TLDR;
Cloudflare and OpenAI say network signals from roughly 1% of Cloudflare’s network helped OpenAI discover and index hundreds of millions of additional pages for ChatGPT Search. The pilot uses metadata about discovery, changes and safety, rather than handing page content over for model training. Brands may become discoverable faster, but indexing still does not guarantee a citation, visit or sale.
What happened
Cloudflare and OpenAI published the first results from a research pilot launched in July. Cloudflare’s network sends discovery signals that can point OpenAI’s search crawler towards useful pages, including product manuals, local information and other deep URLs that ordinary link-following crawlers may miss. OpenAI still decides whether and how a page ranks in ChatGPT Search.
Cloudflare says the participating cohort covered about 1% of its network and helped OpenAI discover and index hundreds of millions of new pages. The pilot also tests filtering for dead URLs, homepage redirects and duplicate content before the crawler spends resources fetching them. Change signals can indicate when a known page is worth revisiting.
The companies are also testing page-level safety signals, such as identifying a fake login page on an otherwise legitimate domain. Cloudflare says early results suggest these signals can keep a harmful page out of the index without treating the entire site as unsafe.
The boundary matters. Cloudflare says it shares metadata about the web for search and discovery, not page content for training. A discovery signal does not grant access, so robots.txt, Cloudflare controls and the origin response still decide whether OpenAI can fetch the page. The next phase will examine whether discovery produces useful business outcomes.
Why it matters
This pilot changes the discovery layer rather than the ranking algorithm. A useful enterprise page can be difficult to find because it has few links, sits deep in a catalogue or changes outside a conventional publishing workflow. Network-level signals could put that URL into OpenAI’s crawl queue sooner, which is valuable for manuals, availability pages, branch information and frequently updated policies.
It also weakens the assumption that links and sitemaps are the only routes into an AI search index. That does not make either obsolete. OpenAI still evaluates relevance and usefulness, while Cloudflare’s signal mainly helps it decide where to look and when to return. Clear internal linking, accurate sitemaps and stable canonical URLs remain useful controls brands can inspect themselves.
The commercial result is unresolved. Independent coverage highlighted that being indexed is not the same as receiving traffic. ChatGPT may use a page to answer a question, cite it, send a visit or do none of those visibly. Cloudflare has not yet published evidence that these newly discovered pages created leads, sales or publisher revenue.
There is also a governance issue. Security, infrastructure and content teams may each control a different part of access. A marketing team can improve a page while a CDN rule blocks OAI-SearchBot, or permit the search crawler while intending to restrict training. Cloudflare documents separate OpenAI identities for search, user-triggered fetching and model crawling, so policies should reflect the intended use.
How your brand can benefit / be affected
Start with pages where freshness or depth creates a real discovery problem. Product support documents, store details, regulatory updates and inventory-linked information are stronger candidates than a homepage already crawled frequently. Confirm that each page has a stable URL, a clear purpose and enough visible text to answer the customer’s question accurately.
Audit access at the edge and origin. Test OAI-SearchBot separately from GPTBot and ChatGPT-User, then record the response code, rendered content and applicable robots rules. Do not infer crawler access from a dashboard toggle alone. A policy should say which bot can access which content and for what business reason.
Measure three stages separately: discovery, appearance and outcome. Server logs can show crawler activity, controlled ChatGPT checks can show whether pages are cited or represented, and analytics can show referred visits and conversions. A rise in crawl requests is not proof that the content has influenced customers.
Finally, keep expectations proportionate. This is a research pilot with provider-reported scale, not a public ranking programme or a promise of referrals. Treat it as evidence that technical discoverability is changing. Brands that maintain accurate deep content and explicit access policies are better placed to benefit, while still demanding proof that broader indexing creates commercial value.
News date: 15 October 2026. Editorial review: 19 October 2026.