boltTLDR
RESONEO's analysis of 1,200 ChatGPT answers reveals three distinct retrieval layers—OpenAI's own index (labrador), cached Google results, and live-fetched pages—each with different staleness and citation rules.
Search Engine Land publishes findings from RESONEO's dissection of ChatGPT's citation sources, identifying four internal pipelines (labrador, bright, oxylabs, and serp) after OpenAI removed the result_source field from its data stream. The research shows free users on Think mode retrieve primarily from OpenAI's proprietary index while paid Thinking relies mostly on scraped Google results, meaning identical queries can pull from fundamentally different corpora—a critical distinction for AI search visibility strategies.