ChatGPT has its own search index. Which of your questions still run through Google and Bing?

KEY TAKEAWAYS

1. ChatGPT has more than one search route. OpenAI's own index carried most primary searches in the one test that measured it, with outside providers used more for dated and news questions.
2. When ChatGPT switches route, the cited URLs change. Overlap fell by about 45% in repeated runs.
3. For evergreen questions, OAI-SearchBot access and plain HTML facts matter most. For dated and commercial questions, rankings in outside engines still count.
4. Indexing is the entry ticket. Only a small share of retrieved pages get cited.

On 15 October 2026, Cloudflare and OpenAI reported that a joint pilot had helped OpenAI find and index hundreds of millions of new pages for ChatGPT search. Microsoft's Web IQ documentation, last updated the day before, says that Web IQ, built on Bing, powers ChatGPT. Both statements are true at once. ChatGPT does not search one index. It searches several, and the route depends on the question.

That matters because most advice on getting into ChatGPT treats it as one system. Allow the bot, submit to Bing, wait. The evidence points to two separate jobs, and which one you need depends on the questions your buyers ask.

Where does ChatGPT get its search results?

From at least four places, by the best evidence available. Between 21 May and 21 July 2026, ChatGPT's network traffic carried a field called result_source with four values: labrador, bright, oxylabs and serp. Peec AI, Suganthan Mohanadasan and Chris Green all reported it.

Their reading is that Labrador is OpenAI's own index, and that Bright and Oxylabs are outside data providers that scrape search results. SERP looks like an open web baseline used mostly for news. Peec describes Labrador as a family of indexes, with separate parts for news, PDFs, local, shopping and more. OpenAI has not confirmed any of these names.

Bing sits alongside them. Microsoft says the infrastructure behind Web IQ powers Copilot and ChatGPT, in a June 2026 engineering post and again in the October documentation. Neither says how much ChatGPT relies on it. Peec saw Bing mainly in Deep Research traffic. Google is in the picture too. Search Engine Land, citing The Information, reported that OpenAI had used SerpApi's scraped Google results for real time topics.

How much of ChatGPT's searching runs on its own index?

Only one test has counted it. Chris Green ran 1,000 prompts up to 10 times each, for 9,946 completed runs, published on 5 July 2026. The primary source split was:

Treat this as a direction, not a market share. Green collected the data through Bright Data, did not state the account type or dates, and calls it a small test. Mohanadasan's tracker notes that the result_source label had gone from the stored payload by late July, and that Bing appeared for some accounts and not others. Peec saw an experiment named prefer-index-over-serp-v3 running in 8% of chats in mid August. The routing is still being tuned.

The crawl data agrees on direction. Botify's log analysis found OAI-SearchBot events up 3.5 times since August 2025. In Cloudflare Radar data compiled by Seomator, OAI-SearchBot's share of AI bot traffic rose from 1.7% in January 2026 to 3.6% in September. OpenAI's crawl is still small, though. Botify counted 887 million OpenAI events against 18.2 billion for Google in one 30 day window. Botify calls that about 4%. By our arithmetic it is 4.9%.

Which questions take which route?

This is the part most guides miss. Green found that evergreen educational and factual prompts mostly used Labrador. Prompts about current events, regulation, surveys, recent developments or explicit dates were more likely to use Bright, Oxylabs or SERP. Mohanadasan saw Bright play a larger role on commercial, shopping, finance, weather and local queries. Neither published figures for these splits, so read them as patterns.

The route changes the answer. When a prompt switched primary source between runs, which happened for 11.6% of prompts, URL overlap fell from 0.273 to 0.149. That is about 45% lower. A page can be cited on Monday and missing on Tuesday for a reason that has nothing to do with the page.

Some questions take no route at all. Mohanadasan saw ChatGPT classify some prompts as text only and skip search, even ones that sounded current. Indexing cannot help there. Only what the model already knows about you can.

Is Bing the way into ChatGPT or not?

