Cloudflare's Monetization Gateway: pricing the web for agents

The three layers, and what each one meters

A publisher trying to price AI access on Cloudflare has three documented layers to work with. They meter different units, and the units are the story. The canonical home for product reference pages covering AI Crawl Control, Pay Per Crawl, and Monetization Gateway is the Cloudflare Developer Documentation.

AI Crawl Control: classification, no charge

According to Cloudflare, AI Crawl Control lets publishers see and sort AI crawlers by purpose rather than block them wholesale. Nothing is billed at this layer. What changes for a publisher: you stop arguing about whether to block AI and start looking at a breakdown of what each crawler does on your site.

Pay Per Crawl: the request becomes the unit

Cloudflare's Pay Per Crawl lets publishers price crawler access with Cloudflare as merchant of record, using HTTP 402. It is in private beta. What changes here: Cloudflare moves from gate to merchant of record. You name a price per crawl, the edge answers unpaid requests with a 402, and the crawler either pays or leaves.

Monetization Gateway: the resource becomes the unit

The Monetization Gateway, published 1 July 2026, lets Cloudflare customers charge for any protected asset behind the edge: pages, datasets, APIs, MCP tools, settling in stablecoins over x402. The metering unit stops being the visit and becomes the thing itself. A dataset can be priced without a page wrapped around it. An MCP tool can be priced per call.

That progression, page to request to resource, is one move: metering migrating from the document to everything the edge can see.

The part you cannot verify yet

Two things circulate widely that the published Cloudflare post does not contain. The first is a product step commonly called Pay Per Use. The second is a 15 September default. Nobody can confirm either from the sources cited here. The Monetization Gateway announcement of 1 July 2026 does not contain the Pay Per Use announcement, the 15 September deadline, or the billion-402s figure. If you are building a pricing rollout around that date, find the primary source before you commit engineering time to it.

What the mixed-use crawler default changed, and for whom

Nobody can tell from the published sources what a 15 September default did, because the Cloudflare post I can check does not mention that date. What is documented is a shift in the default itself.

AI Crawl Control sorts crawlers by purpose rather than blocking them wholesale. Mixed-use crawlers, the ones that both index for search and train on the same fetch, previously forced a binary decision. Blocking them cost you referrals. Allowing them cost you content. Purpose sorting lets a publisher treat those two functions separately.

For whom that matters most: publishers whose traffic depends on search referrals while their content also feeds model training. The 2025 crawl-to-refer numbers show how unequal that trade was. For the week of 19 to 26 June 2025, Cloudflare measured crawl-to-referral ratios of nearly 71,000 to 1 for Anthropic, 1,600 to 1 for OpenAI, 202.4 to 1 for Perplexity, 40 to 1 for Microsoft and 9.4 to 1 for Google. Those are 2025 figures, and Cloudflare says they have improved by more than an order of magnitude since. Treat them as a baseline.

Why the edge is where this lands rather than the origin

Four reasons.

The edge sees the request first. According to W3Techs, Cloudflare sits in front of 25.8% of all websites as of the September 2026 survey. A 402 returned at the edge costs the origin nothing.

The edge already carries the volume. Cloudflare chief strategy officer Stephanie Cohen, speaking at Consensus in May 2026, put the network's output of 402 responses at over a billion a day, quoted by CoinDesk. That is a quoted remark, not a published Cloudflare statistic, and it should be read that way.

The edge can hold the billing relationship. x402 settlement needs someone to bill, reconcile and absorb disputes. Your origin server has no commercial relationship with a crawler. Cloudflare already has one with you and can extend it outward.

The origin cannot tell who is asking. It sees an address. The edge sees a declared user agent, a purpose signal and a request pattern. Pricing requires knowing which of those you are dealing with.

One protocol detail matters for cost modelling. Per the x402 protocol documentation, the protocol charges no fee of its own: payers cover network fees only. It is blockchain-agnostic, supporting EVM chains and Solana. The fee you pay is the network's, and it moves with the chain and the moment.

What a publisher should check before pricing anything

Five checks, in order.

1. Who actually arrives, and what they do with the content. The dated 2025 crawl-to-refer ratios above are the fastest way to see the spread between a crawler that sends traffic back and one that does not.
2. Whether the traffic is worth pricing at all. Artemis figures, cited by CoinDesk, put real daily x402 volume at around $28,000, with an average payment around $0.20 against roughly 131,000 daily transactions. A $0.20 average is the number that decides whether a metered paywall covers its own overhead for you.
3. Whether the trend is your friend. BlockEden.xyz reports daily x402 transactions falling from about 731,000 in December 2025 to about 57,000 in February 2026, a decline of over 92%. Coinbase's own cumulative count in late April 2026 was roughly 165 million transactions, about $50 million of volume and 69,000 active agents, reported by CoinDesk the week Agent.market launched. Cumulative growth and daily decline can both be true.
4. Who will govern the rails. The Linux Foundation announced the x402 Foundation on 2 April 2026 at the MCP Dev Summit North America in New York, with Coinbase contributing the protocol. The operational launch followed on 14 July 2026 with 40 member organisations. The 17 premier members are Adyen, AWS, American Express, Circle, Cloudflare, Coinbase, Fiserv, Google, Mastercard, Monad Foundation, MoonPay, Ripple, Shopify, Solana Foundation, Stellar Development Foundation, Stripe and Visa. Pricing decisions you make now get governed by that group later.
5. What is actually settling, and in what. According to TRM Labs, 198.9 million settlements have moved $52.7 million since May 2025, of which $25.62 million screens as genuine commerce and 0.6% to 7.5% of that plausibly agentic. 99.6% of value settled in USDC. Chainalysis reports payments of $1 or more rising from 49% of value in early 2025 to 95% by early 2026. Small payments are the pitch. Larger ones are the current reality.

The three layers are one product in three stages, and the stage you are pricing at determines whether you are selling access, requests or resources.

If you want the wider picture of how answer engines describe this category, where AI engines cite crypto covers it. The Kunzum research corpus holds the raw responses and scoring scripts under CC BY 4.0. If you would rather start with the method, the free playbook sets it out, and what an engagement involves explains the scoped version.