"Serving Ads Only to Humans": Hakuhodo's Blockchain Trial and the Promise—and Limits—of Ad-Fraud Defense in the AI Era

What Was Done

Three companies—Hakuhodo's new-business arm (Mirai Design business unit), LG Electronics, and Tools for Humanity (which involves Sam Altman and others)—trialed a "Human-Verified Ad Network" as a countermeasure to ad fraud in the AI era. The design is three-layered: Hakuhodo's mini-app "boba" as the user touchpoint, Tools for Humanity's "World ID" to prove one is a unique human, and LG Electronics' "Web3 Ad Network" to record ad-impression history on a blockchain—aiming for a transparent economy where "cost is incurred only for ads genuinely shown to humans."

The trial ran in Japan from July to August 2025, with over 3,500 users and 10 advertisers. Compared with conventional Web2 advertising, human-verified ads without incentives raised CTR by about 50% and improved the bounce rate by about 15 points. Adding a "Watch to Earn" reward feature pushed CTR up to roughly sevenfold.

Why "Now": AI Mass-Produces Fake Humans

The backdrop is a surge in AI-driven fraud. With generative AI, fake ads, fake users, fake clicks, and fake devices can be produced cheaply, making them hard to distinguish from legitimate activity. In 2025, Google blocked or removed 8.3 billion ads and suspended 24.9 million advertiser accounts, of which 602 million were directly tied to fraud. Market-wide, invalid traffic causes an estimated US$63 billion (about ¥9.5 trillion) in annual losses, and Juniper Research projects ad-fraud losses to exceed US$100 billion in 2026. Bots surpassed human traffic for the first time in 2024 and, by some estimates, account for about 37% of all web traffic. This is precisely why "Proof of Personhood" has emerged as a countermeasure.

The Core: A Two-Layer Design of Human Verification and On-Chain Recording

The novelty lies in combining two layers. First, World ID uses the Orb (iris scan) to prove only that one is a "unique human," without revealing personal information. Second, LG's blockchain ledger permanently records every ad impression in a tamper-proof form. Together, they aim to exclude bots and invalid traffic while bringing transparency to the flow of ad spend.

Two Structural Limits

Excessive expectations, however, are unwarranted. The first limit is a misunderstanding of what blockchain does. It faithfully and permanently records the data entered, but does not guarantee that data's "correctness." Reliability depends entirely on the verification layer that precedes recording (identity proofing). If the human-verification layer is breached, fraudulent IDs obtain the same "tamper-proof record" as legitimate ones. What the chain guarantees is "not being altered," not "being true"—a decisive distinction.

The second limit is industry structure. Giants like Google, Meta, and Amazon control their own measurement systems and have little incentive to adopt a neutral, blockchain-based mechanism. Unless the "walled gardens" where most ad spend flows move, even an excellent neutral platform struggles to become an industry standard. This is the same adoption wall common to new fair infrastructures: "value that destroys the middleman cannot move the very incumbents who could spread it."

In sum, this trial is a promising step toward advertising trust in the AI era, but the CTR gains may also partly reflect the quality of the sample (verified and incentivized users). Between PoC success and industry standardization lies the remaining distance of hardening the verification layer and proving value beyond the walled gardens.

[Business Development Insights]

1. A new technology's true value lies in discerning "what it guarantees and what it does not." Blockchain guarantees "tamper-resistance," not "the correctness of inputs." In evaluating adoption, separating the problems a technology solves from those it does not (such as the upstream verification layer), and designing who secures the system's weakest link, is the key to avoiding overestimation.
2. Favorable PoC results must be read together with their "conditions." Here, the CTR gains may be mixed with the effect of sample quality—verified and incentivized users. In new ventures, unless you separately assess whether results stem from "the intervention itself" or from "how the target was chosen and the attached conditions," they will not replicate at scale and will mislead investment decisions. A validation design that isolates the "source" of the effect is essential.
3. A "neutral, fair common platform" will not standardize unless it moves the incumbents who are not its primary beneficiaries. If the players holding the market's volume have no incentive to adopt, even a superior platform will not reach industry-standard status. The breakthrough lies in creating demand from outside them—regulation, advertiser-side requirements, the open domain beyond the walled gardens—and forming a current too large to ignore.

[Sources]

• CRYPTO TIMES, "Hakuhodo verifies ads with blockchain | 50% CTR improvement via identity verification" (June 13, 2026)
• Hakuhodo press release, "Hakuhodo, with LG Electronics and Tools for Humanity, conducts a trial of the 'Human-Verified Ad Network' to suppress ad fraud and serve ads only to humans" (October 14, 2025) — primary source
• RTB SQUARE / plus-web3 (boba; Orb verification; ~7x CTR with Watch to Earn; 3,500+ users, 10 advertisers)
• Lunio, "Global Invalid Traffic Report 2026" (8.51% invalid traffic ≈ US$63 billion; ~US$740 billion in digital ad spend)
• Juniper Research (forecast of US$100 billion+ in ad-fraud losses in 2026); Imperva (bots ~37% of all web traffic, surpassing humans in 2024)
• Google Ads Safety Report 2025 (8.3 billion ads blocked/removed; 24.9 million accounts suspended; 602 million fraud-related)