result_source field carried exactly four values — Labrador, Bright, Oxylabs, and SERP. Three are external scraping providers; Labrador is OpenAI’s own index.
Labrador is not one index but a family of vertical ones: general web, PDF, YouTube, news (with separate tiers for the last day, last 7 days, and older), arXiv, Wikipedia, local, finance (including a separate PDF index), legal, medical, shopping, and images. It stores what Google’s index stores — full page content, crawl date, and publication date. This is the same architecture Google spent two decades building: a general web index plus specialized verticals layered on top.
Corroborating evidence from the Sep 2026 investigation:
- OpenAI job postings. One asks for “designing and operating indexing systems, retrieval pipelines, and serving layers”; another describes running indexing services behind ChatGPT at exabyte scale with vector search. A research role targets dense, sparse, and hybrid retrieval — the setup for the Reciprocal Rank Fusion (RRF) merging that Peec researcher Metehan Yesilyurt documented first.
- Antitrust testimony. During Google’s antitrust trial, Nick Turley (head of ChatGPT) testified that OpenAI had “significant quality issues” with non-Google search API partners, that Google refused to license its index, and that OpenAI began building its own search index in 2023 — targeting 80% of queries from it by end of 2023. He also said that even with full access to Google’s data, it would take at least five years to determine whether answering 100% of queries from its own index is achievable.