Cloudflare is helping ChatGPT find more pages. Found is not cited.

Cloud AI cites who it can see. On October 15, Cloudflare and OpenAI published the clearest statement yet of the step before that: the AI has to find the page first.

The post is Cloudflare and OpenAI: Making AI discovery better for everyone on the Cloudflare blog (HTTP 200 this fire). It's written by Matthew Conroy and Jack Galilee of Cloudflare and Henry Aspegren and James Balfour of OpenAI. Its first line is the whole thesis of GEO in one sentence:

"Optimizing your site for AI is of little use if AI can't find it."

Source: Cloudflare Blog, Cloudflare and OpenAI: Making AI discovery better for everyone, October 15, 2026.

The headline result is about discovery. It says nothing about citations. I want to keep those two apart, because the trade-press version of this story will blur them by Friday.

The short version

• What happened. A research pilot announced July 8 feeds signals from Cloudflare's network to OpenAI's search crawler. The companies say the pilot, "which used a participating cohort of ~1% of Cloudflare's total network, has enabled OpenAI to discover and index hundreds of millions of new web pages."
• What kind of pages. Their examples: the manual for an exact washing-machine model, current passport requirements and fees, a restaurant's latest menu and opening hours. Their line: these pages "can be valuable even if few other sites link to them."
• What it is not. "OpenAI then independently evaluates and ranks how those pages appear." Being indexed is not being chosen.
• Access still rules. "A signal about a page doesn't grant permission to access it." Your robots.txt and your firewall still decide.
• What wasn't reported. As Search Engine Journal noted today, the post doesn't say how often the new pages were cited in ChatGPT or how many visits they sent.
• Monitoring is not execution. Knowing ChatGPT's index got bigger doesn't put your page in it. Shipping a findable, distinct, crawlable page does.

What Cloudflare and OpenAI actually said

Claim	Where (opened this fire)
Pilot launched July 2026; announced as a "first-of-its-kind research pilot" using "content freshness, traffic quality, and actual page changes"	Cloudflare press release, July 8, 2026
Cohort of ~1% of Cloudflare's total network; OpenAI discovered and indexed "hundreds of millions of new web pages"	Cloudflare blog, Oct 15, 2026
OpenAI "independently evaluates and ranks" the pages	Cloudflare blog, Oct 15
Signals filter out URLs that don't exist, sites that "redirect almost any address to their homepage," and "copies of the same content"	Cloudflare blog, Oct 15
Change signals tell the crawler "when a page is worth another visit, and when it isn't"	Cloudflare blog, Oct 15
Testing page-level flags for harmful pages, e.g. "a fake login form planted by an attacker," so one page is flagged "rather than the whole domain"	Cloudflare blog, Oct 15
"Finding new pages can initially increase crawling as those pages are evaluated and indexed."	Cloudflare blog, Oct 15
Uses "metadata about the web for search and discovery, rather than sharing page content for model training"	Cloudflare blog, Oct 15
Next phase: "improve the selection of pages, test faster and more reliable change signals, and measure what happens after content is discovered"	Cloudflare blog, Oct 15
Cloudflare says more than 20% of the web sits behind its network	Cloudflare press release, July 8 and Cloudflare blog, July 1
Cloudflare says its data "suggests that more than 50% of crawl traffic from good bots goes to re-fetching pages that haven't changed"	Cloudflare blog, July 1
July 1: "We plan to make the capability broadly available later this year." The Oct 15 post gives no date.	Cloudflare blog, July 1; Oct 15

Search Engine Journal's write-up (ChatGPT Search Can Draw On More Pages After A Cloudflare Pilot, Matt G. Southern, October 20, 2026, HTTP 200 this fire) adds the right skepticism. Neither announcement explains who participated or whether the new pages are on those participants' sites. The Oct 15 post also gives no figures for crawl volume or server load.

Why this matters for a business that isn't a publisher

Look at the examples again: a manual, a requirements-and-fees page, a menu and hours. Those aren't viral articles. They're the boring, specific pages that answer one buyer's exact question, and they usually have almost no backlinks.

For years the honest advice was "that page probably won't rank because nobody links to it." This pilot is OpenAI and Cloudflare saying, in public, that they're building discovery that doesn't depend only on links. That's a real change in the conditions. It isn't a promise about outcomes.

The operator read is simple. If AI search is getting better at finding specific pages, the businesses that win are the ones that have the specific page. The page has to be live, crawlable, distinct, and current. If the answer to "what are your hours on holidays" or "do you install model X" lives only in a PDF, a Facebook post, or a staff member's head, better discovery has nothing to find.

The checklist: make your pages worth finding

1. Open both doors: robots.txt and the network. OpenAI's crawler documentation (HTTP 200 this fire) says: "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links." It recommends allowing OAI-SearchBot in robots.txt "and allowing requests from our published IP ranges." It also says it "can take ~24 hours from a site's robots.txt update for our systems to adjust." Search and training are separate settings. You can allow OAI-SearchBot and disallow GPTBot, and OpenAI gives that exact example. If you're on Cloudflare, AI Crawl Control (HTTP 200 this fire, "Available on all plans") is where you see which AI services hit the site and allow or block each one. A correct robots.txt still fails if a firewall rule blocks the crawler.

