ChatGPT Doesn't Pull From Google. It Pulls From Bing.

Most businesses optimize AI visibility the same way they optimize Google search. Fix schema, build backlinks, improve page authority. The assumption is that the signals driving Google rankings also drive AI citations. For ChatGPT, that assumption is wrong -- and the error is specific enough to be fixable.

The Seer Interactive finding

In February 2025, Seer Interactive published a study of citations in SearchGPT -- the ChatGPT Search preview mode that has since been integrated into the full ChatGPT product. Their core finding: approximately 87% of the sources ChatGPT cited in answers overlapped with Bing's top-10 results for the same queries.

Search Engine Land covered the study under a headline that put the implication directly: "Bing, not Google, shapes which brands ChatGPT recommends."

We confirmed this as a real primary study -- not practitioner speculation -- in our session 35 research (2026-06-02, platform-citation-behaviors.md). A separate Zyppy signal-strength analysis of ChatGPT ranking factors corroborates it from a different angle: "traditional search rank" scores 9.4 out of 10 in signal strength, the second-most-important factor behind URL accessibility (9.5). Multiple independent analyses point in the same direction.

One caveat we carry: the Seer study examined SearchGPT in preview mode in early 2025, not the fully-integrated ChatGPT Search product. Whether 87% holds precisely in 2026 is untested. We rate the directional claim medium-high confidence -- the Bing-ChatGPT retrieval relationship is well-established; the exact percentage should be treated as indicative, not precise.

Why Google rank doesn't give you the same guarantee

The inverse is equally important: ranking well on Google does not reliably predict AI citation behavior on ChatGPT, Perplexity, Gemini, or Google's own AI surfaces.

Our June 2026 review of the 5W "GEO vs. SEO: The 2026 Venn Diagram" report -- built from 680 million citations across five major AI platforms (session 35, 2026-06-02, platform-citation-behaviors.md) -- documented a collapse in that overlap. Pages appearing in Google's top 10 used to account for roughly 70% of AI citations. As of May 2026, that figure is under 20%.

Three structural reasons for the collapse:

**Tone.** Google rewards persuasive, brand-forward writing that converts searchers. AI engines prefer neutral, factual prose. Content written to rank on Google often reads as promotional to AI citation layers and gets deprioritized.

**Format.** AI retrieval prefers comparison tables, definition-first paragraphs, step-by-step guides, and FAQ blocks. Standard long-form SEO content -- narrative structure optimized for keyword density -- does not produce the extractable units AI citation layers can work with efficiently.

**Source architecture.** AI platforms draw heavily from third-party directories, Wikipedia, and niche expert comparison sites. These are not the signals that drive Google PageRank. Strong domain authority does not predict presence in the specific directories Perplexity reads or the comparison articles ChatGPT cites in category queries.

The result: Google ranking and AI citation have become structurally different optimization problems. For ChatGPT specifically, the relevant upstream signal is Bing -- which means businesses with the largest ChatGPT gaps are not necessarily the ones with the weakest Google presence.

What Bing's new citation data makes visible

On February 10-11, 2026, Microsoft released the AI Performance report inside Bing Webmaster Tools -- the first first-party, publisher-accessible tool from any AI provider showing how content gets used in AI-generated answers.

Our June 5, 2026 methodology rec on this tool (2026-06-05-bing-ai-performance-fix-plan-addendum.md, session 38) notes that it shows which URLs on your domain were used as grounding sources by Microsoft Copilot, the internal sub-queries Copilot generated when retrieving your content, and citation counts per URL over time.

One practitioner case documented in that session: a site with 44,469 Copilot grounding events but only 169 visible in-answer citations. Ninety-nine-point-six percent of that site's AI influence was happening at the retrieval layer, invisible to any standard analytics setup.

The tool is useful for understanding your Bing-level retrieval activity. One clarification we flag consistently to clients: Copilot and ChatGPT are not the same pipeline, even though both draw from Bing's web index. The AI Performance report shows Copilot's grounding behavior -- not ChatGPT's. Microsoft's AI surfaces represent roughly 10-20% of total AI search activity, and the report covers that slice only.

That said: pages that are not being retrieved by Copilot are likely underperforming across the Bing-correlated retrieval surface -- which includes the layer ChatGPT draws from. A weak Bing footprint affects both.

What to actually check

The practical implication of the 87% overlap is straightforward: Bing rankings are a valid diagnostic variable for ChatGPT citation weakness.

Most businesses focus exclusively on Google and ignore Bing entirely. In traditional SEO terms, that is defensible -- Bing's search market share is small. In AI citation terms, it is a meaningful blind spot. A business ranking in Google's top 5 for "accountant in Burlington" but absent from Bing's top 10 for the same query is underrepresented in the retrieval pool ChatGPT draws from.

Bing ranking responds to many of the same structural signals as Google: NAP consistency, schema markup, indexed backlinks, directory presence. But Microsoft-ecosystem properties carry additional weight -- LinkedIn, Bing Places for Business, and Bing Webmaster Tools verification all contribute signal that Google ranking does not reflect. A business with strong Google SEO that has never claimed Bing Places or verified Bing Webmaster Tools is leaving a ChatGPT-specific fix unaddressed.

On the content side: the format preferences that AI citation layers favor (comparison tables, definition-first sentences, FAQ structure) overlap significantly with what improves Bing retrieval. Aligning service pages to those patterns addresses both the Bing ranking and the AI citation surface at the same time.

What a cross-platform audit covers

When we run a Sourcepull audit, ChatGPT citation behavior is tested directly -- brand recognition, category recommendation, and service evaluation queries run against the live ChatGPT product, with each response classified for mention presence and accuracy. The Bing ranking correlation is upstream context; the audit tests the actual output.

If ChatGPT-specific gaps appear in the results, Bing ranking is one of the first upstream variables worth checking. The free Signal Check at sourcepull.ca returns a per-platform score across ChatGPT, Perplexity, Gemini, and Claude in under three minutes. If ChatGPT is your weakest surface, that is where to start before deciding how to diagnose why.