Let’s be clear: a study just dropped showing that over 35% of new web pages now carry fingerprints of AI authorship. The number is staggering, but the real shock is the silence around it. The internet is being diluted by synthetic content, and we’re treating it like a UI update. The data suggests we’ve already crossed the threshold where most new information is machine-generated. The question isn’t whether this is happening—it’s whether anyone can still verify what’s human.
I’ve spent the last three years auditing smart contracts, not scraping web pages, but the pattern is identical. In both cases, the system lacks a native trust layer. Web pages have no on-chain birth certificate. A blog post, a news article, a product review—all just bytes served over HTTP. The server that signs the TLS certificate has no idea if the content was written by a person or a prompt. This is the same class of oracle problem we see in DeFi: a data feed that can’t be cryptographically verified at the source. The industry’s response so far is to build more detection tools, which is like patching a reentrancy bug with a modifier that checks gas left. It’s a temporary fix, not a protocol upgrade.
The Core Mechanism: Attestation, Not Detection The only durable solution is to move from probabilistic detection to deterministic attestation. In practice, this means embedding a cryptographic proof of human authorship at the moment of publication. The stack works like this: a content creator uses a lightweight client (a browser extension or a CLI tool) that generates a zero-knowledge proof of the writing process. The proof captures keystroke dynamics, editing patterns, and tool usage—telemetry that reveals a human cognitive signature without exposing raw text. The proof gets posted to a smart contract on a low-cost rollup. The content itself is hashed, and the hash is stored on-chain alongside the attestation. A reader can verify, in real time, that the page they’re loading matches a hash that was signed by a human prover.
This isn’t science fiction. During my audit of a privacy-focused DAO in 2024, I reverse-engineered a zk-circuit that measured typing entropy. The original goal was to prevent bot accounts from voting, but the same primitive applies to content. The circuit constrains the statistics of inter-key intervals and backspace usage to a range that is statistically impossible for a language model to replicate. The proving time on a consumer laptop was under 200 milliseconds. The gas cost to verify the proof on Optimism was less than 40,000 gas—cheaper than a standard ERC-20 transfer. The bottleneck isn’t the cryptography; it’s the adoption.
The Contrarian Angle: Why Attestation Will Fail Without a Liquidity Incentive Here’s the uncomfortable truth: the technical stack is ready, but the economic model is broken. Content creators won’t adopt a human-attestation protocol unless it pays them. The current web runs on attention, and attention is a function of volume, not veracity. An AI-generated listicle can generate more ad revenue per minute of effort than a deeply researched, human-verified piece. The only way to tilt the equilibrium is to create a liquid market for verified human content. Imagine a bonding curve where each verified article mints a non-transferable attestation NFT, and the protocol buys back and burns the NFT based on page views or curation signals. The creator gets a direct revenue stream, and the protocol bootstraps a reputable corpus of human-generated data. Without this, attestation is a public good with no funding—a classic tragedy of the commons.
Code does not lie, but it often forgets to breathe. The emotional layer of the internet is not logic; it’s incentive. A miner doesn’t secure the chain because they love decentralization; they do it for the block reward. The same principle applies here. If we want humans to keep writing, we need to pay them in a way that machines can’t game. A naive attestation system without economic incentives will be ignored, just like the “no AI” badges on art platforms that are already being spoofed.
Takeaway: The Halving of Human Content The 35% figure is a lagging indicator. By the end of 2025, I expect the ratio of new AI-generated to human-generated web pages to exceed 2:1. The real question is whether the remaining human content will be cryptographically distinguishable. If not, the internet becomes a low-fidelity simulacrum of itself, and the value of any piece of text trends toward zero. The only countermeasure is to make human authorship provable and profitable. The protocols that build this attestation layer will capture the value of truth itself. The ones that don’t will be left indexing a ghost town of synthetic text.