The best prompts for Claude

Claude is the AI assistant made by Anthropic. It is available as a chat app, inside developer tools and through an API, and the same prompting habits carry across all of them.

Anthropic's documentation offers a useful mental model: write for a very capable new colleague who has no background on your project. That person would need to know what the work is for, who will read it, what a good result looks like and what to avoid. A prompt that answers those questions usually beats a short command, and it beats a long prompt stuffed with emphasis and repeated warnings.

There are two places to put guidance. A system prompt, or in the app the instructions attached to a project and your saved preferences, sets a role and standing rules. The conversation itself carries the task. Reference material can be pasted, uploaded, or stored as project knowledge so it is available in every chat within that project.

Claude handles long inputs well when they are organized. Anthropic recommends wrapping distinct parts of a prompt in XML-style tags, for example one tag around a contract and another around your instructions, so each part is unambiguous. The tag names are yours to choose; what matters is that they are consistent and descriptive. The labeled sections used in the prompts on this site serve the same purpose and can be converted to tags if you prefer.

For complicated work, one large prompt is not always the right tool. Splitting a job into a sequence, where the output of one step becomes the input of the next, makes each step easier to check and correct.

Specific model names, limits and plan features change frequently and are left out here on purpose. Anthropic's documentation is the place to confirm current details.