Prompting for Operators

Last verified: 13 February 2026 | Applies to: All plans

In 30 seconds

Prompting is the single highest-leverage skill you can develop as an operator using Claude. The difference between a vague prompt and a structured one is the difference between “this is okay, I guess” and “this is exactly what I needed.” This page covers practical patterns — no academic theory, just what works for business use.

The operator’s prompting framework

Most prompting guides are written for developers. Operators need different patterns. Here’s the framework:

1. State the role and context

Tell Claude who it’s working as and what it knows about your situation.

Weak: “Write a status update.”

Strong: “You’re drafting a weekly status update for my CEO. We’re a 15-person ecommerce company mid-way through a platform migration. The CEO cares about bottom-line impact first, then risks.”

2. Specify the output format

Don’t let Claude guess what you want. Describe the format explicitly.

Weak: “Summarise these meeting notes.”

Strong: “Summarise these meeting notes as: (1) decisions made, (2) action items with owners and deadlines, (3) open questions. Use bullet points. Keep it under 300 words.”

3. Give examples of good output

If you have a previous version of what you want, share it. Claude learns from examples faster than from descriptions.

Weak: “Write our quarterly report.”

Strong: “Write our Q1 report. Here’s last quarter’s report for format and tone — [paste previous report]. Update with this quarter’s data: [paste data]. Keep the same structure but add a section on the platform migration.”

4. Set constraints

Tell Claude what NOT to do. Constraints prevent the most common issues.

Weak: “Draft an email to the client.”

Strong: “Draft an email to the client about the project delay. Keep it under 150 words. Don’t apologise more than once. Don’t promise a specific new date — say we’ll confirm by end of week. Tone: professional, direct, not defensive.”

5. Ask for options, not answers

When you need Claude to think, ask for multiple approaches rather than a single answer.

Weak: “How should we handle this customer complaint?”

Strong: “Give me three options for handling this customer complaint, with pros and cons for each. Consider: (1) the customer has been with us for 3 years, (2) the issue was our fault, (3) the fix costs us about $500.”

Patterns that work for operators

The briefing pattern

Use this when you need Claude to process information and give you the essential takeaways.

Read [document/data/email thread].

Brief me on:
- What's the key issue?
- What decision do I need to make?
- What are my options?
- What would you recommend and why?

Keep it under 200 words. Lead with the recommendation.

The delegation pattern

Use this when handing Claude a task you’d normally delegate to a team member.

I need you to [task].

Context:
- [Who this is for]
- [What they care about]
- [Any constraints — budget, timeline, tone]

Deliverable: [exactly what you want back]

Quality check: Before you deliver, verify [specific things to check].

The review pattern

Use this when you want Claude to critique or improve existing work.

Review this [document/plan/email] for:
1. [Specific thing to check — e.g., "factual accuracy"]
2. [Second thing — e.g., "tone appropriateness for the audience"]
3. [Third thing — e.g., "anything I'm missing"]

For each issue found, suggest a specific fix.
Don't rewrite the whole thing — just flag and fix.

The extraction pattern

Use this when you need structured data from unstructured content.

Extract the following from [document/email/transcript]:

| Field | Description |
|-------|------------|
| [field 1] | [what to look for] |
| [field 2] | [what to look for] |
| [field 3] | [what to look for] |

Output as a table. Flag anything that's unclear or missing with [UNCLEAR].

The iteration pattern

Use this when the first output isn’t quite right — instead of starting over.

This is close. Adjust:
- [Specific change 1]
- [Specific change 2]
- Keep [thing that worked] as-is

This is almost always better than reprompting from scratch. Claude remembers the context and makes targeted changes.

Common mistakes operators make

Too vague: “Help me with my financials.” Claude doesn’t know if you want a budget, a forecast, a reconciliation, or a variance analysis. Be specific about the task.

Too much context, not enough instruction: Pasting an entire document with “What do you think?” gives Claude no direction. Tell it what to look for.

Not using artifacts: If you ask Claude to “create a spreadsheet” in chat, you may get text. Ask Claude to “create a spreadsheet as an artifact” or “create a downloadable Excel file” and you’ll get a file.

Restarting instead of iterating: If the first response is 70% right, don’t reprompt from scratch. Tell Claude what to fix. Iteration is faster and produces better results.

Forgetting to specify the audience: “Write a summary” without saying who it’s for produces generic output. “Write a summary for my CFO who wants specific numbers, not ranges” produces focused output.

Power moves

Chain prompts for complex work: Break large tasks into steps. “First, analyse the data and identify the top 3 trends. Then, for each trend, draft a paragraph explaining the business impact. Finally, compile into a one-page brief.”

Use Claude’s memory: If you’ve set up the Productivity plugin with Workplace Memory, Claude already knows your team and context. You can skip repeating background info and go straight to the task.

Save your best prompts as skills: If you find a prompt pattern that works well, turn it into a custom skill. That way Claude uses it automatically — you don’t have to remember the prompt each time. See Skills.

Use extended thinking for complex analysis: In Chat, toggle on extended thinking when you need Claude to reason through a complex problem. You’ll see its thinking process, which helps you verify the logic — not just the conclusion.

Workplace Memory — set up context so you don’t have to re-explain
Skills — save your best prompt patterns as reusable skills
Models — match the right model to the right task
Chat — where most prompting happens

Something wrong or outdated? Let us know →

Get weekly workflows — subscribe to the newsletter.