Notice: This page requires JavaScript to function properly.
Please enable JavaScript in your browser settings or update your browser.
Learn Roles And Framing — Giving AI The Right Lens | How Prompts Work
Prompt Engineering for Work

bookRoles And Framing — Giving AI The Right Lens

Swipe to show menu

The four components tell the AI what to produce. Roles and framing tell it how to approach the task — from what perspective, with what expertise, and through what lens.

This is one of the most effective and underused techniques in everyday prompting. It doesn't require any additional length or complexity — just a single line at the start of the prompt that changes everything that follows.

What Role Prompting Does

When you assign a role, you're not asking the AI to pretend or perform. You're giving it a frame of reference that shapes:

  • The vocabulary and level of technical detail it uses;
  • The assumptions it makes about the audience;
  • The perspective it brings to analysis or recommendations;
  • The tone and register of the output.

A role can be a job title, a type of expert, a communication style, or even a mindset. All of these are valid:

  • Act as an experienced employment lawyer...;
  • You are a plain-language editor whose job is to make complex text accessible to non-experts...;
  • Take the perspective of a skeptical investor reading this for the first time...;
  • Respond as a direct, no-nonsense manager who doesn't have time for long explanations....
Screenshot description: A chat window showing the same base request sent three times with different role instructions, producing visibly different responses. The base request is: "What should I consider before launching a new product feature?" First version: no role — AI responds with a generic bulleted list of common considerations. Second version: prefixed with Act as a product manager with 10 years of B2B SaaS experience — AI responds with a more specific, prioritized list referencing user research, stakeholder alignment, and rollout risk. Third version: prefixed with Take the perspective of a customer who has been burned by half-finished features before — AI responds with a skeptical, direct list of questions a frustrated user would ask. Each version labeled with its role instruction. The shift in tone and perspective is clearly visible across the three responses.

Framing Beyond Job Titles

Roles don't have to be professional titles. Some of the most useful frames are behavioral or attitudinal:

  • Expertise level of the audienceExplain this as if you're talking to someone with no background in finance;
  • Communication styleWrite this in the style of a direct, confident executive who uses short sentences;
  • Critical perspectiveReview this proposal as a devil's advocate — find the weakest points;
  • Specific relationshipRespond as a trusted advisor who is being honest with a client, not just telling them what they want to hear.

When Role Prompting Is Not Necessary

Role prompting adds value when the perspective or expertise matters to the output. It's unnecessary when:

  • The task is simple and the output format is self-evident;
  • You're working on a generic task with no specific audience or expertise requirement;
  • Adding a role would make the prompt longer without making the output more useful.

The test: ask yourself — "would a different expert give a meaningfully different answer to this?" If yes, a role is worth adding. If no, skip it and keep the prompt focused.

1. What is the main effect of assigning a role to the AI in a prompt?

2. Which of the following scenarios are cases where adding a role to your prompt does not add value?

question mark

What is the main effect of assigning a role to the AI in a prompt?

Select the correct answer

question mark

Which of the following scenarios are cases where adding a role to your prompt does not add value?

Select all correct answers

Everything was clear?

How can we improve it?

Thanks for your feedback!

Section 1. Chapter 4

Ask AI

expand

Ask AI

ChatGPT

Ask anything or try one of the suggested questions to begin our chat

Section 1. Chapter 4
some-alt