Study: 'You Are an Expert' Persona Prompting No Longer Improves AI Accuracy
A new study finds that assigning expert personas to frontier AI models — "you are a great physicist," "you are a lawyer" — no longer produces measurable accuracy improvements on domain-specific tasks. Current models' baseline instruction-following capabilities have rendered the technique redundant. The study also found persona prompting caused no measurable harm, indicating models have become robust to conflicting self-identity framing. The research was highlighted by Wharton professor Ethan Mollick.
Why It Matters
Invalidates a widely cited prompt engineering technique standard since 2023, requiring updates to enterprise prompt libraries and best-practices documentation. Engineering teams over-relying on persona framing may be adding tokens without benefit.