Prebid.org today published a field guide that follows the money through four models by which artificial intelligence companies pay publishers for content, setting a reported OpenAI agreement with News Corp worth more than $250 million over five years beside a Cloudflare setting that lets a site charge AI crawlers from $0.001 per page.

The article, "Getting Paid in the AI Era: A Field Guide to the New Monetization Models," was written by Anton Nashkerskyi of Sigma Software and Krish Arvapally of Dappier for the Prebid LLM Taskforce. It is the latest entry in a publisher series that began on September 30 with a survey of AI traffic, continued on October 8 with crawler business rules, and on October 16 introduced the four payment models under the headline "Who Gets Paid When an AI Reads Your Page?" Today's installment, according to its authors, "follows the money through each model: who pays whom, and on what basis."

In Short

AI companies read news and other web pages to answer people's questions, and publishers want to be paid for that. A group from Prebid, the organization behind widely used open-source ad auction software, wrote a guide that explains the four ways that payment works now: per page, per use, a share of the AI company's revenue, or a negotiated contract. Each model hands the publisher a different price, a different amount of control and a different answer to the question of who is actually paying.

Four ways the money moves

Prebid is best known for the header bidding wrapper that runs auctions on publisher pages. Its taskforce work on large language models extends that interest from ad auctions to content access, and the guide treats AI licensing in the same practical register: fees, intermediaries, logs and payouts.

The four models are defined by who pays and what triggers the charge. In pay per access, the AI company pays for each page its crawler fetches. In pay per use, a buyer pays for a permitted use of the content, such as summarizing or displaying a page. In revenue share, an AI product shares part of its own revenue and divides it according to its own measure of each publisher's contribution. In a direct deal, the AI company pays the publisher under a contract with no intermediary.

The guide gives each model a "main limit." Those limits are, in places, more informative than the descriptions themselves, because they show where the money stops flowing.

Pay per access: a price on every fetch

The simplest model charges the crawler at the door. According to the guide, Cloudflare's Pay Per Crawl, which it describes as being in closed beta, lets the publisher set a price of at least $0.001 and vary it by content. Cloudflare acts as merchant of record, billing the AI company and paying the publisher, and "does not publish its share." That share is, in effect, Cloudflare's take rate, and the guide offers no figure for it.

Amazon Web Services offers a similar per-request charge in its CloudFront network, "settled in stablecoins," according to the guide, which cites a July 2026 InfoQ report on Cloudflare and AWS adding x402 micropayments. Cloudflare's own open-source x402 proxy template, updated in January 2026 with per-route pricing, handles signed payments typically made in USDC, a dollar-pegged token, on the Base network.

For a site already behind Cloudflare, the guide notes, "it is a setting rather than an integration," and "each charge matches a request the publisher can see in its own logs." That auditability is the model's selling point. No separate report from the buyer is needed to check what was billed.

Its weakness is the unit of sale. "The main limit is that the price is per fetch, whatever the AI company then does with the page," the guide states. "A crawler that caches a page pays once, however many answers it later serves from that copy." It adds that "Cloudflare has not named the crawlers that pay through it."

At the minimum price, a million paid fetches would produce $1,000 before Cloudflare's undisclosed share, by PPC Land's calculation. The figure is not new to the industry. When the IAB Tech Lab formed its Content Monetization Protocols working group in August 2025, its cost-per-crawl example was $0.001 per HTTP request, and TollBit, Dappier and ProRata were among the startups at the workshop.

Where Cloudflare has moved since

Cloudflare opened pay per crawl in private beta on July 1, 2025, built on the rarely used HTTP 402 "Payment Required" status code, with crawlers receiving either the page or a 402 response carrying a price. In August 2025 it expanded AI Crawl Control with customizable 402 responses that could carry licensing terms.

A year after the launch, the company signaled that charging per crawl was not the end point. On July 1, 2026, Cloudflare announced a move from charging per crawl toward paying when content is used in an answer, naming Ceramic.ai and You.com as partners without disclosing pricing for either. In the same announcements, Cloudflare said that from September 15, 2026, crawlers classified as Training and Agent would be blocked by default on ad-carrying pages for newly onboarded domains.

