ChatGPT's in-house search index still leans on Google, Bing

Peec AI’s Tomek Rudzki reports that OpenAI has built its own retrieval index, internally named Labrador, spanning separate vertical indexes for web pages, PDFs, video, news and other content types. The finding draws on OpenAI job postings describing “exabyte scale” indexing infrastructure, Google antitrust testimony in which ChatGPT lead Nick Turley said OpenAI began building the index in 2023 after quality problems with non-Google data, and a since-removed “result_source” field visible in ChatGPT’s server-side events between May 21 and July 21, 2026. That field showed ChatGPT answers still drawing on Google, Microsoft’s Bing, and two third-party scraping services alongside Labrador, and Peec AI says ChatGPT had crawled 6 million of a billion test pages, at roughly 35,000 requests per hour, as of early September 2026.