Sources split on this. Older guides, such as Martech Zone's, say ChatGPT relies on Bing, and HubSpot's guide recommends submitting your sitemap to Bing because ChatGPT sometimes uses Bing's index. Microsoft's own documentation backs a link. On the other side, Peec says Bing visibility is not a proxy for ChatGPT, and Green's routing data puts OpenAI's index first.

Ranking studies support the second view. In Ahrefs' July 2025 test of 15,000 long tail prompts, 8.0% of ChatGPT's in-text citations ranked in Google's top 10 for the prompt, and 8.1% in Bing's. A preprint posted on 11 October 2026, reported by Relevant Audience, found Google's top five results held only 6.4% of the exact URLs GPT-5.5 Instant cited.

Both camps are right about different questions. Bing is one input, and on dated or commercial questions the outside route matters more. For evergreen questions, OpenAI's own crawl matters more. Submitting a sitemap with honest lastmod dates and using IndexNow costs little, as Bing's own guidance describes. Just do not read your Bing rank as your ChatGPT standing.

What did the Cloudflare pilot change?

It showed how OpenAI finds pages that few sites link to. The pilot began in July and covered about 1% of Cloudflare's network. Cloudflare sent signals about pages: filtering signals that flag URLs that do not exist or redirect elsewhere, change signals about whether a page is worth revisiting, and harm signals for pages such as a fake login form. OpenAI then evaluated and ranked the pages itself. Search Engine Journal noted that the examples given were pages like a washing machine user guide or a restaurant's menu and hours.

Three limits apply. No citation or traffic figures were published. The post does not say how sites joined. Tech Times describes participation as on by default with an opt out in AI Crawl Control, but Cloudflare's post does not say so, and we could not confirm it. And Cloudflare is clear that a signal about a page does not give permission to fetch it. Your robots.txt still decides.

The practical lesson holds for any site. Signals like these help crawlers skip pages that are dead or redirected. Real 404 and 410 codes, no catch-all redirects to the homepage, and one URL per piece of content all help.

Does being in the index get you cited?

No. It gets you considered. In Mohanadasan's sample of 57 conversations, ChatGPT retrieved 3,554 pages and cited 110, or 3.1%. Brands that ChatGPT wrote into its own search query reached the answer 68.9% of the time. Brands it only fetched reached it 2.1% of the time. One brand was fetched 66 times and never cited.

So index access is the floor. What lifts you off it is being the name ChatGPT already expects, which comes from press coverage, a clear brand entity and search presence. Those are pillars of digital authority for that reason. OpenAI's own line is plain: "Any public website can appear in ChatGPT search." Appearing is not the same as being chosen.

A worked example

This example is hypothetical. A regional accounting firm tracks 30 buyer questions. 20 are evergreen, such as how to register a new business. 10 are dated or commercial, such as this year's filing deadlines and typical fees.

Over five runs each, it is cited in 9 of 20 evergreen questions and 0 of 10 dated ones. Its logs show OAI-SearchBot fetching its guides, so the evergreen route works. The fee page loads its prices with JavaScript, and its deadline pages sit on page three of Google and Bing. The dated questions run through outside providers more often, so the firm is not in that pool.

The fix order follows the route. First, prices in plain HTML, since Mohanadasan saw ChatGPT fall back to third-party pages when prices sat behind JavaScript. Second, update the deadline pages and work on their rankings in Google and Bing. Third, coverage on the outside pages that do get cited for fee questions.

How do you test which route you are missing?

1. Check access. Allow OAI-SearchBot in robots.txt and at your CDN. OpenAI says changes take about 24 hours to apply to search. Our guide to AI crawler access covers the layers.
2. Read your logs. Confirm OAI-SearchBot fetches your key pages, verified against OpenAI's published IP ranges. ChatGPT-User visits do not decide search eligibility.
3. Check the HTML. Prices, specifications and dates should be in the page source, not loaded by script.
4. Clean up signals. Return real 404 or 410 codes, remove catch-all redirects and merge duplicate URLs.
5. Cover the outside route. Submit a sitemap to Bing with true lastmod dates, use IndexNow, and check where your dated and commercial pages rank.
6. Split and sample. Tag your questions as evergreen, dated, commercial or local. Run each several times, as set out in how to measure AI visibility, and record the cited domains.
7. Diagnose by type. Missing everywhere suggests access. Missing only on dated or commercial questions suggests outside rankings. Fetched but not cited suggests authority. Workspace users can also try offline web search, which HubSpot describes as a check against OpenAI's cached index.

