Curie Brief
Turn on cookies to sign in
Signing in saves your progress to your Curie account. We can only do that with cookies on — turn them on to continue.
News
AI medical advice isn't equally reliable for everyone. A new study found that four major AI tools gave less accurate, less comprehensive GI guidance — and were more likely to miss emergencies — when responding in languages other than English. With 68 million U.S. residents speaking a non-English language at home, researchers say language equity is a patient safety issue that can't be ignored.
Millions of Americans are turning to AI chatbots for health advice before ever seeing a doctor — but a new study presented at the ACG Annual Scientific Meeting reveals that this advice is significantly less reliable when asked in a language other than English.
Researchers tested four major large language models (LLMs) — GPT-5 Mini, Gemini 3 Flash, Claude Haiku 4.5, and Llama 3.3 — on 50 gastroenterology questions translated into Spanish, Chinese, Tagalog, Vietnamese, Arabic, and French. The results were striking: accuracy and completeness dropped sharply in non-English languages, and the models were far more likely to miss medical emergencies.
By the Numbers:
Why it matters: As AI becomes a go-to health resource, its performance gaps across languages create real risks for non-English-speaking patients — especially in urgent situations. Researchers are calling on GI practices to evaluate AI tools in the languages their patients actually use before deploying them.
