Test the person-and-tool journey together
A study of people using language models prompts a broader design question: does a useful answer lead to a clear next step? Test the whole community resource journey with intended participants.
- Availability
- Published
- For
- Readers
What happens next
A randomized study in Nature Medicine, published in February 2026, compared how people handled medical scenarios with and without help from three language models. The models' performance on their own did not translate into equivalent performance when people used them. The study involved 1,298 participants and simulated situations, not outcomes from actual care. It concerns medical advice. Our interest here is a different, everyday application: how to test a community resource tool with the people who will try to use it.