1. Home
  2. Glossary
  3. Persona prompting
AI glossary · Prompting

Persona prompting

Persona prompting: Persona prompting is the technique of telling an AI model to respond as a specific role or character, such as 'You are a senior tax accountant,' to shape the expertise, tone, and assumptions in its answer.

Assigning a persona is the fastest way to change how a model answers. 'Explain this contract clause' gets a general summary. 'You are a commercial lease attorney reviewing this for a small-business tenant; explain this clause and what it exposes them to' gets an answer that knows what to look for. The persona tells the model which body of knowledge, vocabulary, and level of caution to use.

Personas work best when they are specific about the role, the audience, and the situation. 'Act as an expert' does little. 'Act as a hospital billing manager explaining a denied claim to a front-desk employee' does a lot. You can also assign a persona to the reader ('Write for a CFO who has five minutes') or ask the model to critique a draft from several personas in turn: a skeptical customer, a compliance officer, a new hire.

A persona does not create knowledge or credentials. Telling a model it is a licensed physician does not make its medical claims reliable, and it may make the model sound more confident than it should. Use personas to shape framing and tone, then verify facts the same way you would with any AI output.

Example at work

A product manager drafting release notes asks the model to rewrite them three times: once as a support agent who will field the calls, once as a sales rep who will pitch the feature, and once as a security-conscious IT buyer. The three versions surface questions she had not thought to answer, and the final notes address all of them.

Why it matters

Persona prompting is a cheap way to get answers tuned to your role and audience instead of generic ones. Knowing its limits, especially that a persona adds framing rather than expertise, keeps you from trusting a confident-sounding answer that is simply wrong.

Related terms