Writing Clear Claude Prompts: Instructions, Examples, Roles and Output Format

1.Write the prompt a new colleague could follow

Claude starts every request without knowing your conventions, your audience, or what a finished answer looks like. Anthropic's guidance says the more precisely you explain what you want, the better the result, and it gives a simple test, the golden rule. Show your prompt to a colleague who has minimal context on the task and ask them to follow it. If they would be confused, Claude will be too.

Three habits follow from that test. First, be specific about the output format and constraints instead of leaving them implied. Second, if you want more than the minimum, ask for it. "Create an analytics dashboard" gets you a dashboard. Adding "Include as many relevant features and interactions as possible. Go beyond the basics to create a fully-featured implementation" asks explicitly for the fuller version. The guidance tells you to request "above and beyond" behaviour directly rather than relying on Claude to infer it from a vague prompt. Third, when the order or completeness of steps matters, write the steps as a numbered list or bullet points.

The documented fix is to give a reason, not to shout louder. The best-practices guide rewrites the rule as: "Your response will be read aloud by a text-to-speech engine, so never use ellipses since the text-to-speech engine will not know how to pronounce them." Giving the context or motivation behind an instruction helps Claude understand your goal and respond more precisely. The guide's explanation is short: Claude is smart enough to generalize from the explanation.

Sources1

2.Steer format and tone with a few well-chosen examples

Sometimes describing the output isn't enough, and it's clearer to show it. The guide calls examples one of the most reliable ways to steer Claude's output format, tone and structure. It says a few well-crafted examples, a technique known as few-shot or multishot prompting, improve accuracy and consistency. What makes them work is quality, not quantity, and the guidance names three properties to aim for.

Sources1

3.Separate instructions, context and inputs with XML tags

Examples aren't the only content that can blur into your instructions. Retrieved documents and variable inputs can too. According to the guide, XML tags help Claude parse complex prompts unambiguously, particularly when a prompt mixes instructions, context, examples and variable inputs. Putting each type of content in its own tag, such as <instructions>, <context> or <input>, reduces misinterpretation.

The guide lists two best practices. Use consistent, descriptive tag names across your prompts, and nest tags when the content has a natural hierarchy: documents go inside <documents>, and each one sits in its own <document index="n">. When every source has its own indexed tag, Claude can still tell the sources apart. If they are pasted in as one undifferentiated block, it can't.

Sources1

4.Put the role in the system prompt and the request in the user turn

Where an instruction goes matters as much as how it is worded. The guide says that setting a role in the system prompt focuses Claude's behaviour and tone for your use case, and that even a single sentence makes a difference. In the Messages API, the system parameter says who Claude is for this application, and the messages array carries the specific request:

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-opus-5-5",
    max_tokens=1024,
    system="You are a helpful coding assistant specializing in Python.",
    messages=[
        {"role": "user", "content": "How do I sort a list of dictionaries by key?"}
    ],
)

print(message.content)

The system prompt is also where standing guidance on tone belongs. On Claude Sonnet 5, setting temperature, top_p or top_k to a non-default value returns a 400 error. The Sonnet 5 guide says to remove those parameters and use system-prompt instructions to guide tone and variety instead.

Sources12

5.Constrain the output with an explicit format

The final part of a clear prompt is the shape of the answer. The consistency guide says to define your output format precisely, using JSON, XML or a custom template, so that Claude follows every formatting element you require. A JSON instruction can name the keys and their allowed values. The guide's customer-feedback example asks for "sentiment" (positive/negative/neutral), "key_issues" (list) and "action_items" (list of dicts with "team" and "task"). A template goes further and can fix the structure and length of each field. This market-intelligence prompt puts the input in <data> tags and then shows the exact shape to follow:

<competitor>
  <name>Rival Inc</name>
  <overview>A 50-word summary.</overview>
  <swot>
    <strengths>- Bullet points</strengths>
    <weaknesses>- Bullet points</weaknesses>
    <opportunities>- Bullet points</opportunities>
    <threats>- Bullet points</threats>
  </swot>
  <strategy>A 30-word strategic response.</strategy>
</competitor>

The template uses several techniques from this page at once. It is an example that also serves as a format specification, it uses XML to keep the data separate from the instruction, and it sets explicit length limits ("A 50-word summary", "A 30-word strategic response").

Sources31

Exam traps

Each one states something that sounds right. Open it to see what is actually true.

1. 1.If Claude ignores a rule, restating it in capitals ("NEVER ...") is the best way to enforce it.Why is that wrong?

   The guidance improves the rule by adding its reason. Once Claude understands the goal behind an instruction, it follows it more precisely and can generalize from the explanation.

   Covered in Write the prompt a new colleague could follow

2. 2.Few-shot examples work best when they are near-identical, so Claude sees one consistent pattern.Why is that wrong?

   Examples should vary and include edge cases. Uniform examples can teach patterns you never intended.

   Covered in Steer format and tone with a few well-chosen examples

3. 3.Concatenating several source documents as plain text is fine because Claude will keep them apart on its own.Why is that wrong?

   When a prompt mixes content types, give each its own tag and nest documents in indexed tags. This reduces misinterpretation.

   Covered in Separate instructions, context and inputs with XML tags