TLDR;
Cloudflare has put Pay Per Use into beta. AI companies can offer to pay when a publisher’s content appears in an answer, report or recommendation, while publishers choose which offers to accept. This is a more useful model than charging for every crawl, but the buyer defines and self-reports each paid use. Enterprise content owners should test the economics and evidence before treating the dashboard as settled revenue attribution.
What happened
Cloudflare launched the Pay Per Use beta on 30 September. A participating AI company identifies its crawler, defines what counts as a paid use, proposes a price and connects a payment account. Publishers see each offer in Cloudflare’s dashboard and decide whether to accept it. Cloudflare then aggregates the reported uses, charges the buyer and pays the publisher monthly.
The model is different from Pay Per Crawl. A crawler may fetch thousands of pages that never influence an answer, so charging for access can create cost before the buyer knows what it needs. Pay Per Use waits for a downstream event, such as an excerpt appearing in AI search or a review shaping a shopping recommendation. Cloudflare first outlined that direction in July; the new event is the working beta with offers, reporting and settlement.
There is a material limitation. The AI buyer reports each use through an API with a timestamp, source URL and event ID. Cloudflare checks that the event maps to an enrolled publisher, but it does not independently observe the full reasoning path inside the buyer’s product. The commercial record is therefore only as strong as the agreed definition of use, the buyer’s reporting and the audit rights around both.
Cloudflare’s wider agentic-web announcement positions the beta as part of a shared identity, metering and settlement layer. Publishers retain separate control over crawler access and can stop participating in an offer. The same content can also carry different prices for different uses, although Cloudflare says the beta is still testing whether buyers and publishers find the terms worthwhile.
Why it matters
This moves AI visibility beyond mentions and referral traffic. A brand may receive no visit when its research, review or guidance changes an AI answer, but that contribution can still have commercial value. Pay Per Use creates a route to record and price that value without negotiating a separate technical integration with every AI company.
Cloudflare also launched a closed Monetization Gateway beta for resources where every request is the use, including APIs, MCP tools and datasets. That product uses HTTP 402 and x402 payments at request time. The distinction matters: reusable editorial content may be fetched once and used many times, while a live product-data query or calculation incurs value and cost on each call.
The opportunity is real, but the buyer currently holds much of the meter. Publishers must understand whether a paid event means a citation, a quoted passage, a recommendation input or something broader. They also need to know what is excluded, how corrections work and whether training rights are separate. Earlier independent coverage of Cloudflare’s crawler-payment experiments showed why access and compensation attract attention; this beta now has to prove reporting quality and useful returns.
How your brand can benefit / be affected
Start with content whose value can be explained. Original research, product testing, specialist guidance, live pricing and authoritative reference material are easier to price than generic pages. Map each content type to the possible AI uses and decide which uses the brand would license, restrict or refuse. Keep training permission separate from citation, retrieval and recommendation rights.
Review every offer like a commercial data agreement. Record the buyer, crawler identity, definition of use, price, permitted downstream activity, reporting frequency and exit terms. Ask how disputes, missed reports and corrected URLs are handled. Do not accept a low headline price without estimating volume, cannibalised traffic, subscription impact and the operational cost of maintaining the content.
Build a reconciliation process before using the numbers in planning. Compare Pay Per Use records with crawler logs, referral data, citation monitoring and first-party conversions. The datasets will not match perfectly because they measure different stages. The useful question is whether the reported use is plausible, valuable and incremental enough to justify continued access.
Treat the beta as a negotiation and measurement exercise. It may become a scalable alternative to one-off licensing deals, but it is not proof that AI companies will pay sustainable rates or report every valuable use. Brands that define their rights, content economics and evidence requirements now will be better placed to accept good offers and reject weak ones.
News date: 30 September 2026. Editorial review: 1 October 2026.