Cloudflare has begun the beta version of its Pay Per Use service, which targets a growing problem in the content economy: AI products may read publishers’ pages and provide their summaries to users without the user visiting the site, depriving the publisher of the visit and the revenue associated with it. Instead of charging for every crawl, the service ties payment to the actual use of content within an AI product.
The idea is for the AI company to define the type of use it will pay for and its price, after which the publisher reviews the offer and accepts or rejects it. The company then reports each use, while Cloudflare aggregates the records, bills the buyer, and pays the publisher’s dues monthly through the payment account linked to the service.
Paying for use, not crawling
Cloudflare believes that charging for crawling alone does not reflect the final value of content, because search engines and AI products may retrieve many pages that they do not later display or use. By contrast, Pay Per Use links financial compensation to what happened after the page was accessed, such as displaying a snippet in a research answer, using a product review in a recommendation, or quoting a paragraph in a report produced by a research agent.
This mechanism differs from Pay Per Crawl, which Cloudflare launched in 2025 and which charges for access to content. Under the new model, publishers can choose the model that best suits them, and they can also accept different offers for the same article depending on the type of use.
How does the service work?
The AI company begins by creating a program within Cloudflare, identifying its crawling bot, defining the paid use, setting the price, and specifying the payment account. Publishers see the offers in the dashboard under Monetize, then Pay Per Use, along with the company’s name, the type of use, and the proposed price. Joining does not require changing the source or building a separate technical integration with each company.
Publishers retain control over access to content and subsequent uses, including restrictions related to training, according to each program’s terms. When an offer is accepted, the purchasing company sends a report for each use in a JSON record containing the time of use, the content URL, and an event identifier. Cloudflare says usage is self-reported, but it verifies that each record is associated with a publisher participating in the program.
What changes in practice for publishers?
The service gives publishers visibility beyond the number of visits by AI bots, including how many times the content was actually used and how much revenue it generated by domain and over time. Cloudflare plans to add more context to the reports, such as the keywords that led to the quotation, the topic of the request, or the product that benefited from the content, without exchanging personal data.
The service combines this data with the Answer Engine Optimization tool, which shows how assistants answer questions related to the publisher’s content. This may help the publisher make business decisions, such as identifying the types of content that generate the most revenue, and editorial decisions, such as what should be updated or made easier to discover.
Limitations and open questions
Pay Per Use is still in the beta phase, and Cloudflare will work directly with participating buyers and publishers to assess the model’s viability. Its success depends on the accuracy of self-reporting, publishers’ confidence in measurement, and whether the revenue is sufficient compared with the cost of making content available. The definition of “paid use” is also set by the buyer, making each program’s terms and price a decisive factor in the participation decision.
Cloudflare says its subsequent goal is to enable new buyers to join with different products and payment models without rebuilding the registration, reporting, and settlement mechanisms for each publisher. It also wants to give publishers greater ability to accept, reject, or negotiate offers, allowing content to be priced according to its recency, value, and type of use.