Early results published October 15, 2026 show a joint research pilot — running since July across about 1% of Cloudflare's network — fed OpenAI's crawler discovery signals that surfaced pages link-following never reached, including ones almost nothing links to.
2026-10-20 · by Moe Ameen
On October 15, 2026, Cloudflare and OpenAI published early results from a joint research pilot that expanded ChatGPT Search's index by hundreds of millions of web pages. The pilot — first announced July 8, 2026 and running since July across a cohort of about 1% of Cloudflare's network — feeds OpenAI's search crawler network-level discovery signals so it can find pages it never reached by following links alone. Cloudflare supplies the signals; OpenAI decides what to index and how to rank it.
The signals are metadata about the web, not page content. Cloudflare flags which pages are worth a crawler's time: it filters out URLs that don't exist, pages that redirect to a homepage, and duplicate content, and it feeds "change signals" that indicate when a page is actually worth revisiting. An experimental signal flags individual harmful pages — the companies' example is a fake login form on an otherwise legitimate site — rather than blacklisting a whole domain. Cloudflare says the research "uses metadata about the web for search and discovery, rather than sharing page content for model training," and that "a signal about a page doesn't grant permission to access it" — existing crawler-access rules still apply, and publishers can allow or block AI systems through Cloudflare's AI Crawl Control, available on every plan.
The headline for creators is a single line from the post: these newly found pages "can be valuable even if few other sites link to them." That is a break from the backlink-gated model of classic search discovery, and it echoes earlier evidence that [ChatGPT's index already surfaces small, unlinked sites](/news/chatgpt-search-indexes-small-sites) alongside big publishers. But being found is not being cited — OpenAI "independently evaluates and ranks pages by relevance and helpfulness," so an indexed page still has to earn its place in an answer. Cloudflare also cautioned that "finding new pages can initially increase crawling," and the companies have not yet reported citation rates or referral traffic. The next phase is better page selection, faster change detection, and measuring what discovery actually sends back to publishers.
The old game was earning backlinks so a crawler would bother to reach you. This pilot rewrites it: ChatGPT's index now finds pages almost nothing links to, as long as they're worth finding — and OpenAI then ranks on helpfulness, not authority. Two things suddenly outweigh link-building. How many genuinely useful, specific pages you publish, and how fresh you keep them. Most creators lose on both, because the real bottleneck was never finding an angle — it's producing on-brand content across surfaces on a cadence.
[Kompozy](/) is built for exactly that output. Point it at a topic and it generates net-new content across every surface an answer engine draws from: [Blog Articles](/glossary/hyperframes) and Email Newsletters for the long-form, citable pages, plus Carousels, Quote Graphics, Photo Posts, and [Persona Shorts](/glossary/persona-shorts) avatar video for the social footprint — all held to one voice by the [Persona Brief](/glossary/persona-brief) and shipped on [Autopilot](/glossary/autopilot) across the eight social platforms plus blog and email behind a per-post [review gate](/glossary/quality-gates). Because the engine produces and schedules on a recurring cadence, the pages it publishes keep changing — which is precisely the "change signal" Cloudflare now feeds the crawler. When the index rewards breadth, freshness, and specificity over backlinks, the win goes to whoever can reliably produce all three. For tuning an individual page to earn the citation once it's found, see [how to get your content cited by ChatGPT](/how-to/get-your-content-cited-by-chatgpt).
Running since July 2026 across about 1% of Cloudflare's network, the pilot had Cloudflare supply network-level discovery signals that helped OpenAI's search crawler find pages it had not reached by following links. OpenAI said on October 15, 2026 that it discovered and indexed hundreds of millions of new web pages. OpenAI decides independently what to index and how to rank it.
No. Discovery and citation are separate steps. OpenAI independently evaluates and ranks pages by relevance and helpfulness after they are found, so an indexed page still has to be the clearest, most useful answer to earn a citation. The expanded pilot widens who is eligible to appear; it does not guarantee anyone a place in an answer.
Per Cloudflare, the research uses metadata about the web for search and discovery, not page content for model training, and a discovery signal does not grant a crawler permission to access a page. Existing crawler-access rules still apply, and you can allow or block AI systems through Cloudflare's AI Crawl Control on any plan.
Less than before. Cloudflare's post says newly discovered pages "can be valuable even if few other sites link to them," so a thin-link or brand-new page can now enter the index. The emphasis shifts from link authority toward publishing useful, specific, frequently updated pages — the kind of breadth and freshness an engine like Kompozy is built to produce across blog, email, and social.