2. Give each answer its own page. Hours, holiday hours, service area, each service, install or care guides, warranty, policies, requirements. One question, one URL, one clear answer. Cloudflare's examples are exactly this kind of page. Link those pages from your own navigation and list them in your sitemap. That is my inference about how to be findable, not something Cloudflare or OpenAI prescribed.

3. Stop making junk look like pages. Cloudflare names three patterns its signals filter out: URLs that don't exist, sites that redirect almost any address to the homepage, and duplicate copies of the same content. If your site sends every missing URL to the homepage, or serves the same page under five addresses, you're producing the noise these systems are built to throw away. Return a real 404 or 410 for pages that are gone, and point duplicates at one canonical page. Google's own HTTP status documentation (HTTP 200 this fire) says that when a 200 page looks like an error or an empty page, Search Console reports a "soft 404." Neither company said clean status codes improve ChatGPT ranking. The point is to keep your real pages from looking like noise.

4. Make "changed" mean changed. The pilot's change signals come from Cloudflare's network, not from your sitemap. The habit is the same everywhere, though. Google's sitemap guide (HTTP 200 this fire) says Google uses <lastmod> "if it's consistently and verifiably … accurate," and that it should reflect "the last significant update to the page." Bing's sitemaps post for AI search (July 31, 2025; HTTP 200 this fire) says to avoid setting lastmod to the sitemap generation time "unless the content on that URL was actually updated." When hours, prices, or requirements change, update the page that day and let the date mean something.

5. Don't let one bad page become your page. Cloudflare and OpenAI are testing page-level flags for compromised pages, like a fake login form planted on a legitimate site. That's better than punishing the whole domain, but you still don't want the flag. Keep the CMS, plugins, and forms patched, and remove pages you didn't publish.

6. Prove it in two separate columns. Column one: is OAI-SearchBot fetching your important URLs? Check server logs or AI Crawl Control. Column two: is ChatGPT citing and sending visits? Check citations and referrals, labeled by surface and date. Crawled isn't indexed, indexed isn't cited, and cited isn't clicked. The pilot reported only discovery and indexing.

That's the proof loop: open access, one page per answer, clean signals, honest freshness, a clean site, and two columns of evidence.

What this is not (honesty block)

• Not independent evidence. The ~1% cohort and "hundreds of millions" of pages are the companies' own early results. No methodology, page list, or third-party check was published in anything opened this fire.
• Not a citation or traffic result. The Oct 15 post reports discovery and indexing only. It doesn't say how often the new pages were cited or how many visits they sent (SEJ confirms the same).
• Not a ranking signal. "OpenAI then independently evaluates and ranks how those pages appear." Nothing opened this fire says pilot sites rank or get cited more.
• Not a reason to move to Cloudflare. Nothing opened this fire says sites outside the pilot, or outside Cloudflare, are less visible in ChatGPT. OAI-SearchBot is OpenAI's crawler for the whole web.
• Not something you can join today. The posts don't name participants or say whether a given site is in the ~1% cohort. The Oct 15 post doesn't describe a sign-up for site owners. The July 1 post said broader availability was planned "later this year." The Oct 15 post gives no date.
• Not a load guarantee. No crawl-volume or server-load figures were published. The companies say new discovery "can initially increase crawling."
• Not a training deal. The companies say the research uses metadata for search and discovery, not page content for model training. That's their statement. We did not see the data flows.
• Not OpenAI's own announcement page. OpenAI's newsroom and help center returned 403 this fire. OpenAI's side here is the two OpenAI co-authors on the Cloudflare post plus OpenAI's crawler documentation. The post mentions GPT-6 and Intelligent UI. We did not open an OpenAI page on either and make no claim about them.
• Not Pay Per Crawl or Pay Per Use. Those are separate Cloudflare programs described in the July 1 post. This piece doesn't cover them.

Related New Reward Field Guides

• OpenAI's API is not ChatGPT's citations. Stop measuring the stand-in.
• ChatGPT UTM tags are missing. Direct traffic is not a visibility crash.
• Google is signing some agent requests. That is identity, not a citation.
• ChatGPT can now use tools on your page. Getting cited did not buy you that.
• Google is testing a top slot for the official page. Make sure it's yours.
• GEO measurement is noisy, so build a proof loop
• Why your business isn't showing up in ChatGPT, and how to fix it

Want the fast read first?

I'm not going to pretend a scan is a strategy. The free AI-visibility score is a quick check: email only, no credit card, a 0–100 read across the engines we track. It won't tell you whether your site was in a Cloudflare pilot. It will tell you whether AI can see you at all, and that still comes first.