Skip to main content
ChatGPT Shopping is where OpenAI’s own search index is most visible in production. Peec’s investigation (full study, Sep 2026) spotted an A/B experiment named “prefer-index-over-serp-v3” in ChatGPT’s server-side events in mid-August 2026, affecting 8% of chats at the time — OpenAI testing its own product index against scraped search results on traffic slices from single digits up to 20%. As of September 2, 2026, at least five separate shopping experiments were running: prefer-index-over-serp-v3, chatgpt-shopping-noamazon, shopping-index-q2qb, shopping-hqi-v2, and shopping-hq-v1. Whether “hqi” stands for “high quality index” is not confirmed; what is confirmed is that both prefer-index-over-serp-v3 and shopping-index-q2qb run on ChatGPT’s own index — see ChatGPT has its own search index. The ranking stack behind one experiment (“shopping-reduced-topn-v4b”), visible in the same events:
  • BM25 lexical component — narrows the field to 10 sources for ranking.
  • rerank400 — reranks a pool of 400 candidates.
  • prod400 — 400 products used as the input pool.
  • ann4096 — approximate nearest neighbor (vector search), run at two dimensions: 12,288 and 4,096.
Multiple ranking signals blended, then A/B tested against real traffic until the mix earns its place in production. The study’s advice for anyone selling products: watch how you show up in ChatGPT’s shopping results specifically — that is where the index push affects product selection today. Source: ChatGPT built its own search index, Peec (Tomek Rudzki), Sep 2026.