Peec AI researcher David Konitzny says he identified 12 providers spanning 43 domains in ChatGPT’s partner-content pipeline. In the material he observed, partner pages arrived as pre-generated summaries, ordinary pages as content chunks, and partner entries appeared earlier in the prompt. Across 1,000 summaries of ranked lists, items ranked 1–3 were named 70% of the time versus 23% for positions 7–10. This is one researcher’s observation, not OpenAI’s complete search specification or a measurement of bias in every final answer. It supports checking original lists and omitted alternatives when comparing recommendations.