How To Prompt Ai For Legal Work

The difference between useless AI output and partner-quality work product is entirely in how you prompt. IRAC-structured prompts, role-based instructions, and few-shot examples turn Claude and ChatGPT from unreliable novelties into genuine productivity tools. Most lawyers who "tried AI and it wasn't good enough" gave it a one-sentence prompt and expected a finished brief.

Legal prompting isn't complicated, but it is specific. The same AI that produces garbage from a vague prompt produces excellent first drafts from a structured one. The gap between a bad legal prompt and a good one is 10x in output quality — and it takes about 30 seconds more effort. Here's the framework that works.

The IRAC Prompting Framework

Lawyers already think in IRAC (Issue, Rule, Application, Conclusion) — use it in your prompts and AI output quality jumps immediately. Structure: "I need a [document type] addressing [Issue]. The applicable rule is [statute/standard/test]. The relevant facts are [Application facts]. Argue toward [Conclusion]." Example: "Draft a motion for summary judgment on the breach of contract claim. The standard is [state]'s three-element test: valid contract, material breach, and damages. Facts: written contract signed 3/15/2024, defendant stopped payments on 8/1/2024 with $45,000 remaining, no cure notice was required under Section 12. Argue that all three elements are met as a matter of law." This prompt produces a structured, usable first draft. "Write me a motion for summary judgment" produces generic filler.

Role-Based Prompting: Tell AI Who It Is

Starting your prompt with a role instruction dramatically improves output quality. "You are a senior litigation associate at a mid-size firm specializing in commercial disputes in [state]." This single sentence changes the AI's vocabulary, tone, formality level, and legal framework. It stops producing generic legal text and starts producing jurisdiction-specific, practice-area-appropriate work product. More specific roles produce better results: "You are a senior associate preparing a brief for Judge [Name] in the Southern District of New York, who is known for preferring concise, fact-focused arguments" produces a dramatically different (and better) brief than no role instruction. Match the role to your actual situation.

Few-Shot Examples: Show AI What Good Looks Like

The most powerful prompting technique for legal work is providing examples. Paste a previous motion you've written and say: "Here's an example of my firm's motion style. Draft a new motion following this same structure, tone, and citation format for the following facts: [new facts]." The AI will match your voice, your formatting conventions, your argument structure, and even your preferred citation style. This works for everything: demand letters, client communications, discovery responses, contract provisions. One example is good; two examples are better; three is diminishing returns. Keep a folder of your best work product to use as few-shot examples — it's the highest-leverage prompting technique that exists.

Context Loading: Give AI Everything It Needs

Claude's 200K context window exists for a reason — use it. The number one mistake in legal prompting is giving AI too little context. Before asking for a motion: paste the relevant statute, the key case authorities (full text, not just citations), and the specific facts. Before asking for a contract clause: paste the full agreement so the AI understands defined terms, existing provisions, and the document's structure. Before asking for a client letter: paste the engagement letter so the AI knows the scope, and the relevant case documents so it knows the details. More context = better output. The attorneys getting the best results from AI are pasting 10-20 pages of context before asking a single question. It feels excessive until you see the output quality difference.

Five Prompts Every Lawyer Should Steal

1. Case analysis: "Analyze this complaint from defendant's perspective. Identify the 5 strongest and 5 weakest allegations. For each weak allegation, suggest a specific defense and the evidence needed to support it." 2. Contract review: "Review this contract from [client]'s perspective. List every provision that creates risk, rate each 1-10 for severity, and suggest specific alternative language for anything rated 7 or above." 3. Motion drafting: "Draft a motion to [type] using this structure: Statement of Facts (narrative), Legal Standard (from the cases I've provided), Argument (IRAC for each element), Conclusion. Cite only the cases I've provided below." 4. Deposition prep: "Based on this complaint and these discovery responses, generate 25 deposition questions for [witness] organized by topic. For each question, note what admission you're seeking." 5. Client letter: "Draft a letter to [client] explaining [legal concept] at a non-lawyer reading level. Use analogies. Keep it under 500 words. End with specific next steps and deadlines."

The Bottom Line: Use IRAC structure, assign a specific role, provide examples of your work, and load full context — these four techniques turn mediocre AI output into genuinely useful legal work product.