Fifty-one percent of web traffic is now non-human. AI-driven scraping grew 7,851% in 2025. The ratio of pages scraped to human visitors has shifted from 2:1 to 6:1, with some AI systems hitting 250:1. Publishers are being read more than ever — and paid less for it than ever. The infrastructure question everyone is rushing to answer is: how do we meter this?
The question is wrong. Metering is two problems, and the industry is solving the easy one while ignoring the hard one.
What's Already Solved
The first problem — who is consuming, and what are they taking? — is essentially an engineering exercise, and the solutions are arriving fast.
Identity. The IAB Tech Lab Agent Registry launched in March 2026 with 10+ registered agents across DSPs, identity resolution providers, CDPs, and measurement firms. Amazon, Optable, and Dstillery are already in. The registry is itself an MCP server — Claude and Cursor can query it directly. OpenAttribution's AIMS spec adds a decentralized identity framework: verify which agent is requesting content before serving it. And the file-standard ecosystem — ai.txt, W3C's TDMRep (legally binding in the EU via HTTP headers), llms.txt (adopted by Vercel, Stripe, Shopify, Anthropic) — gives publishers granular purpose-based permissions: allow RAG, block training, permit inference.
Usage tracking. Microsoft's Publisher Content Marketplace (PCM), launched February 2026, maps Copilot consumption back to individual content assets. People Inc is a co-design partner alongside AP, Condé Nast, Hearst, and Vox. OpenAttribution's Telemetry spec tracks retrieval events, citation events, display events, conversation turns, and commerce outcomes. The IAB's AAMP (Agentic Advertising Management Protocols) layers agent-aware extensions onto OpenRTB. PMG has already integrated it into their Alli operating system.
These are real systems, shipping now, backed by major players. The "who" and "what" of content consumption in the agentic web will be tracked within 12-18 months at reasonable fidelity. This is not the problem.
What Isn't Solved
The second problem — what was the content worth? — is not an engineering exercise. It may not be solvable at all.
When a user asks Copilot "what's the best cast iron skillet?" and gets a synthesized answer drawing from Serious Eats, Wirecutter, and Food & Wine, what was Food & Wine's contribution worth? 33%? 60%? Zero, because the user never clicked through?
This is not a metering question. It's a valuation question. And the answers proposed so far reveal how unsettled it is:
| Model | Who | Attribution Logic | Problem |
|---|---|---|---|
| Flat-fee licensing | OpenAI | None — all-you-can-eat | Doesn't scale with usage |
| Usage-based metering | Microsoft PCM | Per-asset retrieval | Measures retrieval, not value |
| Citation-based | OpenAttribution | Track citation events | Cited ≠ influential |
| Intent-based | D/Cipher (People Inc) | Semantic intent signals | Only works on owned traffic |
Every model picks a proxy for value. None of them are value. This is the same problem digital advertising spent 15 years on — last-click attribution was wrong, multi-touch attribution was expensive and still wrong, and the industry mostly gave up and settled for "good enough" probabilistic models.
Content metering in the agentic web is at the last-click stage. "We retrieved your article" is the 2026 equivalent of "someone clicked your banner ad." It tells you something happened. It doesn't tell you what it was worth.
Why This Matters for People Inc
People Inc's position here is better than most publishers realize — and worse than their licensing deals suggest.
The advantage: D/Cipher's infrastructure — 12,000-term taxonomy, proprietary identity graph, 12 billion intent signals — was built to solve a structurally similar problem for advertising. Knowing what content a user consumed and attributing purchase intent to that consumption is what D/Cipher does. Their Switchboard integration unifies 200 data streams into 2 billion rows per day. The plumbing for metering AI consumption exists. It just needs a different outlet.
The problem: D/Cipher works because People Inc controls the environment. They own the page, the ad slot, the identity signal, and the conversion pixel. In the agentic web, they control none of that. The content leaves. The response is synthesized elsewhere. The conversion happens in a chat window People Inc will never see.
The engineering lesson: attribution infrastructure built for an environment you control does not transfer to an environment you don't. D/Cipher is an advantage for owned-and-operated metering. It is not an advantage for agentic metering.
What the Engineering Team Should Actually Do
1. Ship PCM integration immediately. Microsoft's Publisher Content Marketplace is the least-worst option. It's live, People Inc is already a co-design partner, and usage-based metering — even imperfect — establishes a revenue baseline that flat-fee licensing does not. The engineering lift is modest: content asset IDs mapped to PCM's reporting API, plus instrumentation to track which assets get retrieved and cited. This is weeks of work, not months.
2. Do not build a proprietary attribution system. The instinct will be to extend D/Cipher into agent metering. Resist it. The attribution problem is unsolved industry-wide, and building a bespoke system means betting on a valuation model before the industry has converged. OpenAttribution, AAMP, and PCM are all competing approaches. Participate in all three. Build for none of them exclusively.
3. Instrument for optionality. What is worth building: a content telemetry layer that captures every retrieval event — which agent, which asset, what context, what time — and stores it in the same warehouse Switchboard already feeds. When the industry eventually settles on an attribution model, the publisher with the deepest retrieval history wins. This is not attribution infrastructure. It's logging.
4. Watch the ad analogy. The agentic web will likely follow the same path digital advertising did: attribution will remain imperfect, and the market will settle on a mix of flat-fee access (like CPM) and performance-based payments (like CPA) negotiated bilaterally. The publisher who can demonstrate intent value — "users who consumed our content through your agent converted at 2.3x the baseline" — will set prices. That's a D/Cipher capability. But it requires the agent platform to share conversion data, which none of them do today.
The metering problem is not a technology problem. It is a market structure problem wearing technology clothes. The infrastructure to track who took what is arriving on schedule. The infrastructure to prove what it was worth may never arrive at all — and that's fine. The advertising industry made $600 billion last year with imperfect attribution. Content metering will follow the same path: good-enough measurement, bilateral negotiation, and market power determining who gets paid. People Inc's job is to be the publisher with the most data when that negotiation starts.