The Core Framework
Every good legal prompt has five main parts:1. Task Definition
Your opening instruction needs to be specific about what action you want and what you’re looking for. Vague requests lead to inconsistent results. Too Vague: “Review this agreement” Better: “Identify and redline any unlimited liability provisions in this vendor agreement” Start with action verbs (identify, redline, extract, compare, draft) and be specific about scope. Instead of “financial terms,” say “payment terms only.” When possible, explain why it matters: “…that could expose us to uncapped risk.”2. Context Anchoring
The AI needs background information to make good decisions. Think of this as the setup you’d give a new team member before they review a contract.Essential Context Elements
- Party Position: Who are you in this deal?
- “I am the vendor reviewing a customer’s MSA”
- “We are the covered entity under this BAA”
- Document Type: What kind of agreement is this?
- “SaaS subscription agreement”
- “Mutual NDA for partnership discussions”
- Industry & Regulations: What special rules apply?
- “Healthcare company subject to HIPAA”
- “Government contractor under FAR/DFARS”
- Leverage: What’s your negotiating position?
- “Low leverage - competitive RFP situation”
- “High leverage - we’re their sole supplier”
3. Constraints
Constraints are the boundaries you set to prevent over-editing or off-target responses. Think of these as the guidelines you’d give to keep someone from going overboard with changes.- Edit Style: How should changes be made?
- “Use qualifiers rather than deletions”
- “Add new provisions as separate sections”
- Scope Limits: What should be ignored?
- “Focus only on payment and liability provisions”
- “Ignore stylistic or formatting differences”
- Quantity Controls: How many changes are appropriate?
- “Limit to 5 substantive changes maximum”
- “Flag all issues but only redline critical ones”
4. Output Format
Tell the AI exactly how you want the results presented. Different tasks need different formats. For Analysis:- “Create a table with columns: Clause | Risk Level | Issue | Recommendation”
- “Provide numbered list with clause references”
- “Show changes with explanatory comments”
- “Include fallback language for each edit”
- “Draft email to counterparty explaining our position”
- “Create executive summary in bullet points”
5. Audience
Who will use this output? The audience shapes both tone and detail level.- Legal team: Can handle technical legal language and detailed analysis
- Executives: Need business impact and risk summary, not legal theory
- Counterparty: Requires professional tone with justified positions
- Business teams: Need plain English explanations of practical implications
Putting It All Together
Here’s a complete prompt using all five elements:Advanced Techniques
Multi-Step Prompting: Break complex tasks into steps. First identify issues, then assess risk, then draft fallbacks, then add comments. Negative Prompting: Explicitly state what you DON’T want. “Do NOT change payment terms or add new obligations for counterparty.” Role-Based Perspective: Add a viewpoint to shape analysis. “As conservative inside counsel, flag any provision that could require board approval.”Common Mistakes and Fixes
| Problem | Solution |
|---|---|
| Getting different results each time you run the same prompt | Add more specific constraints and clearer context |
| AI makes too many changes or wrong types of changes | Use negative prompting to exclude what you don’t want |
| Output is too technical or too basic for your audience | Always specify who will use the output |
| Results don’t match your negotiating position | Include leverage and relationship context |
Quick Reference
When writing any prompt, check that you’ve covered:- ✓ Clear task with action verb
- ✓ Your party position and document type
- ✓ Any relevant regulations or industry norms
- ✓ Your leverage level
- ✓ How many changes to make
- ✓ What format you want
- ✓ Who will read this