How to prompt Claude

Show your prompt to a colleague who has none of your background. A prompt that would confuse them will confuse Claude as well. The guide describes Claude as a new employee who lacks context, so be explicit about what you want.

Three habits follow from the test:

• Start with a verb. Anthropic's blog says to lead with a direct action verb such as "Write" or "Analyze".
• Give the reason. When you say why a rule exists, Claude can apply it to cases you did not list. The blog adds that purpose and audience help the model decide.
• Say what to do. Tell Claude what you want, not only what to avoid.

Scrubbr can write that background in from your own documents, so the new colleague arrives already briefed.

Examples. The guide calls examples one of the most reliable ways to steer format, tone and structure. It says to make them relevant, diverse and clearly marked, and suggests three to five inside example tags. The blog says to begin with a single example and add more only if the answer still misses. For a first try, one is enough.

Tags. An XML tag is a pair of words in angle brackets that wraps a piece of text, such as a pair named "document" around a pasted contract. The guide says tags for instructions, context and input reduce misreading when one prompt holds all of them. It also says to use the same tag names every time.

When tags are worth it. Use them when your prompt holds several documents, or an example next to a task. For a three-line prompt, plain labels on their own lines do the same job. The cut-off for a short prompt is ours, since the guide gives none.

• Long documents go first. Put long material at the top, above your question. Anthropic's guide says this for documents of about 20,000 tokens or more, which is many pages.
• One sentence of role is enough. A role in the system prompt focuses the tone, and even a single sentence helps.
• Permission to be unsure. Tell Claude it may say it is not sure. Anthropic's blog reports fewer made-up answers when you do.
• Define success first. Anthropic's overview assumes you already know what a good result looks like and how you will test for it. In the six-part prompt, that is the "great answer" line.
• Format and limits in words. Be specific about the output format and the constraints. Use numbered steps when order or completeness matters.

The prompting hub has the common core, and the habits above are the ones Claude's guides stress.

Sources checked on 2026-11-03.

• Prompt engineering overview (Anthropic)
• Prompting best practices (Anthropic)
• Best practices for prompt engineering (Anthropic blog)