OpenAI partner content gets priority in ChatGPT, report says

OpenAI’s licensing partners appear to receive a privileged route into ChatGPT: their articles are supplied as pre-generated summaries, while much of the rest of the web arrives as raw text snippets, according to an analysis reported by The Decoder.

The report draws on research by David Konitzny, a GEO researcher at Berlin-based startup Peec AI. His analysis identified 12 publishers across 43 domains in one of ChatGPT’s news pipelines, including Condé Nast, Hearst, Axel Springer, News Corp, Reuters and the Associated Press. The pipeline also includes material outside traditional news, such as recipes, product reviews and buying guides.¹

The important distinction is how those sources reach the chatbot. ChatGPT does not necessarily fetch a partner page live when answering a query. Instead, partner content can arrive as a pre-generated summary, while non-partner pages are represented by raw snippets. The partner entries are then placed ahead of other web results in the prompt used by the system.¹

Konitzny’s analysis found that the advantage becomes particularly visible in ranked lists. Across 1,000 cases, partner summaries accounted for 70 per cent of entries appearing in positions one to three, compared with 23 per cent of entries in positions seven to 10. That does not prove every ranking is manually manipulated, but it does suggest that the format and ordering of the supplied material can shape what users see.¹

The arrangement also raises a reliability question. OpenAI has previously promised licensing partners more prominent placement in ChatGPT, but earlier research found that some partner publishers appeared far more often through the web interface than through the API. The Decoder also points to evidence that even partner sources are not always cited correctly.¹

For readers, the practical lesson is simple: a result appearing near the top of ChatGPT is not automatically an independent measure of quality or relevance. It may reflect a publisher relationship, the way the material was summarised, the surface being used or the ranking system’s own interpretation of the query. Those distinctions matter more as ChatGPT becomes a front door to news and specialist information.

Our opinion

This is less a story about OpenAI secretly picking winners than about the quiet power of plumbing. Give one group of publishers clean summaries and a preferred place in the prompt, while everyone else arrives as unprocessed scraps, and the ranking contest is no longer happening on equal terms. OpenAI may have sensible reasons for using licensed feeds, but users deserve to know when an answer’s apparent authority is partly a distribution deal. The more ChatGPT becomes a news interface, the more clearly that distinction needs to be exposed.