Claude Prompts: 12 Examples and the Patterns Behind Them

Claude prompts that work: 12 copy-paste examples for research, writing, code and automation, built on the patterns in Anthropic's own prompting docs.

Josh Pocock8 min read

Claude prompts cover

AI Coding

Claude prompts

Stride AI Academy

The best Claude prompts are specific, give the reason behind each instruction, show an example or two, and separate your instructions from your data with XML tags. Those patterns come straight from Anthropic's own prompting docs, and they work across chat, the API and Claude Code. Below are 12 copy-paste Claude prompt examples built on them, grouped by research, writing, code and automation.

Each section names the pattern, links the exact docs section it comes from, then gives prompts you can adapt. Everything here was checked against Anthropic's prompting guide for current models and its prompting overview on November 22, 2026. Anthropic describes that page as its living reference, so check it when new models ship.

Before you write a prompt

Anthropic's overview says to have three things first: a clear definition of success, a way to test against it, and a first draft. If you can't say what a good answer looks like, no wording fixes that. Write one line that starts "A good answer..." before you write the prompt.

Pattern 1: be clear and direct

The docs suggest treating Claude like a brilliant new employee who doesn't know your norms. Their test: show the prompt to a colleague with little context. If they'd be confused, Claude will be too. Use numbered steps when order matters, and ask for more if you want more. (Be clear and direct)

Prompt 1, research brief:

Research how three competitors price their entry plan. For each one:
1. Name the plan and its monthly price.
2. List what's included in five bullets or fewer.
3. Link the pricing page you used.
Finish with a table comparing all three. If a price isn't published, say so instead of estimating.

Prompt 2, a build that goes past the basics:

Create an analytics dashboard for a small agency's monthly client report. Include as many relevant features and interactions as possible. Go beyond the basics to create a fully featured version.

The second sentence is Anthropic's own example. Without it, you tend to get the bare minimum.

Pattern 2: explain why

Giving the reason behind a rule helps Claude apply it in cases you didn't spell out. The docs contrast "NEVER use ellipses" with a version that says the text will be read aloud by a text-to-speech engine. (Add context)

Prompt 3, a voice script:

Write a 60-second voicemail script for a dental clinic's after-hours line. It will be read aloud by a text-to-speech voice, so spell out numbers and times in words, avoid abbreviations, and keep sentences under 15 words so the pacing sounds natural.

Pattern 3: show examples

Examples are one of the most reliable ways to steer format, tone and structure. The docs recommend 3 to 5 examples that are relevant, varied enough to cover edge cases, and wrapped in <example> tags inside <examples>. (Use examples effectively)

Prompt 4, subject lines in your voice:

xml
Write 10 email subject lines for our webinar on client onboarding. Match the style of these examples.

<examples>
<example>Your first client call, minute by minute</example>
<example>The 3 questions we ask before every proposal</example>
<example>Why our onboarding takes 9 days, not 30</example>
</examples>

Pattern 4: structure prompts with XML tags

When a prompt mixes instructions, context and input, wrap each in its own tag such as <instructions>, <context> and <input>. Keep tag names consistent and nest them when content has a hierarchy. (Structure prompts with XML tags)

Prompt 5, rewrite with clear boundaries:

xml
<context>
We sell bookkeeping to restaurants. Our readers are owners, not accountants.
</context>

<instructions>
Rewrite the draft below at a 7th-grade reading level. Keep every number. Cut anything that repeats.
</instructions>

<input>
{{PASTE DRAFT HERE}}
</input>

Pattern 5: put long documents first and ask for quotes

For inputs over about 20,000 tokens, the docs say to put long documents at the top, above your question and instructions. In Anthropic's tests, queries at the end improved response quality by up to 30%. Wrap each document with its source, and ask Claude to pull relevant quotes before answering. (Long context prompting)

Prompt 6, contract review:

xml
<documents>
  <document index="1">
    <source>vendor_contract.pdf</source>
    <document_content>{{CONTRACT}}</document_content>
  </document>
  <document index="2">
    <source>our_standard_terms.docx</source>
    <document_content>{{TERMS}}</document_content>
  </document>
</documents>

Find quotes in the contract that differ from our standard terms on payment, termination and liability. Put them in <quotes> tags. Then explain each difference in plain English in <summary> tags.

Pattern 6: give Claude a role

A role in the system prompt focuses tone and behavior. The docs say even one sentence makes a difference. (Give Claude a role)

Prompt 7, a role set through the API:

python
import anthropic

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-opus-5-5",
    max_tokens=1024,
    system="You are a senior support lead at a SaaS company. You write short, warm replies that solve the problem first.",
    messages=[{"role": "user", "content": "Draft a reply to a customer who was double-charged this month."}],
)
print(message.content)

In chat, put the role in a Project's instructions so it applies to every conversation there. Our Claude API tutorial covers getting a key and making your first call.

