What you'll learn
You've controlled the instruction, the role, and the input. The last lever in the toolkit is the output itself — its length, format, tone, and where it starts. Brief the model the way you'd brief a colleague, and you'll stop fighting walls of text you didn't want.
By the end of this module you'll be able to:
- Specify length, format, and tone so the answer arrives ready to use
- Steer the response by prefilling how it should begin
- Stop the model from over-explaining or under-answering
Specifying length
"Briefly" and "in detail" mean different things to a model than to you. Give a concrete target — a word count, a sentence count, or a number of bullets — and the output lands where you expect.
| Vague | Concrete |
|---|---|
| “Briefly” | “In 2 sentences” |
| “A short summary” | “5 bullet points, max 12 words each” |
| “Go into detail” | “About 300 words, with a subheading per section” |
| “One-liner” | “A single sentence under 15 words” |
Specifying format
The shape of the answer matters as much as the content. Say whether you want prose, a bulleted list, a table, or structured data — and the model will oblige. If you'll feed the output into software, ask for a machine format explicitly (we go deep on JSON in Module 14).
Prompt
Compare cats and dogs as pets.
Response
Prompt
Compare cats vs dogs as pets in a Markdown table with columns: Trait, Cats, Dogs. Include exactly 4 rows: independence, exercise needs, space, upkeep cost.
Response
| --- | --- | --- |
| Independence | High | Lower |
| Exercise needs | Low | High |
| Space | Small OK | More is better |
| Upkeep cost | Lower | Higher |
Controlling tone & style
Tone is a spectrum you can dial precisely: formal or casual, warm or neutral, playful or terse. Combine it with the audience persona from Module 6 for tight control. And prefer showing over telling — one line of an example style is worth several adjectives.
| Ask for | You'll get |
|---|---|
| “Formal, no contractions” | Buttoned-up, business-appropriate prose |
| “Warm and encouraging” | Friendly, supportive phrasing |
| “Punchy, like a landing page” | Short, high-energy marketing lines |
| “Neutral and factual” | Just the information, no color |
Prefilling the answer's start
A powerful trick: begin the answer for the model. Because it continues text (Module 2), the first few words steer everything after them. Start the response with { and it will keep writing JSON; start with 1. and it will produce a numbered list.
Prompt
List 3 productivity tips. Begin your reply exactly like this and continue: 1.
AI response
2. Turn off notifications during deep-work blocks.
3. End each day by writing tomorrow's top three priorities.
Tip
Stopping over- and under-answering
Two opposite failure modes, each with a simple fix:
- Over-answering (rambling, caveats, restating the question): ask for only the answer, set a length cap, or prefill past the preamble.
- Under-answering (one-word replies, missing parts): ask it to address each numbered point, or require a minimum ("at least 3 reasons").
Key idea
Recap & quick check
Key takeaways
- Give concrete length targets (word/sentence/bullet counts) instead of 'briefly' or 'in detail'.
- Name the format — prose, bullets, table, or structured data — especially if software will read it.
- Tone is a precise dial; a one-line style sample beats a pile of adjectives.
- Prefilling the first tokens of the answer steers its format and skips preamble.
- Over- and under-answering are usually a blank output spec, not a content problem — specify the shape.
Quick check
1. Why prefer 'in 2 sentences' over 'briefly'?
2. How does prefilling the answer's start steer the output?
3. You need output a program can parse. What should the prompt do?
4. The model keeps rambling with caveats and restating your question. Best fix?
That completes the everyday toolkit. Now we push into harder problems — getting the model to reasonbefore it answers. Next up: Phase 3, Module 10 — Chain-of-Thought Prompting.