What is still unknown?

Most of the routing evidence comes from May to July 2026, before GPT-6. SISTRIX saw ChatGPT's answer format change on 9 October, with linked pages per answer rising from 3.3 to 4.1. Whether routing changed with it is not known. The share of each route by question type has not been published by anyone. Rerun your own question set after each model change and compare matched questions only.

If you want this mapped for your brand across ChatGPT, Perplexity and Google, join the waitlist.

Questions

Does ChatGPT have its own search index?

Yes, by the evidence available. Researchers reading ChatGPT's network traffic between May and July 2026 found a field naming four result sources, and one of them, labelled Labrador, behaves like OpenAI's own index. OpenAI has not confirmed the name. Its October 2026 pilot with Cloudflare described pages being added to its index.

Does ChatGPT still use Bing or Google?

Partly. Microsoft's Web IQ documentation, updated 14 October 2026, says Web IQ powers ChatGPT, without saying how much. Two outside scraping providers also appear in ChatGPT's traffic. In the one test that counted, outside sources were the primary route for under 12% of primary searches.

How do I know if my page is in ChatGPT's search index?

There is no Search Console for ChatGPT. Check that OAI-SearchBot is allowed and appears in your server logs on the page. Then ask the question the page answers several times and record whether it is cited. Workspace accounts with offline web search can run a further check.

Will being in ChatGPT's index get my page cited?

Not on its own. In one researcher's sample, 3.1% of the pages ChatGPT retrieved were cited. Brands that ChatGPT wrote into its own search query reached the answer far more often than brands it only fetched. Indexing gets you into the pool. Reputation decides who is picked.

Sources

1. Cloudflare Blog, Cloudflare and OpenAI. Making AI discovery better for everyone blog.cloudflare.com
2. Chris Green, What can we learn from ChatGPT search sources? chrisgreenseo.substack.com
3. Search Engine Land, ChatGPT citations change when hidden search pipelines switch searchengineland.com
4. Peec AI, ChatGPT built its own search index peec.ai
5. Suganthan Mohanadasan, How ChatGPT picks sources. A live tracker suganthan.com
6. Microsoft Learn, Microsoft Web IQ overview learn.microsoft.com
7. Microsoft Command Line, Grounding system for the agentic web commandline.microsoft.com
8. OpenAI, Overview of OpenAI crawlers developers.openai.com
9. OpenAI Help Center, Publishers and developers FAQ help-lb.openai.com
10. SISTRIX, What changed in ChatGPT in October. GPT-6, UTM tracking and sources sistrix.com
11. Search Engine Journal, ChatGPT search can draw on more pages after a Cloudflare pilot searchenginejournal.com
12. Tech Times, ChatGPT search gained hundreds of millions of pages link crawlers never found techtimes.com
13. Botify, OpenAI has tripled their crawl of the web. An analysis of 7B+ log files botify.com
14. Seomator, Monthly AI crawler report. September 2026 seomator.com
15. Relevant Audience, ChatGPT citations. Google and API proxies miss most URLs relevantaudience.com
16. Ahrefs, Only 12% of AI cited URLs rank in Google's top 10 for the original prompt ahrefs.com
17. Search Engine Land, ChatGPT's answers came from Google Search after all. Report searchengineland.com
18. HubSpot, How to get indexed by ChatGPT blog.hubspot.com
19. Bing Webmaster Blog, Keeping content discoverable with sitemaps in AI powered search blogs.bing.com
20. Otterly.AI, OpenAI is building its own search index otterly.ai
21. Martech Zone, ChatGPT visibility and Bing IndexNow martech.zone

NEXT STEP

See where your brand stands.

Your Digital Authority Score, who AI recommends for your target questions, and the placements that close each gap. Your first report is free.

Get your free report →