Pattern 7: say what to do, not what to avoid

To control formatting, the docs suggest telling Claude what you want instead of what you don't, and matching your prompt's own style to the output you want. Removing markdown from your prompt can reduce markdown in the reply. (Control the format of responses)

Prompt 8, prose for a report:

Write the summary as smoothly flowing prose paragraphs, the way a consultant would in a client letter. Use headings only to separate the three sections.

Prompt 9, no preamble:

Respond directly without preamble. Do not start with phrases like "Here is" or "Based on". Return only the finished product description.

That second line is close to the docs' own advice for cutting introductory text, now that prefilled assistant replies aren't supported on Claude 4.6 and later models.

Pattern 8: be explicit about action in Claude Code

Current models follow instructions precisely. "Can you suggest some changes" can get you suggestions rather than edits. Say "change" or "make these edits" when you want action. (Tool usage)

Prompt 10, an edit, not advice:

Change the signup form so the email field validates on blur, not on submit. Read the form component and its tests first, then make the edit and run the tests.

Prompt 11, stay in scope:

Fix only the date bug in invoice totals. Don't refactor surrounding code, add comments to code you didn't change, or add error handling for cases that can't happen. If a test looks wrong, tell me rather than working around it.

Prompt 11 condenses the docs' samples on overeagerness and avoiding hardcoded test fixes. If you reuse a prompt like this often, save it as a command: see Claude Code slash commands. Rules that apply to every session belong in your CLAUDE.md file.

Pattern 9: let Claude think, and chain when you need checkpoints

Current models use adaptive thinking, deciding how much to think based on the effort setting and how hard the task is. You can guide it in plain words, for example asking Claude to reflect on tool results before its next step. To rein in thinking, the docs suggest lowering the effort setting, or using max_tokens as a hard limit. (Thinking and reasoning)

When you need to inspect each step, chain prompts. The most common chain in the docs is self-correction: draft, review against criteria, refine, each as a separate call. (Chain complex prompts)

Prompt 12, the review step of a chain:

xml
<criteria>
- Every claim has a source link
- No paragraph over four sentences
- The first line answers the reader's question
</criteria>

<draft>
{{DRAFT FROM STEP 1}}
</draft>

Review the draft against each criterion. For each one, say pass or fail and quote the lines that fail. Don't rewrite yet.

Then send the review back with "Revise the draft to fix every failed item." Splitting review from rewrite lets you check the review before anything changes.

Claude prompting tips that apply everywhere

  • Ask for sources and a "say so if unsure" rule in research prompts. The docs' research guidance recommends clear success criteria and checking across sources.
  • Soften all-caps rules. The docs note that Opus 4.5 and Opus 4.6 respond more strongly to system prompts than earlier models, so "CRITICAL: You MUST" can cause overuse. "Use this tool when..." is usually enough.
  • Keep one prompt per job. A prompt that researches, writes and formats at once is harder to fix than three short ones.
  • Test against your success line, then change one thing at a time.

If Claude still pushes back on a reasonable request, our explainer on Claude's constitution covers the published principles behind how it decides. To make your prompts produce something you can share, see Claude artifacts.

Learn to build with Claude Code

The Stride AI Coding course teaches Claude Code and Codex for building client tools, written for non-programmers, with weekly live calls to review your builds. Many of the free guides cover Claude Code setups, and the Claude Code community is where members compare theirs.

FAQ

What makes a good Claude prompt?

A clear goal, the reason behind each rule, a few examples of the output you want, and XML tags separating instructions from your data. Anthropic's docs also suggest defining what success looks like before you start.

Should I use XML tags in Claude prompts?

Yes, when a prompt mixes instructions, context, examples and input. Anthropic's docs say tags reduce misinterpretation. There are no required tag names; use descriptive ones and keep them consistent.

How many examples should I include?

Anthropic recommends 3 to 5 examples for best results. Make them relevant to your real task, varied enough to cover edge cases, and wrap them in <example> tags.

Where should long documents go in a prompt?

At the top, above your instructions and question. Anthropic's tests found that putting the query at the end improved quality by up to 30% on complex, multi-document inputs.

Do these prompts work in Claude Code?

Yes. The same patterns apply, and the docs add one for coding: say "change" or "make these edits" when you want Claude to act, not "can you suggest". Save prompts you reuse as slash commands.

Can I still prefill Claude's response?

Not on Claude 4.6 and later models, where a prefilled last assistant turn returns an error. The docs suggest direct instructions, XML output tags or structured outputs instead.

Build it with us in Stride AI Academy

Four courses, weekly live calls and a community of people building and selling AI. Or start with 320+ free guides.

Keep reading

Free AI Coding guides