OpenAI 合作夥伴內容以預先生成摘要進入 ChatGPT，一般網頁則以原始片段提供
OpenAI's partner content arrives in ChatGPT pre-summarized while the rest of the web gets raw scraps
David Konitzny 的分析指出，OpenAI 授權合作夥伴內容透過專用管道以預先生成的摘要進入 ChatGPT，而一般網頁則以原始文字片段提供。該管道涉及 12 家出版商、43 個網域，合作夥伴條目在提示詞中排於所有網頁結果之前。在分析的 1,000 個案例中，摘要提及排名第 1 至 3 位項目的比例為 70%，第 7 至 10 位則為 23%。
Oct 27, 2026
Content from OpenAI's licensing partners appears to flow into ChatGPT through a dedicated pipeline that gives it priority over the rest of the web. That's the finding of an analysis by David Konitzny, a GEO researcher at Berlin-based startup Peec AI. Of three configured news engines, only one called "labrador" populates a field named "provider." Through that field, Konitzny identified 12 publishers with 43 domains, including Condé Nast, Hearst, Axel Springer, News Corp, Reuters, and AP. Beyond news, the pipeline also feeds recipes, product reviews, and buying guides from sites like Allrecipes, Tom's Guide, and Welt into the chatbot.
ChatGPT doesn't fetch partner pages live. Instead, it receives them as pre-generated summaries, while regular web pages arrive as raw text snippets. In the prompt, partner entries are placed ahead of all web results. For ranked lists, the summaries name 70 percent of entries in positions 1 through 3 but only 23 percent of those in positions 7 through 10, according to 1,000 cases Konitzny analyzed.
OpenAI had promised partners more prominent placement in ChatGPT back in 2024, and a study from Hamburg later found Springer media appearing far more often in the web interface than through the API. But even partners don't always get cited correctly.
AI News Without the Hype – Curated by Humans
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.
來源：The Decoder · the-decoder.com