ChatGPT Search Faces Scrutiny: Study Reveals High Inaccuracy in News Source Citations

December 3, 2024
ChatGPT Search Faces Scrutiny: Study Reveals High Inaccuracy in News Source Citations
  • OpenAI has introduced ChatGPT Search, marking a significant advancement in AI-powered search technology, but recent findings raise serious concerns about its accuracy.

  • A study conducted by Columbia University's Tow Center for Digital Journalism revealed that ChatGPT Search frequently fails to provide accurate answers, often delivering incorrect or misleading information.

  • Out of 200 news citations tested from 20 publishers, ChatGPT provided incorrect or partially incorrect source information in 153 instances, indicating a troubling level of inaccuracy.

  • The study highlighted that ChatGPT Search struggles particularly with news content, which raises questions about its reliability in sourcing current events.

  • Notably, these inaccuracies were observed even for publishers that have partnerships with OpenAI, such as The New York Post and The Atlantic.

  • The chatbot's failures included misattributing quotes and linking to plagiarized content, showcasing significant issues with its data sourcing.

  • When access to certain publications is blocked, ChatGPT may resort to unreliable sources, resulting in over a third of its responses being fabricated or incorrect.

  • In response to the study, OpenAI described the findings as stemming from an 'atypical test' and defended its methodology, stating that misattribution issues are challenging to resolve without specific data.

  • Despite these shortcomings, OpenAI remains committed to improving the accuracy of ChatGPT Search, suggesting that future iterations may enhance performance.

  • The implications of ChatGPT's inaccuracies pose a threat to trust in AI and journalism, potentially affecting OpenAI's partnerships and revenue from licensing deals.

  • Launched in October 2023, ChatGPT Search was designed to combine conversational capabilities with web search tools, aiming to provide real-time, useful information.

  • Looking ahead, OpenAI plans to launch a new AI agent, code-named Operator, in January 2024, which will allow ChatGPT to perform tasks on behalf of users, potentially streamlining various processes.

Summary based on 5 sources


Get a daily email with more Tech stories

More Stories