The guide's caching caveat explains why the shift matters. Cloudflare said at the time that more than half of legitimate bot crawl traffic re-fetched unchanged pages. A toll that counts fetches measures crawler behavior, not the value an AI product extracts from a page.

Pay per use: pricing the purpose

The second model shifts the charge from the request to the purpose. The buyer, which the guide defines as "the AI company paying for content," pays for a permitted use. A marketplace verifies the buyer, bills it and pays the publisher. Five operators are described.

TollBit

On TollBit, the publisher sets a rate per 1,000 pages for each license type. TollBit suggests pricing against current RPM, defined in the guide as "what 1,000 human pageviews earn in ads," so that AI access is priced against what the pages already earn. Its examples run from $1 to $5 per 1,000 pages. By PPC Land's calculation, that is $0.001 to $0.005 per page, which puts the bottom of TollBit's range at exactly Cloudflare's minimum. According to the guide, "TollBit takes no share of the rate and charges the buyer a fee on top."

Cloudflare Pay Per Use

Cloudflare's Pay Per Use has been in beta since September 30, according to the guide. Its mechanics differ from TollBit's in one important respect: the buyer proposes a price for each use and reports its use through an API. The publisher's record of what was used therefore depends on what the buyer reports, a contrast with the log-matched charges of Pay Per Crawl.

Microsoft's Publisher Content Marketplace

Microsoft's Publisher Content Marketplace is described as a pilot with a public interest form, in which publishers set licensing and usage terms for grounding uses, meaning content fed into AI answers at the moment they are generated. According to the guide, Microsoft says it pays for the value content delivers and in February "was still testing pricing models." Gannett announced in October 2025 that it was partnering with Microsoft on the marketplace's launch, describing structures that included direct licensing fees and revenue-sharing components without giving amounts.

Dappier and FORA

On Dappier's RAG Marketplace, publishers list their content and set a price per query. FORA is presented as "an open protocol for the same kind of transaction." Publishers set their prices, and the exchanges that run the protocol earn a margin on each access. A publisher or group can run its own exchange, "taking on work a marketplace would otherwise handle," according to the guide.

What pay per use offers, and who is buying

The model's advantage, as the guide puts it, is that "the price can follow the use, so a full display of an article can cost more than a summary of the content." Marketplaces also keep records of each use, such as TollBit's transaction log of crawler, page and price.

The obstacle is on the other side of the market. "The main limit is that the demand side of the marketplace is not quite built out, largely due to alternative access points through web scraping tools like Firecrawl," the guide states. Why pay a marketplace for a page that a scraping service will fetch anyway?

The early buyers it can name are few. Cloudflare names Ceramic.ai and You.com as initial partners. Microsoft names Yahoo as an early demand partner after testing in its own Copilot. TollBit named the news app Particle as an early buyer in 2024. FORA has a working proof of concept.

PPC Land's coverage points to the same gap. When Next Net and Sundial launched a licensing ledger for ESSENCE, Refinery29, AFROPUNK and Beautycon in July 2026, financial terms were undisclosed and no AI company was named as a participant. The Copyright Clearance Center had earlier added pay-per-use AI Transactional Rights, starting with content summarization, in March 2026. On the standards side, the IAB Tech Lab's CoMP v1.0 specification, released for comment on March 10, 2026, deliberately avoided picking one model and listed pay-per-crawl, pay-per-use and outcome-based approaches side by side.

Revenue share: pools set by the platform

In the third model, the publisher does not set a price at all. An AI product shares part of its own revenue and allocates it by its own measure of contribution.

ProRata, according to the guide, shares 50% of its revenue with content partners. Perplexity's Comet Plus, a $5 monthly subscription, pays publishers according to human visits, search citations and agent actions. Axios reported that publishers receive 80% and Perplexity keeps 20% for computing costs, with an initial pool of $42.5 million. On the standalone price, that is $4 per subscriber per month flowing to the publisher pool, by PPC Land's calculation. PPC Land reported when Perplexity announced Comet Plus in October 2025 that the subscription was also included at no additional cost in its Pro and Max plans, with CNN, Condé Nast, Fortune, Le Figaro, Le Monde, the Los Angeles Times and the Washington Post among the launch publishers.

