Cloudflare reports that daily requests from AI agents on its network grew by more than 1,700% over the past year. The figure points to a growing machine audience, but it does not tell a publisher how many readers, referrals or paying customers those requests produced. For SEO and editorial teams, measuring that difference is the practical challenge.
In its September 30, 2026 post, Cloudflare also reports human traffic falling by as much as 40% in under a year across some heavily crawled categories, including retail, software, IT services and financial services. Separately, it says requests associated with AI training, according to crawlers’ declared purpose, rose from 22% of observed crawler requests in spring 2025 to 52% in June 2026.
Read the figures with their denominators
These are Cloudflare’s observations and classifications. The agent-growth figure describes requests on its network. The training percentage concerns crawler requests, and the human decline concerns selected categories. They cannot be combined into a single measure of web-wide audience replacement.
The reported patterns also do not establish how much of a particular website’s traffic decline was caused by AI. A publisher would need its own evidence to assess that question. A large increase in automated requests can coexist with very different outcomes for individual sites.
Separate fetching from business outcomes
A useful measurement framework begins with distinct questions. Who requested the content? Was the request allowed and successfully served? Did it lead to a human visit, a qualified enquiry or a transaction? Each question measures a different stage, and a rise at the first stage does not establish growth at the last.
Consider a comparison page fetched repeatedly during automated research. Its request count could increase while visits to the publisher remain unchanged. Alternatively, an agent’s research might eventually lead a user to a product or service. Without evidence connecting those events, the publisher should treat the commercial outcome as unknown.
In our analysis, teams should review automated demand alongside human acquisition and revenue, while keeping the measures separate. That can help reveal whether heavily requested pages also receive valuable referrals, or mainly incur additional delivery costs. It also avoids presenting crawler activity as an improvement in reader engagement.
SEO now needs an access decision as well as a visibility goal
The question for a publisher is which uses of its material support its business. A site may value discoverability while placing different conditions on bulk collection, recurring retrieval or commercial reuse. The appropriate decision depends on its content, audience and ability to measure the results.
That makes access policy an editorial and commercial choice. Teams should assess a proposed change against a clear purpose, then examine what happened after implementation. Blanket assumptions that every agent is valuable, or that every automated request is harmful, make that evaluation less useful.
Keep the content useful to the person behind the task
For SEO teams, the report supports investigating how content is retrieved and used. It does not establish a new ranking factor or a guaranteed route to AI citations. Clear product details, current facts and identifiable sources remain practical ways to make a page easier to evaluate, whether the immediate reader is a person or software acting for one.
For editorial businesses, the harder issue is converting use into value. More requests can be a sign of demand, but the business case depends on attribution, access terms and outcomes. Cloudflare’s figures make those questions more urgent; a publisher’s own measurements must determine the answer.