Prebid

Who Gets Paid When an AI Reads Your Page?

by

By Constantine Mirin and Anton Nashkerskyi

For the Prebid LLM Taskforce. Part of the publisher series.

Earlier in this series, The State of AI Traffic: What Publishers Are Seeing in Crawls, Clicks, and Costs showed how much of a publisher’s traffic is now bot-generated, and what it costs. “Block or Allow Was Never the Whole Decision” looked at which bots to allow, restrict, price or block, and this piece picks up at pricing: once you let a machine in, how do you get paid for it?

From letter to launch partner

In July 2024 Condé Nast told Perplexity to stop using content from The New Yorker, Vogue and Wired. It never sued Perplexity. Fifteen months later it was contributing the same content as a launch partner in Comet Plus, Perplexity’s publisher compensation program. Other publishers took the same AI companies to court. Some did both.

Beneath the headlines, quieter machinery is being installed: per-fetch meters, per-use licenses, revenue-share pools. Add direct deals and you have four models, each answering one question differently. When a machine reads your page, who gets paid, how much, and who decides?

Access or licensing: which problem do you have?

Breaking news, sports and live coverage have an access problem. They get heavy bot traffic at publication and earn little or no revenue from it. They need control at the door.

Evergreen explainers, reference material and deep archives have a licensing problem. The value is in the work itself. AI tools can answer from it, so fewer readers need to visit the page.

Publishers with each problem should be shopping in different halves of the market. Pay per access fits the access problem: it prices the fetch. Pay per use fits the licensing problem: it prices what the AI does with your content.

Four monetization models

Pay per access. A verified crawler gets a price instead of the page (HTTP 402) and pays per fetch, whatever it does next. Seen with Cloudflare’s Pay Per Crawl (a closed beta), Akamai with Skyfire’s payment tokens, and the open x402 protocol. Metering search crawlers along with AI crawlers may affect your indexing.

Pay per use. The price follows the use: a broad AI license for grounding, grounding with citation, or display of full text. On TollBit you set rates per page and license type, and Akamai and Fastly can redirect AI bots there. Cloudflare’s Pay Per Use beta reverses this: the AI company proposes the use and the price. Microsoft’s Publisher Content Marketplace (a pilot) also pays by use but hasn’t said how it sets prices. FORA and Sphere use this model too. After delivery, you rely on the buyer’s reports.

Revenue share. You are paid by the operator’s own measure: contribution to each answer for ProRata’s Gist, visits, citations and agent actions for Perplexity’s Comet Plus, and what Google judges “contributed significantly” in its invite-only pilot. Nothing to integrate, but no per-read receipt.

Direct deal. Two parties, a negotiated fee. Maximum leverage, but months per deal, so they reach the largest publishers, not the long tail.

Eight questions that define every monetization model

Every model answers the same eight questions, in roughly the order a buyer meets them.

•  Finding you. Discover: how does a buyer find your content, and who controls the ranking?

•  The door. Identify: which agent is asking? Authorize: what grants this fetch, and can it be forged or replayed?

•  Terms and money. Declare: how are your terms, including permitted uses, expressed so a machine can read them? Price: one price or many, attached to what unit, the page or the passage? Settle: how does money reach you, from whom, and how often?

•  Afterwards. Report: what can you see, and could you audit it? Dispute: what can you do when you disagree?

Discovery matters more than it looks: whoever ranks search results decides which publisher gets paid, and no public documentation we found says how ranking is set. Some AI companies are also building their own indexes for AI use rather than licensing search ones.

The models differ mainly in who answers these questions. Pay per access identifies the crawler, charges per fetch and settles the payment, but says nothing about how the content is used. Pay per use adds terms and a price for each use, but relies on the buyer to report what happened. A revenue-share pool answers almost all of them for you, including the price. A direct deal answers them in a contract. The arrangements that ask the least of you tend to be the ones where one company answers most of the eight questions.

One arrangement sits outside these four models: brands pay publishers to shape how AI assistants describe and cite them, led by BCN’s GEO Brand Impact. It is brand money on the advertising side of the house, so it can run alongside any of the four models.

What you can do

A few checks help place any offer on the map before you sign.

•  Work out whether you have an access problem or a licensing problem, and how you value your content for each.

•  Pull a week of CDN logs and count unique AI crawlers, including ones that don’t declare themselves, to see how much demand you have before you set a price. That number is a floor.

•  Ask each potential partner to answer the eight questions above, and put the answers side by side. Questions they leave open are yours to address separately.

•  Find out who collects the money and pays you, how often, and how much you must earn before a payout.

•  Read all the terms in the agreement, not just the pitch deck, including downstream use, indemnification, liability and exclusivity.

Then ask the vendor who is actually paying through them today, and whether you can start with a short, non-exclusive test.

What comes next

This piece is the map. The next articles zoom in. Provenance and Reporting looks at how content can carry proof of who made it, and how AI companies can report what they did with it. Getting Paid in the AI Era walks through each monetization model and how the money moves. The New Tollbooths covers the marketplaces and intermediaries in the middle. Later pieces cover protocols, measurement, copyright and regulation, and why AI systems might prefer permissioned content.

Sources

Public research

Category: Tag:

Reader Interactions

Leave a Reply