Google's invite-only pilot "pays sites whose content it judges to have contributed significantly to an AI answer," according to the guide. In September, The Information reported that Google pays about 100 publishers through that pilot, with one early entrant earning more than $1 million a year and several small sites receiving less than $1,000 over several months.

The guide's verdict on the model is blunt. "The main limit is that there is no uniformity in business model," it states, which "leads to fragmented relationships between demand partners looking to consume unique content and publishers supplying it." It notes that "Perplexity has not said how it weights its three factors, Google does not publish its calculation, and pooled payouts rise and fall with the operator's own revenue." A publisher in a revenue-share pool is, in other words, exposed to the commercial performance of an AI product it does not control.

Direct deals: the full fee, for a few

The fourth model removes the intermediary entirely. "The AI company pays the publisher under a contract, with no intermediary, so the publisher keeps the full fee," the guide states.

Its examples carry the largest figures in the document. Dotdash Meredith, now People Inc., receives at least $16 million a year in fixed payments from OpenAI, plus a variable component. Meta will reportedly pay News Corp up to $50 million a year for three years, covering Meta AI answers and training. Amazon reportedly pays The New York Times at least $20 million a year. The News Corp agreement with OpenAI, reportedly worth more than $250 million over five years, averages $50 million a year by PPC Land's calculation, the same as the reported annual ceiling of the Meta deal.

Under a direct deal, "the publisher negotiates terms directly, including covered uses and how payment is calculated." That control comes at a cost in time and access. "The main limit is that deals take months to negotiate and have so far gone mainly to the largest publishers, leaving most of the long tail outside them, as the first article in this series found," the guide states.

PPC Land's earlier reporting shows how such agreements reach the books of large groups. USA TODAY Co. said its AI agreements with Meta and Microsoft had a notable impact on its first-quarter 2026 results, as digital other revenues, which include AI partnerships and licensing, grew 125.6%. News Corp chief executive Robert Thomson has described a strategy of wooing and suing AI companies, signing agreements with some while litigating against others.

Answer engines on publishers' own pages

The guide also describes a route to AI revenue that sits outside the four models, because the AI company is not the one paying. "Some publishers also earn from AI on their own pages," it states. An on-site answer engine lets readers ask follow-up questions about an article. Advertisers pay rather than AI companies: ads run with the answers, and the publisher either shares that revenue with the provider or sells the space itself.

Taboola's DeeperDive, ProRata's Gist Answers, Direqt, Inline AI and Dappier's AskAI work this way, according to the guide. Taboola launched DeeperDive on June 11, 2025 with Gannett's USA TODAY Network and The Independent, and Reach selected it for the Express and Daily Star in February 2026.

Only one provider's mechanics are set out in detail. "Dappier shared how its version works," the guide states. AskAI answers from the publisher's content, licensed content or both. Ads in the conversation include programmatic ads, product listings and sponsored conversations, "in which a brand's AI agent answers readers' questions." Dappier collects revenue from what the guide calls agentic ads "and typically shares it 50/50 with the publisher, with payouts based on earned advertising revenue, similar to a traditional ad network." Dappier also "covers LLM, token and hosting infrastructure costs, allowing publishers to get started without taking on those expenses."

The section closes with a link back to the four models. "An engine that draws on other publishers' articles has to license them, so these products also add buyers for pay per use." For the marketplace operators described earlier, on-site answer engines could become a source of the demand the guide says is missing.

The guide's five starting points

The guide ends with five steps for publishers. The first is to know the volume by counting a month of declared AI crawler fetches in the logs, so that any quoted rate can be tested against the publisher's own traffic. The second is to assess the audience opportunity, weighing "reader engagement, ad revenue and any licensing cost together" for an on-site answer engine. The third is to ask any network or marketplace "which AI companies pay through it today."

The fourth concerns the take: whether a fee comes out of the publisher's rate or is added on top for the buyer, and how a revenue share calculates the publisher's part. The fifth covers contract terms: which uses are covered, for how long, whether the arrangement is exclusive, how payments are booked, the minimum payout and "any tax withholding."

