AI prompting guidelines

For any role interacting with an AI agent

These principles help you get faster, more accurate, and more useful AI outputs — whether you’re building a UI, reviewing a contract, generating reports, or performing underwriting tasks.

1. Be specific, not vague

Bad: Summarize this case 
Better: Summarize this insurance claim in 3 bullet points, highlighting claim amount, cause, and recommended next steps

AI works best when you provide clear details, scope, and format.

2. Use real-world references

Instead of saying “be creative” or “do it better,” anchor your request in existing examples: 
Format this report like last quarter’s summary but add a risk analysis section

AI understands known patterns and formats more easily than abstract requests.

3. Break large requests into smaller parts

Good: Draft the introduction section of the underwriting report 
Avoid: Write the full underwriting report in one go

Smaller, well-defined tasks produce more accurate, focused results.

4. Use specific, relevant keywords

Generic words = generic results.

• Instead of clear → concise and bullet-pointed
• Instead of professional → formal tone, industry-standard terminology
• Instead of summary → executive summary with 3 key risk indicators

Specific keywords help AI lock onto the exact style and scope you want.

5. Include domain context

Generate an underwriting checklist for commercial property insurance > Generate a checklist

The more you share about your industry, role, or process, the more relevant and accurate the output.

6. Collaborate and iterate

Share drafts and partial outputs with colleagues or stakeholders.

Feedback from others often improves AI results faster than refining prompts alone.

7. Define output for different use cases

Don’t just say “make it suitable for mobile” or “make it readable.” Be explicit:

• For email: short, plain language, under 150 words
• For report: formatted in table with columns for risk, score, and notes

Clarity on final format ensures the output fits directly into your workflow.

8. Follow the 80/20 prompt rule

Most high-quality results come from defining:

1. Task or deliverable type
2. Style or tone
3. Domain or context
4. Structure or format

9. Give actionable feedback

Avoid: Make it better 
Use: Add a section on claims history, remove duplicate entries, and use industry-standard risk categories

Treat AI like a capable assistant — clear, step-by-step feedback gets better results.

10. Build in layers

Instead of a single, complex prompt, add details over several iterations:

1. Draft a standard policy review checklist
2. Add risk scoring methodology
3. Include compliance requirements for commercial properties

Iterative prompting keeps outputs accurate and adaptable.

Key takeaway

AI won’t replace skilled professionals — it enhances those who think in systems, structure, and intent.