The Lawyer's Guide to AI Prompt Engineering: Get Reliable Output Every Time

Why Prompt Engineering Matters for Lawyers

The single biggest mistake we see lawyers make with AI is treating it like Google. They type three words, hit enter, get a generic answer, and conclude that "AI isn't there yet for legal work."

AI is there. Your prompts aren't.

Prompt engineering - the practice of structuring requests to get reliable, professional output - is the most important AI skill a lawyer can develop. It's the difference between AI that drafts garbage and AI that drafts work product you'd actually file.

This guide walks through the exact prompt structure we teach in our AI training program, with before-and-after examples from real legal use cases.

The Five-Part Prompt Framework

Every effective legal prompt has five components. Miss any one and quality drops sharply.

1. Role

Tell the AI who it is.

Weak: "Write a demand letter."

Strong: "You are a senior personal injury attorney with 15 years of experience in California. You write demand letters that are firm but professional."

2. Context

Give the AI the facts it needs.

Weak: "Client was in a car accident."

Strong: "Client is a 42-year-old paralegal who was rear-ended at a red light on March 3, 2026. She suffered cervical strain, a concussion, and missed 6 weeks of work. Total medical bills: $18,400. Lost wages: $9,200. Defendant's insurance has acknowledged liability."

3. Task

Specify exactly what you want produced.

Weak: "Help with the letter."

Strong: "Draft a demand letter to State Farm requesting policy limits of $100,000."

4. Format

Tell the AI how to structure the output.

Strong: "Format: header with parties and date; one paragraph each for liability, damages, and demand; close with a 30-day deadline. Maximum 1.5 pages."

5. Examples and Constraints

Show what good looks like, and what to avoid.

Strong: "Match the tone of the attached prior demand letter. Do not use 'in light of the foregoing' or 'be that as it may.' Cite no statutes or cases - I'll add citations after."

The Verification Pattern

Every legal prompt should end with one more line: "After drafting, list any factual assumptions you made and any places where I should verify accuracy."

This single addition reduces hallucination risk by an order of magnitude. The AI surfaces its own uncertainties, which is exactly what you need to catch errors before they reach a client or court. See our deeper analysis in AI hallucinations in legal work.

Before and After: A Real Example

Real prompt from a litigation associate.

Before: "Summarize this deposition."

The AI returned a 400-word paraphrase of the table of contents. Useless.

After: "You are a senior trial attorney preparing for cross-examination. Below is the deposition transcript of the plaintiff's expert witness. Identify: (1) the three weakest points in his testimony, (2) any internal contradictions, (3) any statements that conflict with his published work (pasted below), and (4) suggested follow-up questions for cross. Format as a bullet list under each heading."

Result: a usable cross-examination outline in 90 seconds. The associate billed 0.2 hours instead of 4.0.

Five Prompt Templates Every Lawyer Should Save

Copy these into a note in your phone or password manager. Reuse them daily.

1. The Triage Prompt: "Categorize the following [emails / documents] as URGENT, ROUTINE, or DEFERRED. For URGENT items, draft a one-paragraph response I can review."

2. The Summary Prompt: "Summarize the following [document type] in 5 bullet points covering: the legal question, the facts, the holding, the reasoning, and any dissents or limitations."

3. The Comparison Prompt: "Compare the following two [contracts / depositions / cases]. Identify (a) areas of agreement, (b) areas of conflict, (c) anything missing from one but present in the other."

4. The Drafting Prompt: "Draft a [document type] in the style of the attached example. Use the facts below. Tone: [professional / firm / collegial]. Length: [target]. Flag any places where you needed to make assumptions."

5. The Verification Prompt: "Review the following draft for: (1) factual claims I should verify, (2) legal citations that may be inaccurate, (3) any logical gaps, (4) tone inconsistencies. Do not rewrite - only flag."

What NOT to Put in Prompts

Privileged client information should never go into consumer AI tools that train on user data. For confidential matters, use:

• An enterprise plan (ChatGPT Team or Enterprise, Claude Team or Enterprise, Microsoft Copilot for Business)
• A private deployment through services like Claude Connect
• A dedicated legal AI platform with appropriate BAAs

Read our full breakdown in AI ethics for lawyers before pasting anything sensitive.

The Iteration Loop

Your first prompt is rarely your best prompt. Pros iterate:

1. Run the prompt.
2. Note what's wrong with the output.
3. Add a constraint or example to the prompt.
4. Run it again.
5. Save the working version.

Over six months, you'll build a library of 30 to 50 prompts that handle 80% of your daily work. That library is the actual productivity gain.

Where to Go Next

If you're new to prompting, start with the five templates above on low-stakes tasks (email triage, meeting prep) before moving to anything client-facing. Build the habit first; then introduce it to higher-risk work.

If your firm wants a structured rollout, our AI training program builds firm-specific prompt libraries and trains attorneys on iteration.

And if you want to see how prompt engineering scales into an entire AI operating system, see Building a Complete AI Operating System for Your Law Firm. For the daily habits that turn good prompts into recovered billable hours, see 7 daily AI workflows every lawyer should master.

Better prompts. Better output. Same brain.