Describe the Product, the Process, and the Performance
Effective description covers three things: what you want produced, how you want AI to approach the work, and how you want it to behave while working with you.
More Than a Prompt
Description is how you communicate with AI so it can actually help. Most prompting advice covers only one part of it: describing the output you want. The framework identifies three things worth describing:
| Type | Question it answers | Example |
|---|---|---|
| Product description | What should the result be? | ”A one-page memo for the operations director, under 400 words, with a recommendation in the first paragraph.” |
| Process description | How should AI approach the work? | ”First list the three options with pros and cons. Then pick one and justify it. Flag any assumptions.” |
| Performance description | How should AI behave while working with me? | ”Be direct and skeptical. Push back if my reasoning is weak. Ask before making assumptions about our budget.” |
Most disappointing AI results come from a missing type, not a badly worded sentence.
Product Description
Product description is the most familiar. It covers what the finished thing should be:
- Purpose and audience: who it’s for and what they need to do with it.
- Format and length: memo, table, bullet list, script; word count or page count.
- Content and constraints: what it must include, what it must avoid, and which sources to rely on.
- Style and tone: plain, formal, warm, technical; reading level.
- Examples: “Here’s a past memo that hit the right tone.” One good example often beats a paragraph of adjectives.
If you’ve learned the CRAFT method in AI Fundamentals, its Context, Ask, and Format parts are mostly product description.
Process Description
Process description tells AI how to get there. It matters most for complex or high-stakes work, where how the answer is reached affects whether you can trust it:
- “Work step by step and show your reasoning before the final answer.”
- “Only use information from the attached policy document. If the answer isn’t there, say so.”
- “Before drafting, ask me up to five clarifying questions.”
- “Give me three different approaches first. I’ll pick one before you write anything.”
A good process description also makes discernment easier later, because you can see how the AI reached its answer instead of just the answer.
Performance Description
Performance description sets how AI should behave as a collaborator. It’s the least-used and often the most powerful:
- “Act as a tough reviewer. Don’t soften criticism.”
- “I’m learning this topic. Give hints, not answers.”
- “Keep responses short. I’ll ask for detail if I need it.”
- “If you’re not confident about a fact, tell me instead of guessing.”
Performance description becomes essential in agency mode. When you configure an assistant that others will use, its instructions are mostly performance description: how to respond, what to refuse, when to hand off to a human.
Common mistake: writing a detailed description of the output and saying nothing about process or performance, then being surprised when the AI fills gaps with confident guesses or agrees with every idea you have.
If you need sources respected, reasoning shown, or honest pushback, say so explicitly. AI assistants tend to be agreeable and to keep going rather than stop and ask, unless you describe a different behavior.
- Pick a real request you’d normally make in one sentence, for example “help me write a proposal for a new mentoring program.”
- Expand it into a product description: audience, purpose, format, length, must-includes.
- Add a process description of two or three sentences about how the AI should work: steps, sources, whether to ask questions first.
- Add a performance description of one or two sentences about how it should behave with you.
- Send the full version in a new chat. Then send only the one-sentence version in a separate new chat. Compare:
- Which result needs less editing?
- Which one asked you useful questions or flagged assumptions?
- Which part of your description made the biggest difference?
1. 'Only use the attached handbook, and say "not found" if the answer isn't in it.' Which type of description is this?
It tells the AI how to approach the work (which source to use and what to do when the answer is missing), so it's a process description.
2. 'Be blunt. If my argument has holes, tell me plainly.' Which type is this?
It describes how the AI should behave as a collaborator, which is performance description.
3. Why does process description help with discernment later?
When you ask for visible steps, sources, or assumptions, you get material to evaluate. Description and discernment reinforce each other.
Up Next
The next lesson puts all three types to work on a realistic task, including when to give examples, how to describe what you don’t want, and how to describe when you don’t yet know what you want.