AI Search Engines Fabricate News Citations, Undermining Publishers
A study by the Columbia Journalism Review and Tow Center for Digital Journalism found that AI search engines frequently provide incorrect answers with false or missing citations, with error rates averaging 60% across eight tested models. The tools fail to credit original publishers and instead fabricate links, threatening the integrity of online information and publisher revenues.
What happened
A study by the Columbia Journalism Review examined eight AI search models and found that in 60% of cases, responses were incorrect. Grok 3 performed worst with a 94% error rate, while Perplexity achieved a 37% error rate.
The study tested ChatGPT Search and found it cited wrong sources in nearly 40% of cases and provided no sources at all in 21% of responses. AI search engines aspirate and reformulate publisher content without crediting original authors, instead displaying syndicated or fabricated links.
Reported impact
Users receive incorrect information presented with absolute confidence, creating an illusion of reliability when answers are entirely wrong. Paid versions of these AI tools show even higher error rates despite expectations of better performance.
Publishers lose traffic as AI search engines redistribute content without directing users to original sources, threatening the economic viability of news organizations and the integrity of online information.
Index analysis
The failure modes involve systematic hallucination and citation fabrication across mainstream search tools, with error rates significantly above acceptable thresholds for information-providing systems.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.