Cloudflare says daily traffic from AI agents across its network has increased by more than 1,700% over the past year, and the company is now testing new ways for websites to charge AI services and agents for using their content.
The company disclosed the traffic growth on September 30 as part of its broader push to adapt the web for software acting on behalf of users. Cloudflare says its network is now handling about 115 million HTTP requests per second on average, up from 63 million at the end of 2024.
Alongside that data, Cloudflare has moved two previously announced monetization projects, Pay Per Use and Monetization Gateway, into beta.
Pay Per Use is aimed largely at publishers whose material is incorporated into AI products without necessarily sending a human visitor back to the original website, letting participating AI companies offer payment for defined uses of content, such as including material in an AI-generated answer. Publishers can decide whether to accept each program.
Rather than charging whenever an AI crawler downloads a page, payment is tied to the later use reported by the AI company. Cloudflare then handles billing and pays participating publishers.
There is an important limitation where usage is self-reported by buyers. Cloudflare checks that reported uses correspond to publishers enrolled in the program, but the beta will help determine whether the reporting and payment model works at scale.
Cloudflare’s second system, Monetization Gateway, targets websites, APIs, datasets and tools where an automated request itself has value.
The service uses the HTTP 402 – Payment Required response and the x402 payment protocol to tell an agent that it must pay before receiving the requested resource. Sellers can set prices per request, query or token.
It’s initially limited to eligible buyers and sellers in the United States, with additional regions planned. Payments currently settle using the USDC stablecoin on the Base blockchain.
The payment products sit alongside other Cloudflare efforts to distinguish different kinds of automated traffic. In July, the company separated its controls for AI search, agents and model training, while a September update added a way for sites to remain available to search crawlers while signaling that their content should not be used for AI training.
Cloudflare has also developed technologies including Markdown for Agents and WebMCP to make websites easier for AI systems to read and interact with. If AI agents become a significant audience for the web, Cloudflare wants websites to have options beyond either giving those agents unrestricted access or blocking them completely. Whether enough AI companies and publishers agree on what individual uses are worth remains unresolved, which is why both payment systems are still being tested as betas.
Leave a Reply