OpenAI appears to give licensed publishers a hidden head start in ChatGPT
Cartoon · AI-generated
An analysis of ChatGPT's search pipelines suggests content from OpenAI's publishing partners is supplied as pre-written summaries and placed ahead of ordinary web results.
OpenAI appears to be routing content from its licensed publishing partners through a separate path inside ChatGPT, giving those sources preference over the wider web, according to an analysis by David Konitzny, a GEO researcher at Berlin-based start-up Peec AI. The finding suggests that partner material is not treated in the same way as ordinary web pages, but instead enters the chatbot as a pre-generated summary, with partner entries placed ahead of other search results when ChatGPT assembles an answer.
Konitzny examined the configuration of three news engines used by ChatGPT and found that only one, codenamed "labrador", filled in a field called "provider". Through that clue, he identified 12 publishers across 43 domains, including Condé Nast, Hearst, Axel Springer, News Corp, Reuters and the Associated Press. The pipeline is not limited to hard news: according to the analysis, it also brings in content from sites such as Allrecipes, Tom's Guide and Welt, extending the arrangement into recipes, product reviews and buying guides.
The distinction appears to matter in practice. Whereas regular websites are returned to ChatGPT as snippets of text, partner content is delivered as a ready-made summary, and those summaries are shown before other web results. In a sample of 1,000 cases analysed by Konitzny, partner summaries accounted for 70% of entries in positions one to three in ranked lists, but only 23% of entries in positions seven to ten. That pattern points to a structural advantage for licensed publishers inside the chatbot, rather than a purely neutral ordering based on relevance alone.
The findings also sit alongside earlier indications that OpenAI intended to give publishing partners greater visibility in ChatGPT. In 2024, the company told partners they would receive a more prominent display, and a later Hamburg study reportedly found Springer titles appearing more often in ChatGPT's web interface than through its API. Even so, Konitzny's analysis suggests that prominence does not guarantee perfect handling: the report says partner content is still not always cited accurately.
Inspired by headline at: [1]
Sources by paragraph: - Paragraph 1: [2], [4] - Paragraph 2: [2], [4] - Paragraph 3: [2], [3] - Paragraph 4: [2], [5]
Source: Noah Wire Services · Original