Summary
On October 15, 2026, Cloudflare and OpenAI published early results from the AI-search research pilot they launched in July 2026. Using network signals from roughly 1% of Cloudflare's network, OpenAI's search crawler discovered and indexed hundreds of millions of additional pages, aided by discovery, filtering, change, and page-level safety signals.
What changed
Cloudflare and OpenAI shared results from their joint research pilot: Cloudflare network signals help OpenAI's search crawler find unknown pages (discovery), skip low-value fetches such as dead links, homepage redirects, and duplicate content (filtering), know when a page is worth revisiting (change signals), and flag individual harmful pages like fake login forms without penalizing the whole domain. The pilot used about 1% of Cloudflare's network and surfaced hundreds of millions of newly discovered and indexed pages. Signals carry metadata for discovery, not page content for model training; OpenAI independently evaluates and ranks what it finds.
Why it matters
This operationalizes a cooperative model for the agentic web in which a network provider supplies crawl-efficiency signals to an AI search system, cutting wasted fetches for publishers while improving freshness and coverage for AI answers. It is a notable reversal for Cloudflare, which spent the prior year shipping tools to block AI crawlers, and points toward negotiated, signal-based discovery rather than adversarial crawling or pure licensing.
Evidence excerpt
Cloudflare's network provides signals that can help OpenAI's search crawler discover more useful pages across the web. The pilot used roughly 1% of Cloudflare's network and led to hundreds of millions of new pages being discovered and indexed.