Module 7: Output Formatting and Roles

Why this matters

Most people accept whatever default format the AI produces. That’s a missed opportunity. You can shape output dramatically by specifying format and role — making the model far more useful for your specific work.

Formatting output

You can tell AI to format output as almost anything:

Lists and structure:

• “Bullet points.”
• “Numbered list with sub-bullets.”
• “Outline form, three levels deep.”

Tables:

• “Comparison table with columns: option, price, pros, cons.”
• “Markdown table.”
• “CSV format I can paste into Excel.”

Documents:

• “Slide deck outline, 5 slides.”
• “Executive summary, 100 words, then full detail.”
• “FAQ with 8 questions.”
• “Standard Operating Procedure (SOP) format.”

Code-like formats:

• “JSON object with these keys.”
• “Markdown.”
• “YAML config.”

Communication formats:

• “Email with subject line and body.”
• “LinkedIn post under 1300 characters.”
• “Tweet thread, 5 tweets, each under 280 characters.”
• “Blog post intro, 100 words, hook in the first sentence.”

The format you ask for is the format you get. Specify it.

Using roles and personas

Telling the model to play a specific role unlocks different patterns from its training:

“Act as a senior copywriter with 15 years of B2B SaaS experience. Review this landing page and identify the three weakest sections.”

“You are a startup advisor. I’ll describe my business idea. Give me your honest take, including reasons it might fail.”

“Act as a financial advisor working with a young family. Walk me through how to think about emergency funds vs investing.”

Why roles work:

• They activate domain-specific patterns from training data
• They set tone automatically (a financial advisor sounds different from a comedian)
• They give the model a frame for what’s relevant and what’s not

Tip: Combine roles with the Structured Prompt Formula. The role becomes part of “Context.”

Behavior modes

Beyond roles, you can set modes — short, named patterns of behavior:

• teach — Explanations with examples and analogies
• critique — Pure feedback, no praise
• plan — Steps and structure, not prose
• think — Deeper analysis before answering
• brainstorm — Quantity over quality, no filtering

Example:

“Mode: critique. Review this email and tell me everything that’s wrong with it. Don’t say anything nice.”

Modes are useful because they give you a quick way to set behavior without rewriting the whole prompt. Once you know which modes work, you can use them as shortcuts.

Building reusable prompt templates

A prompt template is a saved prompt with variables you fill in for each use.

Example template — Email Reply Analyzer:

Mode: critique.

You are a communication coach. Review the email below and tell me:
1. What's the writer's likely emotional state?
2. What's their actual ask, in plain language?
3. What's the cleanest reply I can send?

[PASTE EMAIL HERE]

Save five or ten templates for tasks you do often. Suddenly you have a personal AI toolkit.

Common templates worth building:

• Meeting notes → action items
• Long article → 3-bullet summary
• Bad email → rewritten clearly
• Idea → critique from three angles
• Document → flashcards for studying

Key takeaways

• Specify the format — bullet points, table, JSON, slide deck, whatever
• Use roles to activate domain-specific patterns
• Use modes (teach, critique, plan, think) as shortcuts
• Build a personal library of templates for tasks you repeat

Quick Check