Cloudflare and OpenAI: network signals brought hundreds of millions of new pages to ChatGPT search
In a post dated October 15, 2026 and co-written with OpenAI, Cloudflare published the first results of a joint research pilot launched in July. Run with a participating cohort of roughly 1% of Cloudflare's network, the pilot helped OpenAI find hundreds of millions of new web pages and add them to its index. How those pages rank remains OpenAI's call.
First results from a pilot that began in July
The post appeared on the Cloudflare Blog under four bylines: Matthew Conroy and Jack Galilee, plus Henry Aspegren and James Balfour from OpenAI. It describes a research pilot the two companies started in July around one question: could signals gathered across Cloudflare's network help ChatGPT users reach fresh content more efficiently? The authors say the early answer is yes. The pilot pulled pages that OpenAI's search index had never seen into that index, widening the pool of sources ChatGPT can draw on.
The post also gives a sense of scale. The participating cohort made up about 1% of Cloudflare's total network, and through it OpenAI discovered and indexed hundreds of millions of new web pages. The authors frame the stakes in their opening line: "Optimizing your site for AI is of little use if AI can’t find it." In other words, tuning content for AI systems pays off only once those systems can locate the page.
Cloudflare supplies the discovery signal, OpenAI makes the ranking call
The split of roles is clear. Cloudflare's network produces signals that help OpenAI's search crawler find pages. OpenAI then judges, independently, where and how those pages show up, weighing relevance and usefulness. The signal puts a page in front of the crawler; its placement in results is OpenAI's decision. Along the way the team catalogued common sources of wasted crawling: bots repeatedly requesting files that do not exist, sites that bounce nearly every URL to the homepage, and addresses that resolve to duplicates of the same content. Spotting those patterns early lets needless work be dropped before any fetch happens.
For freshness, change signals tell a crawler whether a page merits a return visit. The post notes that a crawler re-downloading an unchanged page adds load for the site owner with nothing gained. On the safety side, signals still in testing aim to mark a single compromised page on an otherwise legitimate site, for instance a fake login form an attacker slipped in, without marking the entire domain, so the site owner is not penalized; early results suggest they can help pick out pages that belong outside an index. The authors also flag a trade-off: discovering new pages can raise crawling at first, while those pages are evaluated and indexed.
In AI search, visibility starts with being found
Visibility in AI search now depends not only on content quality but on whether a crawler finds the page in the first place. The post's examples are everyday ones: the manual for a specific washing machine model, current passport application requirements, a restaurant's latest menu and opening hours. According to the authors, pages like these can matter even when few other sites link to them. Per the post, Cloudflare's signals can help OpenAI's crawler find more useful pages to surface to ChatGPT users. The post also ties the work to GPT-6 and Intelligent UI, which it describes as making ChatGPT more visual and interactive.
The second lesson is technical SEO hygiene. Requests for missing files, blanket redirects to the homepage and duplicate content are treated here as wasted crawling and filtered out. On control, the authors are explicit: "A signal about a page doesn’t grant permission to access it." For search and discovery, the research works with web metadata and does not share page content for model training. Site owners can monitor and manage AI access through Cloudflare's AI Crawl Control. Next, the partners plan to refine which pages get selected, test quicker and more dependable change signals, and measure outcomes once content has been discovered.
The source does not mention Türkiye, so this section is commentary
The Cloudflare post, the only source for this story, contains no information about Türkiye. What follows is therefore not drawn from the source but is explicitly UNALSOFT commentary. The unknowns deserve to be listed one by one. The post does not say which customers or domains were in the 1% cohort, whether sites opted in or were chosen by Cloudflare, when or how site owners could join more broadly, how many clicks or visits participating sites received from ChatGPT, or anything specific to sites in Türkiye.
Our reading: we cannot tell whether any site in Türkiye running behind Cloudflare was part of this pilot, but the problems the post labels as wasted crawling can be fixed on any site regardless. Answering correctly for pages that do not exist, not redirecting every address to the homepage and cutting duplicate content all help a page be read properly by whichever crawler arrives. We covered a different angle, the access rules for AI bots, on August 16; that article is independent of this pilot.
UNALSOFT's take
We read this less as a partnership announcement and more as evidence of how much site architecture weighs in AI discovery. The pilot shows that crawl effort spent on missing files, blanket redirects to the homepage and duplicate URLs can be filtered out before a fetch, and the post reminds readers that pages with few inbound links can still be valuable. The practical takeaway is simple: clean up broken URLs, set redirects up with intent, keep one clear address per piece of content, and make real changes to a page visible. In our web design work we treat that structure as part of the design itself. Because the source carries no data about Türkiye, this article promises no local outcome.
Sources
Cloudflare Blog, Cloudflare and OpenAI: making AI discovery better for everyone
Does your site give AI crawlers a clean path?
A short conversation is enough to review broken URLs, redirects and duplicate content together.