Writing Prompts That Get You a Usable First Draft
A first prompt that's just the task name, "write a product description," "summarize this," "give me some ideas," almost always comes back too generic to use as is. Getting Started with Claude covers the basic habit of describing the situation instead of only the task. This article turns that habit into a framework you can apply the same way every time, whether the task takes ten seconds or ten minutes to think through.
Tip
A prompt that names the task, the audience, the constraints, and the format leaves Claude nothing to guess. A prompt that names only the task leaves it guessing at all three of the others.
You're not writing a cleverer sentence. You're answering four questions before Claude has to guess at them.
The four ingredients of a prompt that works
Every task you'd hand to Claude has four things worth stating up front, even when the answer feels obvious to you.
Task. What you actually want produced or answered. A specific verb and a specific object, not "help with X."
Audience. Who reads or uses the result. A message to your manager and a message to a close teammate call for different words even when the underlying content is identical.
Constraints. Anything the result has to fit inside: a length, a deadline, a tone, information that has to be included or has to be left out.
Format. The shape of the output. A bulleted list, a short paragraph, a table, three options instead of one.
Leave any of the four out and Claude fills the gap with its best generic guess. Sometimes that guess is close enough. It's rarely exactly what you needed on the first try, and the gap usually shows up as an extra round of back and forth that a fuller first prompt would have skipped.
| Ingredient | The question it answers | What Claude does if you skip it |
|---|---|---|
| Task | What should exist after? | Picks the most common interpretation of a vague verb |
| Audience | Who reads or uses it? | Writes for a general reader, usually a shade too formal |
| Constraints | What must it fit inside? | Defaults to its own length, tone, and scope |
| Format | What shape should it take? | Returns prose when you wanted a list, a table, or options |
Weak prompt, strong prompt, same task
The difference is not length. It is how many of the four questions the prompt answers before Claude has to.
| Weak first prompt | What Claude has to guess | Stronger version |
|---|---|---|
| "Write a product update." | Audience, length, tone, what shipped | "Write a 120-word product update for existing customers announcing saved views. Friendly, not salesy. One short paragraph plus a one-line 'how to turn it on.'" |
| "Summarize this report." | Length, focus, who it's for | "Summarize this report in five bullets for an exec who cares about budget impact and timeline, not methodology." |
| "Give me some subject line ideas." | Count, product, angle, length | "Give me seven subject lines under 45 characters for a re-engagement email to trial users who never activated. Curious, not pushy." |
The same framework, three different tasks
The four ingredients don't change based on how big the task is. Here's the same structure applied to three tasks that have almost nothing else in common.
A quick internal message:
Write a two-sentence Slack message to my team letting them know the Tuesday standup is moving to 2pm this week only. Casual tone, no need to explain why.
”A structured comparison:
Compare a monthly and an annual pricing plan for a $15-a-month product, from the buyer's perspective. List the tradeoffs in a short table: cost, commitment, and cancellation risk.
”A creative brainstorm:
Give me eight name ideas for a neighborhood coffee cart that emphasizes speed and quality, not size or seating. Keep each name under three words.
”Notice what's constant. Every one of those names a task, an audience or context, at least one constraint, and a format. The Slack message names casual tone and a two-sentence limit. The comparison names a table with three named columns. The brainstorm names a count and a length limit per idea. None of them are long. None of them leave Claude to guess what "good" looks like here.
- Task: a specific verb and a specific object
- Audience: who actually reads or uses the result
Constraints: a length, a tone, a deadline, or what to include or exclude
Format: the shape you want the answer to arrive in
Run a first prompt against that list before you send it. If one of the four is missing, ask whether you actually don't have an answer yet (in which case, that's fine, Claude can help you figure it out) or whether you just forgot to say it.
Inside Claude Tutorial
This same framework applies to reviewing, comparing, and organizing.
Naming the task, audience, constraints, and format isn't just for writing. The app's lessons carry the same four-part structure into other Claude skills, so it becomes automatic instead of something you have to remember to do.
What "constraints" actually buys you
Constraints are the ingredient people skip most often, usually because the task doesn't feel like it needs one. It almost always does. A length constraint stops Claude from writing three paragraphs when you needed three sentences. A tone constraint stops a casual internal note from coming back sounding like a press release. An inclusion or exclusion constraint stops the answer from wandering into territory you didn't ask about.
Common mistake
Treating constraints as optional polish you'll add later if the first draft needs it. A constraint stated up front shapes the whole draft. The same constraint added after the fact usually means asking for a partial rewrite instead of a clean first pass.
A useful habit: before sending a prompt, ask yourself what would make you reject the first draft outright, too long, too formal, missing one specific piece of information, and state that as a constraint instead of finding out the hard way.
When the task is genuinely small
Not every request needs all four ingredients spelled out in full sentences. A one-line follow-up inside a conversation Claude already has context for doesn't need you to restate the audience again. The framework matters most for the first message in a conversation, or any request where getting it wrong costs you a real round of revision. For something truly quick, task and format alone are often enough.
Give me three alternatives to the second bullet point above, same length, slightly more specific.
”That prompt skips audience and constraints because the surrounding conversation already established both. That's the judgment call: state what Claude doesn't already know, and skip restating what it does.
Common mistakes
- Stacking three requests into one prompt. "Write the email, then a subject line, then a follow-up" gives you three rushed outputs instead of one good one. Ask for the email, get it right, then move on.
- Describing the topic instead of the task. "Something about our new pricing" is a subject, not an instruction. Name the verb: draft, compare, summarize, list.
- Front-loading background and burying the ask. Two paragraphs of context followed by "thoughts?" makes Claude guess what you want done with all of it. Lead with the task, then supply the context it needs.
Frequently asked questions
How do I write a good prompt for Claude?
Answer four things before Claude has to guess at them: the task (a specific verb and object), the audience, any constraints (length, tone, what to include or exclude), and the format you want back. A prompt that names all four is usually enough, even for a quick request.
Should a Claude prompt be long or short?
As long as it needs to be to answer those four questions, and no longer. A three-sentence prompt that names the audience, one constraint, and a format beats a paragraph of background with no clear ask. Length is not the goal; specificity is.
Why does Claude give me generic answers?
Almost always because the prompt named only the task. With nothing said about audience, constraints, or format, Claude fills each gap with its safest general-purpose choice, and the sum of those choices reads as generic. Add one real constraint and the answer sharpens immediately.
Do I need to be polite to Claude for better results?
Politeness doesn't change output quality. Clear structure does. "Please" and "thank you" are fine to include if they feel natural, but they are not what makes a prompt work.
What is the best prompt structure for Claude?
Task, audience, constraints, format, in whatever order reads naturally. For a first message, state all four. For a follow-up inside a conversation Claude already has context for, state only what changed.
Continue reading
- Giving Claude a Role Worth Following: the next technique in this path, for shaping how Claude approaches a request, not just what's in it.
- Using Examples to Show Claude Exactly What You Want: when describing the four ingredients isn't enough, show a sample of the output instead.
- How Claude Actually Thinks, and Why Vague Prompts Get Vague Answers: why an underspecified prompt produces a generic draft in the first place.
- Turning Claude's First Answer Into the One You Actually Need: what to do when even a well-built first prompt doesn't land exactly right.
- Getting Started with Claude: the broader habit this framework builds on, including how to keep a reusable set of instructions.