AI Prompting Best Practices for Lawyers: Getting Better Outputs

AI can help lawyers move faster, but the quality of the output depends heavily on the quality of the prompt, the quality of the source material, and the lawyer’s review.

The difference between a useful AI-generated work product and a generic response often comes down to the instructions provided at the outset. A vague prompt may produce a vague answer. A structured prompt can produce a more focused, reviewable output that aligns with the legal task at hand.

This principle applies across legal work. Whether a lawyer is researching a motion, reviewing a contract, preparing for a deposition, or analyzing diligence documents, AI performs best when it has clear direction, relevant sources, and a defined objective. Additionally, AI should support legal judgment, not replace it. Lawyers remain responsible for evaluating legal strategy, validating sources, applying professional judgment, and reviewing the final work product.

Purpose-built legal AI platforms such as Harvey are designed for legal work that requires speed, precision, source grounding, and security. But regardless of which platform a lawyer uses, better prompting leads to better outputs.

Providing Clarity is Critical When Prompting AI

Many lawyers approach AI the same way they approach a search engine: they enter a short question and expect a useful result. But legal work rarely operates that way.

When assigning work to a junior associate, lawyers typically provide context about the matter, explain the objective, identify relevant documents, specify the audience, and describe what a successful deliverable looks like. AI benefits from the same level of direction.

The “CLAIM” Framework for Better Legal AI Prompts

Providing clarity upfront reduces the back-and-forth that occurs when an AI system lacks sufficient information to understand the task. Instead of refining multiple generic outputs, lawyers can begin with a prompt designed to produce a more useful first draft.

One simple framework is CLAIM:

• Context: Tell the AI what matter, document, transaction, or dispute it is working on.
• Legal task: Specify the task. Summarize, compare, draft, issue spot, extract, rewrite, critique, or analyze.
• Audience: Identify who will use the output. A partner, client, court, business stakeholder, opposing counsel, regulator, or internal team may require different levels of detail and tone.
• Instructions: Provide constraints, jurisdiction, assumptions, governing law, sources to use, review standards, and any limitations.
• Mode of output: Specify the desired format, such as a table, outline, memo, chronology, issue list, deposition digest, or checklist.

The core principle is simple: vague prompts produce generic answers. Structured prompts produce reviewable work product.

Example 1: Legal Research

Vague prompt:

“Research whether this claim is viable.”

CLAIM prompt:

“You are assisting a US litigation associate. Based on the attached complaint and the cited authorities, identify the strongest arguments for and against a motion to dismiss the breach of fiduciary duty claim under Delaware law. Organize the answer by legal element, cite relevant authority where available, and flag any factual gaps that need attorney review.”

The CLAIM prompt gives the AI a jurisdiction, procedural posture, role, source boundary, output structure, and risk-checking instruction. That makes the answer easier to verify and revise.

Example 2: Contract Review

Vague prompt:

“Review this contract.”

CLAIM prompt:

“Review the attached vendor agreement from the perspective of a US-based enterprise customer. Identify provisions that create unusual commercial, privacy, data security, indemnity, limitation of liability, termination, or assignment risk. Provide a table with clause reference, issue, business impact, suggested revision, and priority level.”

The CLAIM prompt turns a broad review into a structured risk assessment that a lawyer can use to prioritize revisions.

Example 3: Litigation Prep

Vague prompt:

“Give me cross-examination questions.”

CLAIM prompt:

“Using the attached expert report, deposition transcript, and case chronology, draft targeted cross-examination questions for the opposing expert. Focus on assumptions, methodology, inconsistencies with record evidence, unsupported conclusions, and admissions useful to our motion strategy. Group questions by topic and include the source passage that supports each line of questioning.”

Prompt Engineering Best Practices for Legal Work

1. Start With the Legal Role and Objective

AI needs to understand the purpose of the assignment before it can generate useful output. A prompt should identify whether the lawyer needs a quick summary, a first-pass draft, a strategic critique, an issue list, a deposition digest, or a research plan. The clearer the objective, the more relevant the response becomes.

2. Give the AI the Right Source Material

Legal prompts perform best when grounded in source documents. Contracts, pleadings, transcripts, regulations, playbooks, diligence materials, precedent agreements, and internal knowledge repositories provide the factual and legal context necessary for meaningful analysis. The strongest legal AI platforms help lawyers work directly from these materials rather than relying solely on general knowledge.

