When Role Prompting Actually Helps
About this lesson
Common advice: start every prompt with “you are an expert in X.” This advice is partly right and partly misleading. Role prompting works, but not for the reason most people think and not in the form most people use it. This video discusses the version that actually helps—functional roles, not credential roles—and the cases where role prompting hurts rather than helps. The Core Idea There are two kinds of role prompts, and they do very different things. Credential Roles: These assert what the model is. Examples include “You are a senior consultant,” “You are an expert in marketing strategy,” or “Act as a world-class engineer.” These assignments give the model a title but no specific framework for organizing its thinking. The model adjusts its tone toward whatever it associates with the credential, but the substance of its output rarely changes meaningfully. Functional Roles: These describe what the model should do and from what perspective. Examples include “Act as an editor reviewing this for clarity,” “Review this from the perspective of someone seeing it for the first time,” or “Evaluate this against these specific criteria.” These assignments give the model a frame for the work: a specific lens, a specific question to answer, or a specific kind of analysis to produce. The substance of the output changes accordingly. Summary The widely-circulated “you are an expert” advice is mostly the credential version. The version that actually moves output is the functional one. The distinction matters because the credential version is what most people use, but the functional version is what actually delivers high-leverage results by providing the model with a specific cognitive framework.
DeepCamp AI