Role prompting

“You are…” is the shortest line in most prompts and one of the most misunderstood. A role doesn’t make a model smarter; it changes who it is talking as, and to.

Same question, different role

Question: “What is a mortgage?”

Prompt: What is a mortgage?

A mortgage is a loan used to buy property. The property serves as collateral, and you repay the loan with interest over a set term, typically 15 to 30 years.

What changed: Correct, neutral, generic. Fine for a quick fact.

Writing a role that works

  • Be specific. “A mortgage broker advising a first-time buyer” beats “a finance expert”.
  • Include the audience. Half of a role’s effect comes from who it is talking to.
  • Add what they care about. “…who always flags hidden costs” steers priorities.
  • Don’t rely on it for facts. Supply the facts in context; the role only shapes how they’re used.

The Role field in the prompt generator and the system prompt generator both put your role where it has the most effect.

Questions

What is role prompting?

Role prompting assigns the model a persona, expertise or point of view, usually with “You are…” or “Act as…”. It changes vocabulary, depth, priorities and tone. Anthropic’s documentation recommends setting a role in the system prompt as one of the most effective ways to shape Claude’s behaviour.

Does “you are an expert” make answers more accurate?

Not on its own. A role changes style and focus, but it does not add knowledge the model lacks, and research on generic “expert” personas shows mixed effects on accuracy. Specific roles tied to an audience and goal help more than flattering titles.

Where should the role go?

At the top of the prompt, or in the system prompt if you are building an assistant. Keep it to one or two sentences and make it concrete: who they are, who they are talking to, and what they care about.