The Attorney's Guide to Prompt Engineering

Master prompt engineering for legal AI. Learn how to write AI instructions that actually work for contract review, research, and client communication.

By Michael Clendening, Founder of EverIntent | November 3, 2025 | 10 min read

Executive Summary

Most attorneys get mediocre results from AI because they treat it like a search engine instead of a trained associate. This guide teaches the 4-part prompt framework that transforms vague requests into precise, reliable legal outputs. You'll learn zero-shot vs few-shot techniques, chain-of-thought reasoning for complex analysis, and real examples that work for contract review, client emails, and legal research. By the end, you'll know exactly how to write prompts that save 45 minutes per task.

Why Most Attorneys Get Bad AI Results

You open ChatGPT. You type "review this contract." You paste 50 pages of dense legalese. The AI gives you a three-paragraph summary that misses the critical indemnification clause and halluccinates a termination provision that doesn't exist.

Sound familiar? The problem isn't the AI. It's the prompt.

Legal AI is exceptionally powerful—but only when given proper instructions. Think of it like delegating to a brilliant first-year associate who knows nothing about your case, your client, or your goals. If you say "review this contract," they'll skim it and guess at what you care about. But if you provide context, specify the task, define the format, and set constraints, they'll deliver exactly what you need.

This is prompt engineering: the art and science of instructing AI to produce reliable, professional-grade legal work.

The Perfect Legal Prompt Framework

Every effective legal prompt has four components:

1. Context

Who you are, what you're working on, and why it matters. Give the AI your role, practice area, jurisdiction, and the situation.

2. Task

What you want the AI to do. Be specific: analyze, summarize, draft, compare, identify risks, extract clauses.

3. Format

How you want the output structured. Bullet points, numbered list, table, memo format, email draft.

4. Constraints

Boundaries and requirements. Cite specific sections, flag risks above certain thresholds, limit length.

Real Prompts That Work

Contract Risk Analysis Example

You are a commercial litigation attorney reviewing a vendor agreement for a mid-size tech company client.

TASK: Analyze the attached contract and identify all provisions that create litigation risk or limit our client's rights.

FORMAT: Provide a numbered list with: Clause title and section number, Risk level (High/Medium/Low), Specific concern, Recommended modification.

CONSTRAINTS: Focus on indemnification, limitation of liability, dispute resolution, and IP ownership. Flag any clause that limits remedies to less than direct damages. Cite exact section numbers. Keep each item to 2-3 sentences.

Client Email Response Example

You are a family law attorney. A client just emailed asking: "Can my ex change our custody schedule without my permission?"

CONTEXT: This is in California. We have a court-ordered custody agreement. The client tends to be anxious and needs clear, calm guidance.

TASK: Draft a professional email response that: 1) Directly answers their question, 2) Explains the legal standard in simple terms, 3) Outlines next steps, 4) Reassures without over-promising.

CONSTRAINTS: Keep under 200 words. Use 8th grade reading level. Include a clear call-to-action. Maintain warm but professional tone.

Advanced Techniques: Zero-Shot vs Few-Shot Prompting

Zero-shot prompting is when you give the AI a task without examples. It works well for common legal tasks the AI has seen many times: "Summarize this contract" or "Draft a demand letter."

Few-shot prompting is when you show the AI 1-3 examples of what you want before asking it to perform the task. This is critical for:

  • Your firm's specific document style
  • Unusual formatting requirements
  • Jurisdiction-specific citation formats
  • Client communication tone

Chain-of-Thought Prompting for Complex Analysis

For complex legal reasoning, explicitly ask the AI to "think step-by-step" or "show your analysis." This technique—called chain-of-thought prompting—dramatically improves accuracy on multi-step problems.

Example: "You are reviewing a merger agreement to determine if our client can terminate due to a material adverse change (MAC). Think step-by-step through the analysis. Consider the exclusions and whether they apply. Apply Delaware MAC case law standards. Give a probability assessment."

5 Common Prompt Mistakes (and How to Fix Them)

  1. Being too vague: "Review this" → "Identify all force majeure provisions and assess enforceability under New York law"
  2. No format specification: Add "Provide as numbered list" or "Draft as client email"
  3. Ignoring constraints: "Keep under 300 words" or "Cite to specific contract sections"
  4. Not providing context: Include jurisdiction, practice area, client type
  5. Single-step for multi-step tasks: Break complex tasks into sequential prompts

Takeaways

  • Use the 4-part framework: Context + Task + Format + Constraints
  • Show examples for firm-specific style (few-shot prompting)
  • Use "think step-by-step" for complex legal analysis
  • Test and iterate—save your best prompts for reuse
  • Private AI environments are essential for client confidentiality