The list reads like a procurement checklist for ad tech, and that is the point. The questions about fees, buyers and reporting are the ones publishers have long asked of supply-side platforms and ad networks.

Who wrote it, and what it leaves open

Two disclosures shape how the guide reads. Arvapally works for Dappier, which appears in the document twice: once as a pay-per-use marketplace and once as an on-site answer engine provider, the only one whose revenue split and cost model are described in detail. Nashkerskyi works for Sigma Software and co-wrote the October 16 article that introduced the four models with Constantine Mirin.

The sourcing is largely secondary. The biggest numbers, for Meta and News Corp, Amazon and The New York Times, and News Corp and OpenAI, are described as reported rather than confirmed, and the guide's source list relies on outlets such as Press Gazette, Engadget, GeekWire and Axios. The Sponsored Conversations reference points to a company announcement distributed through EIN Presswire.

Several questions remain unanswered. The guide does not estimate how much money any of the four models has paid out in total. Cloudflare's share of Pay Per Crawl revenue is unpublished, and so is the list of crawlers paying through it. Perplexity's weighting and Google's calculation are undisclosed. Microsoft was, as of February, still testing pricing. For most of the market, the actual price of AI access remains a number set by the buyer, the intermediary or a confidential contract.

Why it matters for publishers and advertisers

The economics underneath the guide have been visible in PPC Land's coverage for some time. Cloudflare's crawl-to-referral ratio, the number of pages an AI platform collects for each visitor it sends back, has ranged from about 118 to nearly 50,000 crawls per referral on its dashboard. The September 30 article in Prebid's own series cited TollBit's scrape-to-referral ratio of 227 to 1 in the second quarter of 2026, up from 150 to 1 in the first. Every model in the guide is an attempt to price what those ratios leave unpaid.

Compliance remains a weak point for any toll. TollBit data for the first half of 2026 showed that about 15% of identified AI page fetchers in Europe reached URLs that sites had disallowed. A fetcher that ignores the rules never meets the price, which is one reason the guide places the demand-side problem at the center of the pay-per-use model.

For advertisers, the most direct consequence sits in the section outside the four models. On-site answer engines create a new ad surface inside publisher pages, with programmatic ads, product listings and brand-run AI agents answering readers in conversation. That inventory is sold by publishers and their providers, not by the AI platforms that dominate the rest of the discussion.

Prebid's involvement also follows a pattern. The organization took over stewardship of the open-source sales agent publishers need for AI-driven buying in January 2026, and its LLM Taskforce series now applies the same publisher-side lens to content licensing. Google chief executive Sundar Pichai said in December 2024 that a marketplace would emerge and that "people will figure it out". Almost two years later, the guide describes several marketplaces and few named buyers.

What comes next

The series continues. According to the guide, the next article, "The New Tollbooths: AI Content Marketplaces and Intermediaries Explained," will compare the marketplaces behind several of these models.

Timeline

Summary

Who: Prebid.org and its LLM Taskforce, through an article by Anton Nashkerskyi of Sigma Software and Krish Arvapally of Dappier. The guide describes Cloudflare, Amazon Web Services, TollBit, Microsoft, Dappier, FORA, ProRata, Perplexity, Google, OpenAI, Meta, Amazon and publishers including People Inc., News Corp and The New York Times.

What: A field guide to four ways AI companies pay publishers for content: pay per access, pay per use, revenue share and direct deals, plus on-site answer engines funded by advertisers. It sets out who pays whom in each model, the reported figures, from $0.001 per page to direct deals worth up to $50 million a year, and the main limit of each, and closes with five steps for publishers.

When: Published today, November 2, 2026, as the fourth entry in a publisher series that began on September 30, 2026.

Where: On the Prebid.org blog. Most of the companies and deals it names are based in the United States, while Cloudflare's and AWS's network-level tools are offered to sites worldwide.

Why: AI systems collect far more pages than they send readers back, and the guide maps the payment options that have emerged to cover that gap. Its central finding is that the best-paid route, the direct deal, has gone mainly to the largest publishers, while the marketplaces open to everyone else still lack enough paying buyers.