Fact-checking information from large language models can decrease headline discernment
A preregistered experiment comparing how LLM-generated and human fact checks affect belief in and sharing of political headlines.
Authors or creators: Matthew R. DeVerna, Harry Yaojun Yan, Kai-Cheng Yang, Filippo Menczer
Checked: 2026-08-01
Source note: Primary peer-reviewed research. The experiment concerns political headlines and one fact-checking setup, so broader use cases require separate evidence.
Claims connected to this source
- LLM fact checks did not significantly improve headline discernment, while human fact checks improved it in the experiment.
Open original · Proceedings of the National Academy of Sciences
