A Peec AI researcher has identified in ChatGPT's news engines a field identifying twelve partner publishers whose content arrives pre-summary and passes before the rest of the web.
David Konitzny, a researcher at Peec AI in Berlin, examined the configuration of three news engines used by ChatGPT. In one of them, internally called "labrador", a "provider" field identifies publishers who have signed a licensing agreement with OpenAI: twelve groups spread over 43 domains, including Condé Nast, Hearst, Axel Springer, News Corp, Reuters and the Associated Press.
According to its analysis, the content of these partners does not enter ChatGPT in the form of raw snippets but in the form of summaries prepared in advance, which are placed in front of ordinary web results. Out of a sample of 1,000 cases, they occupied 70% of the top three positions, compared to 23% of positions seven to ten — a gap that is difficult to explain by relevance alone.
A challenge for licensing agreements
These findings, which come from an external researcher and not from OpenAI, shed light on what the agreements between the company and the press really monetize: structurally favored visibility. They also raise the question of the neutrality of responses for publishers who have remained outside these agreements. OpenAI has not commented.
Read the full article on the original source
Go to the source →