3. Specify Jurisdiction and Legal Standard

Legal analysis depends heavily on governing law and procedural posture. Prompts should identify the relevant jurisdiction, forum, governing law, transaction type, procedural stage, and legal standard whenever possible. For example, a motion to dismiss under Delaware law requires a different analysis than a summary judgment motion in federal court. Providing this information helps narrow the scope of the response.

4. Ask for a Structured Output

AI does not automatically know how a lawyer intends to use the result. If the desired output is a checklist, chronology, issue list, motion outline, diligence summary, deposition digest, or redline table, the prompt should say so explicitly. Structured outputs are often easier to review, verify, and incorporate into legal workflows.

5. Build in Verification

Lawyers should treat verification as part of the prompt, not an afterthought. Prompts can instruct the AI to:

• Flag uncertainty
• Distinguish facts from assumptions
• Cite source material
• Identify missing information
• Surface follow-up questions
• Highlight areas requiring attorney review

These instructions help create more transparent and reviewable outputs.

6. Iterate Like You Would With an Associate

Few lawyers expect a perfect deliverable after a single conversation with a junior associate. AI should be approached similarly. The first output often provides a foundation that can be refined through additional instructions, clarification, and review. Iterative prompting frequently produces stronger results than attempting to generate a final product in a single request.

Ready-to-Use AI Prompts for Lawyers

The most effective AI prompts combine clear instructions with matter-specific context. The following templates can be customized by replacing the bracketed fields.

Litigation Prompts

Deposition summary prompt

"Summarize this deposition transcript for a [litigation team]. Include key admissions, disputed facts, credibility issues, impeachment points, and follow-up discovery needs. Provide page and line references where available."

Witness credibility prompt

“Analyze this witness statement and deposition excerpt for inconsistencies, evasive answers, unsupported assertions, and facts that require corroboration.”

Cross-examination prompt

"Using the attached [expert report], [deposition transcript], and [case chronology], draft targeted cross-examination questions. Focus on methodology, assumptions, contradictions, and damages. Include the source passage supporting each question."

Motion strategy prompt

"Based on the [complaint], [answer], [key exhibits], and controlling law in [jurisdiction], identify the strongest arguments supporting a motion for [summary judgment/motion to dismiss] and the factual weaknesses opposing counsel is likely to attack."

Case chronology prompt

"Create a chronological timeline from these [emails], [contracts], [pleadings], and [deposition excerpts]. Include date, event, source document, legal relevance, and disputed or undisputed status."

Authority-checking prompt

“Review this draft brief section. Identify unsupported propositions, missing authority, overstatements, and arguments that need stronger factual support.”

Transactional and M&A Prompts

Review purchase agreement prompt

"Review this [purchase agreement] against the [letter of intent]. Identify deviations, retrades, negotiation issues, and provisions requiring business review."

Key provision extraction prompt

“Extract change-of-control, assignment, indemnity, termination, and consent provisions from these material contracts.”

Due diligence issue list prompt

"Analyze the attached diligence materials and create an issue list that includes risk level, business impact, supporting document reference, and recommended follow-up request."

Disclosure schedule drafting prompt

"Draft a first-pass disclosure schedule section using the attached [agreement] and [data room documents]. Identify assumptions and information gaps requiring attorney review."

In-House Legal Prompts

Outside counsel memo summary prompt

"Summarize this outside counsel memorandum for a [business stakeholder]. Explain key risks, recommended next steps, and open questions in plain English."

Vendor contract review prompt

"Review this vendor agreement against our playbook. Flag deviations from preferred positions, explain business impact, and suggest alternative language."

Regulatory update drafting prompt

"Draft a business-friendly explanation of this regulatory development for [product], [sales], and [operations] teams. Focus on practical implications and recommended actions."

How to Prompt AI Without Compromising Confidentiality

A Practical Confidentiality Checklist

Prompt quality matters, but so does protecting confidential information. When using general-purpose AI tools, lawyers should carefully evaluate what information they include in prompts. Even when a task appears routine, prompts may contain privileged information, client identifiers, deal details, personal information, or internal strategy that should not be shared broadly.

A practical confidentiality checklist includes:

• Remove or mask client names, employee names, deal names, account numbers, and personal identifiers whenever possible
• Replace sensitive information with placeholders such as "[Client]," "[Counterparty]," "[Witness A]," or "[Agreement Date]"
• Limit prompts to the information necessary to complete the task
• