentsu’s May 2026 global ad forecast puts worldwide ad spend at $1.06 trillion, with digital accounting for 69% of that total. IAB and PwC reported that US digital ad revenue reached $294.6 billion in 2025, with programmatic advertising up 20.5% to $162.4 billion.
The same automated systems that make programmatic buying efficient also expand the surface area where fake inventory, fake users, and fake outcomes get monetized.
Juniper Research estimated that global ad spend lost to fraud would rise from $84.2 billion in 2023 to $172.3 billion by 2028, as AI enables fraudsters to mimic human behavior and evade detection systems.
DoubleVerify found that bot fraud accounted for 65% of all fraud in CTV environments in 2024, with compromised devices simulating real user behavior to deceive measurement systems.
When a fake device can convincingly impersonate a living room viewer watching premium inventory, the platform’s reported delivery numbers are unverified claims.
Blockchain’s pitch to advertisers in that environment is a receipt: an immutable record of what the system observed, attached to a verified identity and fixed at the moment of delivery.
What blockchain cannot do on its own
A blockchain faithfully and permanently records inputs, but its trustworthiness depends on the verification layer that precedes it.
If the identity verification layer is gamed, the fraudulent identity receives the same permanent record as a legitimate one.
The hard problem is the oracle layer: confirming that the viewer was human before the record is written, that the device was legitimate, that the impression was viewable, and that the downstream action was genuine.
World ID’s design separates proof of personhood from personal identity, allowing users to prove uniqueness without revealing their identity.
Advertising is a trust-sensitive use case, and combining human verification, ad targeting, and wallet behavior into a single system will face regulatory and consumer scrutiny in markets where biometric data collection is actively contested.
The adoption constraint is the third. Google, Meta, Amazon, and the major CTV platforms control their own measurement systems and have little incentive to adopt a neutral blockchain-based receipt layer that would weaken their hold on attribution.
Blockchain’s most practical near-term path runs through markets where platform owners have an incentive to increase advertiser trust: crypto apps, independent CTV inventory, rewards campaigns, wallet-based commerce, and gaming.
Two ways this develops
In the bull case, advertisers running high-value performance campaigns demand verifiable logs as proof that probabilistic measurement can no longer supply.
Blockchain verification integrates with existing ad stacks as a parallel audit trail for campaigns where fraud risk justifies the additional infrastructure.
Juniper projects $172.3 billion in ad fraud losses by 2028, and redirecting even 1% to 3% of that figure through verified proof systems points to a protected value pool of roughly $1.7 billion to $5.2 billion.
| Scenario | What happens | Value pool | Where adoption happens first | What blocks adoption |
|---|---|---|---|---|
| Bull case | Advertisers demand verifiable logs for high-fraud campaigns and performance outcomes. | $1.7B–$5.2B protected value pool if 1%–3% of projected 2028 ad-fraud losses move through proof systems. | Crypto apps, rewards campaigns, independent CTV, gaming, wallet commerce, high-value performance ads. | Integration with existing ad stacks and privacy-safe identity design. |
| Base case | Blockchain becomes a parallel audit trail for specific high-risk channels, not a full replacement for Google or Meta measurement. | Niche but commercially meaningful fraud-protection market. | Web3 apps, CTV experiments, on-chain commerce, affiliate attribution. | Advertiser education and fragmented standards. |
| Bear case | Google, Meta, Amazon, and CTV platforms improve AI fraud detection enough to keep measurement in-house. | Blockchain remains a niche verification layer. | Crypto-native apps and limited proof-of-human pilots. | Platform resistance, biometric scrutiny, weak advertiser adoption. |
The Hakuhodo model scales through mainstream platforms, Spindl extends attribution beyond crypto-native apps, and the user never knows that the infrastructure beneath it is a blockchain.
In the bear case, Google, Meta, and CTV platforms improve AI-based fraud detection fast enough that the marginal value of a blockchain receipt layer stays narrow.
Regulatory pushback against biometric proof-of-human systems slows adoption of the verified attention model in key markets.
Blockchain ad tech stays useful inside crypto apps and niche high-fraud channels but fails to cross into the programmatic mainstream.
The $162.4 billion US programmatic market continues flowing through the existing measurement stack, with its fraud losses treated as an accepted line item.
AI has made fake behavior cheap enough that detection systems may permanently lag behind fraud generation. If advertisers conclude that probabilistic measurement can no longer be trusted, blockchain proof systems are positioned to absorb that budget.
