Is Bing still a side door into ChatGPT? Less than when we wrote this.
We published this in August 2026 arguing that Bing coverage was a practical route into ChatGPT search. Evidence published since shows OpenAI serving from its own index on a growing share of traffic. The experiment is still worth running, for smaller reasons than we first gave.
This page originally argued that improving Bing coverage was a meaningful path into ChatGPT search, resting on research showing most ChatGPT citations matched Bing's top results. That research is now old, and OpenAI has been shown serving from a proprietary index on a growing share of queries. We have rewritten the page rather than quietly editing the claim. The original publication date is unchanged.
What changed since we published this
The strongest version of the Bing argument was always borrowed. It came from an analysis finding that around 87% of ChatGPT search citations matched Bing's top results. If that were still true, Bing rank would be a close proxy for ChatGPT visibility and the work would be obvious.
Between May and July 2026, server-side event data from ChatGPT surfaced a retrieval field naming an OpenAI index, and by early September there were live experiments testing that index against external search results on real traffic. OpenAI's own leadership testified in 2024 that the company began building a search index in 2023. Reporting also indicates ChatGPT draws on Yelp, TripAdvisor and other feeds alongside general web search.
Bing coverage is a practical side door into ChatGPT search, because ChatGPT search leans on Bing's index.
Bing coverage is one of several inputs, its share is shrinking, and nobody outside OpenAI can currently say by how much.
The original idea
Improving discovery in Bing creates another path through which current pages can be found, while allowing OpenAI's search crawler supports inclusion in ChatGPT search. The recommendation was to make both paths healthy rather than assume one controlled the other. That framing survives. The weighting inside it does not.
What the platforms actually document
OpenAI documents three separate crawlers, and they do different jobs. This is the part that has not changed and it remains the highest-value thing on this page.
| Crawler | What OpenAI says it does | Respects robots.txt |
|---|---|---|
OAI-SearchBot | Surfaces websites in search results in ChatGPT's search features | Yes |
GPTBot | Crawls content that may be used to train the foundation models | Yes |
ChatGPT-User | Handles user actions in ChatGPT and custom GPTs | No. OpenAI states that because these actions are user-initiated, robots.txt rules may not apply |
Search inclusion and training consent are therefore separate decisions. A site can allow the first and refuse the second, and four of the forty-four agency sites we surveyed in September had done exactly that. IndexNow, separately, documents a protocol for telling participating engines that a URL was added, updated or deleted. Neither system promises a ranking.
Why the premise weakened
Two things pull in the same direction.
- OpenAI is diversifying retrieval. Its own index is in live testing on a share of traffic, alongside external search and vendor feeds. Every point of share that moves to the proprietary index is a point where Bing rank stops being a proxy.
- Bing is small where our readers are. Bing holds roughly 4.5% of global search and about 1.45% in India. Work justified only by Bing's own traffic is hard to defend on that number alone.
We cannot quantify the first. The figure that would settle it, the current share of ChatGPT citations traceable to Bing, is not published by anyone with access to it.
What still holds
Three reasons survive, and they are cheaper than the original pitch implied.
- Crawler access is still the gate. If
OAI-SearchBotcannot reach the page, no amount of index diversification helps. This is documented, checkable in an afternoon, and independent of which index is used. - Bing Webmaster Tools is free diagnostic data. Index coverage in a second engine tells you whether a page is genuinely retrievable or just indexed by Google, and it costs nothing.
- IndexNow is a one-time setup. A single submission reaches Bing and several other participating engines. It is a notification mechanism, not an influence button, and it should be sold as one.
The honest version is smaller than the version that sells. Bing work is cheap diagnostics and an open gate, not a lever on ChatGPT.
How to run the experiment
- Verify the domain in Bing Webmaster Tools and record current index coverage as a dated baseline.
- Confirm robots.txt and your CDN allow Bingbot and
OAI-SearchBot. A CDN can challenge a bot that robots.txt permits, and that failure is invisible in robots.txt alone. - Implement IndexNow and submit genuinely changed URLs. Submitting unchanged pages teaches you nothing and wastes the signal.
- Improve one page that answers a real buyer question directly, with the answer near the top rather than buried.
- Record the relevant ChatGPT answers before the change using a fixed prompt set across fresh sessions, not a single look.
ChatGPT has been observed serving cached copies of pages. A fix is done when a re-crawl shows up in your server logs, not when you publish it. Checking the live answer too early will tell you the work failed when it has simply not been re-read.
What result to measure
Four outcomes, recorded separately, because they fail for different reasons. A page becoming indexed is a technical result. Being cited is a visibility result. The business being described accurately is an answer-quality result. An enquiry is the only commercial one.
Save index coverage, the prompt-set answers, the cited sources, and referral baseline.
Confirm crawling and indexing after the page change, from logs where you have them.
Re-run the exact prompt set in fresh sessions and compare rates, not single answers.
Keep, revise or stop based on observed movement. Stopping is a legitimate outcome.
Limits
- We have not run this test with a holdout, and a single site cannot separate this work from everything else changing around it.
- The evidence for OpenAI's own index is reconstructed from observed retrieval fields, job postings and testimony rather than an OpenAI announcement of scope. The direction is well supported. The share is not public.
- The 87% Bing-match figure is real but dated, and we would not quote it as current.
- Crawler behaviour and index composition here describe 23 September 2026. These systems change monthly.
- We are an agency that sells visibility work, and this revision argues that one of the things people sell is worth less than claimed. That cuts against our own category, which is the direction a caveat should run.
OpenAI: crawler documentation
Peec AI: evidence for OpenAI's own index, a vendor selling AI visibility tracking
Seer Interactive: the 87% Bing match, now dated
IndexNow protocol documentation
See what is limiting your visibility.
Send the URL and tell us what you sell. You will get the first issue worth fixing, with the reason behind it.
Request an audit ↗Free · No call required