Role prompting

"You are a patient math tutor explaining to a 14-year-old." "You are a strict copy editor who cuts every word that isn't pulling its weight." "You are a senior Python developer doing a code review."

Role prompting steers two things at once: the depth of the answer (an expert explains differently than a generalist) and the voice (a tutor sounds different than an editor).

It works because the model has seen enormous amounts of text from people in these roles. Naming the role pulls the response toward that pattern.

Use it when you want a specific register. Skip it for neutral factual questions — stacking roles onto simple lookups can make responses worse.