A Claude Prompt Template That Works for Almost Any Task

Most bad prompts are short. They look efficient, and in reality they force Claude to guess who you are, who the output is for and what "good" means. When it guesses, you get the most average answer possible. The fix is a simple structure you can fill in quickly, and after a while you stop needing the template at all.

This is the structure I use when I brief Claude on anything that matters, from product copy to a feature spec.

The template

Copy this, fill in what applies, and delete what does not.

<context>
[Paste the real material: the draft, the notes, the data, the email thread,
examples of past work. Long material goes here, at the top.]
</context>

You are helping me as [the kind of expert this needs].
The output is for [who will read or use it, and what they already know].

The job: [one or two sentences, stated as the outcome you want].

Why it matters: [the goal behind the task, so Claude can make good calls].

Constraints:
[length, format, tone, what to include, what to leave out]

Here is an example of the level I want:
<example>
[one short example you like]
</example>

If something important is missing, ask me up to three questions before you start.

That is it. Nothing clever. Each part exists because it removes one specific guess.

Why each part is there

Context first. The single biggest quality jump comes from pasting real material instead of describing it. Anthropic's prompting best practices recommend putting long documents near the top, above your instructions. Wrapping them in simple tags like <context> keeps your material separate from your instructions, which reduces misreadings.

The expert and the reader. "You are a senior product designer" sets the level of the answer. "For a founder who has never shipped an app" sets the vocabulary. You need both. A technical explanation for engineers and one for a client can share every fact and still be completely different texts.

The job as an outcome. "Write an email" is an activity. "Get the client to approve the revised timeline without a call" is an outcome. Claude makes better choices when it knows what winning looks like.

Why it matters. This is the part people skip, and it is one of the most useful. The docs say plainly that explaining the motivation behind an instruction helps Claude deliver more targeted answers. "Keep it short because it will be read on a phone between meetings" beats "keep it short."

Constraints, stated positively where you can. The same guide suggests telling Claude what to do instead of what not to do. "Write in plain sentences a busy person can scan" works better than a list of banned things. Keep one or two hard "never" rules if you really need them.

One example. Examples are one of the most reliable ways to steer tone and format. One good example often does more than a paragraph of adjectives. If you give several, make them different from each other, or Claude will copy the one pattern too closely.

Permission to ask. The last line saves more time than anything else. A few questions up front beat three rounds of corrections.

What about "think step by step"?

You will see this at the end of many prompt templates online. With current Claude models it matters less than it used to, because they already decide how much to reason on their own. If you want more depth in the Claude app, raise the effort setting in the model picker. In Claude Code, add the word ultrathink to a message, which the model configuration docs confirm is the only phrase it treats as a thinking keyword. Plain "think hard" is read as normal text.

A worked example

Here is a lazy prompt:

Write a launch post for my app.

And the same request using the template:

<context>
App: a time tracker for freelancers that logs billable hours automatically.
Launch notes: [paste]
Two past posts of mine that did well: [paste]
</context>

You are helping me as an experienced product marketer.
The output is for freelance designers and developers who already track
time manually and hate it.

The job: a LinkedIn launch post that gets people to try the free version.

Why it matters: my audience ignores posts that sound like ads. It needs to
read like something I learned while building it.

Constraints: under 1,200 characters, short paragraphs, one concrete moment
from building it, the link only at the end, no hashtags.

Match the rhythm of my past posts above.

If something important is missing, ask me up to three questions first.

The second one takes two minutes longer to write. It usually saves twenty minutes of back and forth.

A few follow ups worth keeping

The first answer is a draft. These short follow ups do most of the work after it:

Tell me the three weakest parts of this and rewrite only those.

Make the strongest case against this plan, then tell me what I
should test before committing.

Explain this to someone smart who is new to the field. No jargon in the
first paragraph, then define the two terms they will meet next.

Here is my rough prompt: [paste]. Rewrite it using context, reader,
job, why it matters, constraints and an example. Ask me for anything
you cannot fill in.

That last one is useful when you are in a hurry. Claude interviews you and hands back a proper brief.

Codes are not commands

You may have seen lists of magic codes like /godmode or /10x that supposedly trigger hidden behavior. Claude has no secret list of them. Unless you created a skill with that name, it reads the code as ordinary words and guesses what you meant. A clear sentence always beats a clever word.

If you have a prompt you reuse every day, you can turn it into a real command. In Claude Code, save it as a skill:

~/.claude/skills/sharpen/SKILL.md

---
name: sharpen
description: Tightens a piece of writing. Use when asked to sharpen,
  tighten or cut a draft.
---
Rewrite the text so it is shorter and more specific. Keep every fact.
Cut filler, replace vague words with concrete ones, and keep the
author's voice. Then list the three changes that did the most work.

The folder name becomes the command, so this gives you /sharpen. The skills documentation covers the details. In the Claude app you can upload the same folder as a zip under Customize, then Skills.

Start with the next prompt you were about to send in one line. Add the context, the reader and the reason, and compare the answer with what you usually get. After a week of doing this, you will write the structure from memory.