Cloudflare says the OpenAI discovery pilot indexed a one percent cohort
Cloudflare says the OpenAI discovery pilot covered about 1 percent of its network. A July figure of 20 percent describes the network, not that cohort.
The short version
About one percent. The post says the pilot began in July 2026, that hundreds of millions of new pages were indexed, and that the signal is metadata for search and discovery, not page content for model training. The July press release's figure, more than 20 percent of the web behind Cloudflare, describes the network, not the pilot cohort.
The October 15 results
Cloudflare's October 15 post, by Matthew Conroy, Jack Galilee, Henry Aspegren, and James Balfour, says a research pilot with OpenAI began in July 2026. The cohort was about one percent of Cloudflare's network. Hundreds of millions of new pages were indexed. OpenAI ranks on its own. The metadata is for search and discovery. The post says it is not page content for model training. The distributed-cloud desk is The document named in the sources is the record this page will keep. The document named in the sources is the record this page will keep.Distributed Cloud.
What a signal does not grant
The post says a signal does not grant access. AI Crawl Control remains the customer's control. A site that does not want to be crawled is not enrolled by the existence of the pilot. That is Cloudflare's sentence about its own switch. Privacy, for who holds a record, is Asked directly, Is page content used for training, on this post? The post says the metadata is for search and discovery, not page content for model training. GPT-6 and Intelligent UI are named as OpenAI product context, not as something this pilot measured.the key stays with the holder.
Product names that are context
GPT-6 and Intelligent UI appear as OpenAI product context. The post does not measure them. This page will not treat a product name in a discovery post as a benchmark of that product. Artificial intelligence is The short answer this page will stand on is this: About one percent. The post says the pilot began in July 2026, that hundreds of millions of new pages were indexed, and that the signal is metadata for search and discovery, not page content for model training. The July press release's figure, more than 20 percent of the web behind Cloudflare, describes the network, not the pilot cohort.Artificial Intelligence.
The July figure is the network
Cloudflare's July 8, 2026 press release says more than 20 percent of the web sits behind Cloudflare's network. It quotes Nick Ryder and Matthew Prince. Twenty percent is a claim about the network. One percent is the October post's claim about the pilot cohort. Multiplying them, or substituting one for the other, is how a headline overshoots. The paper is The document named in the sources is the record this page will keep. The document named in the sources is the record this page will keep.Futureweb.
Where The Next Web blurs the two
The Next Web's account of the pilot can be read as the July announcement: a Wednesday, 20 percent of the web, and feeding a crawler. That reading conflates the announcement with the October results. This page labels it as a conflation. The cohort size to cite for the results is the one percent in the October post. Software is The document named in the sources is the record this page will keep. The document named in the sources is the record this page will keep.Software.
Not the D1 control plane
Yesterday's brief on Cloudflare D1 is about a database control plane off the query path. This pilot is a discovery index. When one region is the control plane is the standing question, and it is not the dependency here. That brief is Asked directly, What does the post say was indexed? Hundreds of millions of new pages, from a cohort of about one percent of Cloudflare's network. OpenAI ranks independently. A signal does not grant access, and AI Crawl Control stays the customer's control.when one region is the control plane.
What a reader should quote
Quote October 15 for the cohort, the page count, the metadata limit, and the crawl control. Quote July 8 for the share of the web Cloudflare says sits behind its network, and for the start of the pilot as a press release. Quote The Next Web only with the warning that it may stack those two. Hardware is Asked directly, Does this brief depend on D1? No. A control-plane region is a different Cloudflare document, linked so the pilot is not mistaken for it.Hardware.
Hundreds of millions is the company's count
The page count is Cloudflare's. This paper did not recount an index. Hundreds of millions of new pages, from one percent of one network, is a large sample and still a sample. It is not the web. Open source is Asked directly, Who wrote the October 15 post? Matthew Conroy, Jack Galilee, Henry Aspegren, and James Balfour. It is Cloudflare's account of a pilot that began in July 2026. The document named in the sources is the record this page will keep. The document named in the sources is the record this page will keep.Open Source.
What would be a different story
A customer log showing a crawl that ignored AI Crawl Control would be a different story. A later post that changes the cohort size would replace the one percent. A headline that says 20 percent of the web was indexed would not match the October post. Energy is Asked directly, What is the 20 percent figure? The July 8 press release says more than 20 percent of the web sits behind Cloudflare's network. That is the network, not the pilot. The Next Web's account can be read as mixing the July announcement with the October results.Energy.
Sources
The reports this brief is filing. Futureweb did not republish them.
Questions
Who wrote the October 15 post?
Matthew Conroy, Jack Galilee, Henry Aspegren, and James Balfour. It is Cloudflare's account of a pilot that began in July 2026.
What does the post say was indexed?
Hundreds of millions of new pages, from a cohort of about one percent of Cloudflare's network. OpenAI ranks independently. A signal does not grant access, and AI Crawl Control stays the customer's control.
What is the 20 percent figure?
The July 8 press release says more than 20 percent of the web sits behind Cloudflare's network. That is the network, not the pilot. The Next Web's account can be read as mixing the July announcement with the October results.
Is page content used for training, on this post?
The post says the metadata is for search and discovery, not page content for model training. GPT-6 and Intelligent UI are named as OpenAI product context, not as something this pilot measured.
Does this brief depend on D1?
No. A control-plane region is a different Cloudflare document, linked so the pilot is not mistaken for it.