ChatGPT Search Citations Lean on Bing, Not a Live Google Scrape

What the evidence says

The practical question for anyone chasing visibility in AI answers is which web index ChatGPT actually looks at when it goes online. The strongest available answer still points to Bing.

The most widely cited data comes from a Seer Interactive study of more than 500 citations generated by 100 questions. It compared what ChatGPT Search cited against Google's and Bing's results for the same queries. Bing matched 87% of ChatGPT's citations, mostly within its top 10. Google matched 56%, with a median rank of 17 and an average rank of 28 3. A second write-up frames the same figures more bluntly: when ChatGPT needs current information, it draws on pages Bing has already crawled rather than crawling the web from scratch, so pages missing from Bing's index are effectively invisible to it 4.

That reliance has a direct technical consequence. Practitioners are being told to confirm Bing indexation, allow OpenAI's crawlers in robots.txt, and use IndexNow to push new pages into Bing faster 34. For businesses whose customers increasingly ask ChatGPT instead of searching Google, the guidance is that a missing citation can mean a lost sale before the customer ever reaches the website 3.

The counter-signal from a Google update

Not everyone accepts a clean Bing-only story. A February 2026 study tracked 11 sites hit by a Google algorithm update and found that their ChatGPT citations fell 27.8%. That was a steeper drop than in Google's own AI Mode, which fell 23.8%, while Perplexity barely moved at -2.9% 5. The authors read this as strong evidence that ChatGPT scrapes Google live. They argue that static overlap snapshots, which they put at roughly 6–8% top-10 overlap between ChatGPT and Google, understate how much ChatGPT depends on Google rankings 5.

The two camps disagree on more than their conclusions. They also report Google overlap figures far apart, 56% versus single digits, which almost certainly reflects different methods: citation matching across a full results set in one case, strict top-10 overlap in the other 35. Those numbers can't be compared directly.

Why the Bing reading still holds

The Google-update finding deserves attention, but it does not prove live Google scraping. Eleven sites is a small sample. A Google penalty is also a judgment about content quality, and other ranking systems, Bing's included, may weigh many of the same signals. Sites Google demotes for thin or unhelpful content could lose ground in Bing-backed retrieval at the same time, or within the same period. That would show up as correlated drops without any direct pipe to Google's index. Perplexity's near-flat result fits a pipeline built on a genuinely separate index, which the frase analysis attributes to Brave 5.

The more defensible conclusion is that ChatGPT's retrieval runs mainly through Bing, as the direct citation-matching data shows 34. Google rankings work as a correlated proxy for quality, not as the source feed. The frase study's methodological point still stands, though. Measuring how sensitive citations are to rank changes, engine by engine, is a better diagnostic than a single overlap snapshot 5.

The wider landscape: indexes are not interchangeable

The debate matters because the industry tends to treat "web search" as one thing. A benchmark of 500 queries run through four search APIs against Google's live results found wide divergence. Tavily matched 60% of Google's top-10 URLs, Brave 47%, Perplexity's Search API 33%, and Exa 32% 1. Brave kept Google's ordering best, and its top result was a Google-ranked page 89% of the time 1. Builders of agents and RAG pipelines who assume their API returns "basically Google" are often working with a different slice of the web 1.

Independent indexes are also being positioned as a selling point. Brave runs its own crawler and ranking rather than wrapping Google or Bing, which frees it from their rate limits, terms, and outages, and it powers an answer service the company says handles 22 million answers a day 2. On Artificial Analysis's Search Index, Perplexity Search scored 80, Octen 77, and Parallel Advanced and Brave LLM Context tied at 75 2.

What it means for SEO

The takeaway is to diversify. Bing indexation is no longer optional for anyone targeting ChatGPT, and the technical fixes are cheap 34. Protecting Google rankings still pays off indirectly, because the quality signals Google rewards seem to carry over across engines 5. Perplexity and other Brave-style independent indexes form a third front that responds to neither Bing nor Google in a predictable way 15. AI search is splitting into several retrieval systems, and visibility will have to be tracked separately in each.