Description in Practice
Module 3 · Description

Description in Practice

Apply product, process, and performance description to a realistic task, using examples, boundaries, and interviewing techniques for when you don't know exactly what you want yet.

Four Techniques That Make Description Work

1. Show, Don’t Just Tell

Adjectives like “professional,” “engaging,” or “concise” mean different things to different people, and to AI. An example removes the guesswork:

Here's a past announcement our staff liked. Match its length, tone,
and structure, but not its content:

[paste example]

Say clearly which qualities of the example to copy. Otherwise AI may copy the wrong ones, such as reusing its exact phrases.

2. Describe the Boundaries

What you don’t want is often as important as what you do:

  • “Don’t use jargon. Readers are new volunteers, not staff.”
  • “Don’t invent statistics. Leave a placeholder like [STAT] where a number should go.”
  • “Don’t change the meaning, only the wording.”

Placeholders are especially useful: they stop AI from filling gaps with plausible fiction and show you exactly what you need to supply.

3. Let AI Interview You

When you’re not sure what you want, don’t fake it. Use a performance description that turns the AI into an interviewer:

I need to write a grant summary but I'm not sure what to emphasize.
Ask me questions one at a time, up to eight, to figure out what
matters most. After that, propose an outline for my approval.
Don't write the summary yet.

This is description and augmentation together. The questions often clarify your own thinking.

4. Describe in Layers

For big tasks, don’t describe everything at once. Describe the plan first, approve it, then describe each piece. A layered approach keeps you in control of the important decisions and makes errors easier to catch early.

⚠️ Common Pitfall: The Wall of Instructions

Common mistake: writing a 600-word prompt covering every possible detail, sending it, and then not being able to tell which instruction the AI ignored.

Long descriptions are fine when they’re organized. Group them under headings (Product, Process, Behavior), put the most important instruction first, and for complex work describe in layers instead of all at once.


Lab: Describe a Real Task Three Ways

This lab uses a realistic scenario. Feel free to swap in your own.

Scenario: You coordinate a community library’s volunteer program. Volunteer sign-ups have dropped 30% this year. You need to present options to the library board next month.

🧪 Try It: Build a Full Description, Layer by Layer
  1. Layer 1: interview. Start a new chat and ask the AI to interview you about the situation (up to six questions, one at a time). Answer using the scenario, inventing reasonable details.
  2. Layer 2: plan. Ask for a proposed outline of the board presentation. Add a process description: “Give three options for reversing the decline, each with cost, effort, and risk. Mark anything you’re assuming.” Review the outline and change at least one thing.
  3. Layer 3: product. Ask for the presentation’s opening section only. Include a product description (board audience, five-minute section, plain language) and a boundary: “Use [STAT] placeholders instead of inventing numbers.”
  4. Add a performance description: “Now act as a skeptical board member. What three questions would you ask about this proposal?”
  5. Reflect. Write down:
    • Which layer changed the result most?
    • Where did the AI make assumptions you had to correct?
    • What would you put in your first message next time to save a round?
🧠 Knowledge Check

1. Why ask AI to use placeholders like [STAT] instead of numbers?

2. You know you need help with a strategy memo but aren't sure what it should say. What's the most fluent first step?

3. When you give an example to match, what should you also do?


Up Next

You’ve described the work well, and something came back. Is it any good? Module 4 covers Discernment: judging the product, the process, and the performance.