Quick Answer
Use prompt patterns that make Claude ask, inspect, cite, draft, and review instead of guessing.
Primary-source note: Anthropic’s usage guidance ties quality and limits to conversation length, context, models, and feature use, so prompts should provide enough context without dumping unrelated material.
The six-part everyday prompt
A reliable Claude work prompt has six parts: role, source material, task, constraints, output format, and review criteria. You do not need fancy phrasing. You need enough context for Claude to know what good looks like and where the boundaries are.
Example: “Act as a careful operations assistant. Use only the pasted thread. Summarize decisions and draft a reply. Do not promise dates or discounts. Return bullets plus a draft under 140 words. Flag anything that needs my confirmation.”
- Role: what viewpoint should Claude use?
- Source: what material may Claude rely on?
- Task: what output should exist at the end?
- Constraints: what must be avoided?
- Format: what shape should the answer take?
- Review: how should Claude check its own work?
Source check: How do usage and length limits work? from Claude Help Center.
Use inspection prompts before creation prompts
For files, threads, datasets, and research, the first prompt should often be inspection, not creation. Ask Claude what the material contains, what is missing, what is ambiguous, and what it needs to know before producing the final output.
This pattern works because it turns hidden assumptions into visible review points. It also reduces the chance that Claude will write confidently from incomplete material.
Use review prompts after creation prompts
After Claude drafts an email, report, spreadsheet formula, or agenda, ask for a review pass. The review should check facts, tone, unsupported claims, sensitive data, ambiguity, and whether the output satisfies the original request.
For important work, ask Claude to argue against its own draft: “What could be misunderstood? What claim is weakest? What would a skeptical reader challenge?”
Keep prompts adaptable
A reusable prompt should have placeholders, not hard-coded facts. Write [audience], [source], [deadline], [tone], and [format] so you can reuse it. If you use Projects, store the stable pattern there and leave the fresh facts for each chat.
When a prompt produces a bad answer, revise the instruction that failed. Did Claude need a source limit, a shorter output, a clearer audience, or a review checklist? Treat prompts like working procedures, not magic spells.
Prompts to Copy
Before answering, ask up to five clarifying questions only if they change the output. If not, state assumptions and proceed.
Use only the sources I provide. If a claim is not supported, mark it as assumption, inference, or needs verification.
Review your draft for factual accuracy, unsupported claims, tone risk, privacy risk, and missing next steps. Then revise once.
Create a reusable template for this workflow with placeholders and a final human-review checklist.
Cite this page
Claude Helps, "Claude Prompt Library for Everyday Work", https://claudehelps.com/prompt-library (updated 2026-07-06).
Primary Sources
- How do usage and length limits work? Claude Help Center
- Upload files to Claude Claude Help Center
- Use research on Claude Claude Help Center
- What are projects? Claude Help Center
FAQ
What is the best general Claude prompt format?
Use role, source material, task, constraints, output format, and review criteria. Add examples only when style or structure matters.
Should I ask Claude to ask questions first?
Yes when missing information would change the output. Otherwise, ask Claude to state assumptions and proceed.
How do I make Claude cite sources?
Tell Claude which sources it may use and require each factual claim to cite a source or be marked as unsupported.