Claude Writing Prompts: Reusable Patterns for Better Drafts

Use the seven-part prompt pattern

A strong Claude prompt has a repeatable anatomy. It names the role, provides context, defines the exact task, states the source boundary, specifies the output format, gives review criteria, and tells Claude when to stop.

This pattern is useful because it prevents the assistant from optimizing for the wrong thing. It also makes your own thinking visible. If you cannot fill in a field, the problem is not prompting - the writing project is underdefined.

When the input is long, place documents and source packets first, then the question and instructions near the end. Anthropic documents this as a long-context prompting practice.

• Role: editor, developmental reader, claims reviewer, continuity clerk.
• Context: brief, source packet, audience, project constraints.
• Task: one bounded job, not all writing work at once.
• Boundary: what Claude may use and what it must not invent.
• Format: table, memo, outline, margin notes, rewrite, checklist.
• Criteria: how the answer will be judged.
• Stop: ask before drafting or before rewriting the whole piece.

Brief diagnosis prompt

Use this when you have an idea but not a clean plan. It is deliberately diagnostic: Claude should improve the thinking before it writes.

Source-grounded draft prompt

Use this when factual accuracy matters. Claude should quote or cite source notes first, then draft only from those notes.

Fiction scene pressure prompt

Use this before writing a scene. It keeps Claude focused on dramatic movement rather than decorative language.

Line edit prompt

Use this when the draft exists and you want accountable edits rather than a blanket rewrite.

Turn repeated prompts into a skill

If you keep pasting the same writing checklist, move it into a skill or project instruction. Anthropic describes skills as folders of instructions, scripts, and resources for specialized repeatable tasks.

For writers, a useful skill can hold a house style guide, editorial rubric, approved source hierarchy, and examples of acceptable revisions. Keep it narrow: one repeatable job per skill.