The Digital Press

Researchers evaluated six major large language models on 9,100 health-related claims across 21 languages, revealing strong English performance but weaker results in many non-European languages and variable accuracy by topic and source.

Why it matters: Inaccurate AI health information risks misleading diverse global audiences, affecting public health decisions.

The big picture: Multilingual, domain-specific validation is crucial before AI tools are widely used in global health communication.

Stunning stat: Models underperform significantly in multiple non-European languages compared to English-centric claims.

The stakes: Deploying unvalidated AI in health communication can propagate misinformation and harm public trust worldwide.