ChatGPT search feature reproduces news articles without authorization
OpenAI launched a search feature for ChatGPT on October 31, 2024, that reproduces news content from websites without authorization from publishers. The feature shows article excerpts to paid subscribers while bypassing publishers' stated restrictions, and OpenAI has not paid Brazilian news outlets despite the technology functioning in their regions.
What happened
OpenAI launched a search feature for ChatGPT on October 31, 2024, available to paid subscribers for R$ 115 per month, that delivers exclusive information from news websites and bypasses subscription paywalls while providing a link to the source.
The feature reproduces article excerpts without authorization from publishers. Folha de S.Paulo explicitly prohibits OpenAI's GPTbot from accessing its content via a robots.txt file, but OpenAI used a different bot called OAI-SearchBot to circumvent this restriction.
OpenAI announced partnerships with American and European news outlets involving annual payments between US$ 1 million and 5 million, but made no similar arrangements with publishers in Latin America, Africa, or Asia, where the technology also functions.
Reported impact
Marcelo Rech, president of the Associação Nacional de Jornais (ANJ), characterized the service as plagiarism because the platform is copying proprietary content from news outlets without permission.
No payments have been made to Brazilian news outlets by OpenAI despite the search feature operating in the country. When tested on October 29, the search feature returned a detailed summary of an exclusive Folha de S.Paulo report, including content from an accompanying infographic.
Index analysis
OpenAI stated that sites can appear in the search results even if they restrict the use of their information for training generative AI models, and did not provide explanation in its announcement about how websites can prevent their content from being used.
Governance controls
Publishers can attempt to prevent access through robots.txt files, but OpenAI uses alternative bots to bypass these technical restrictions.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.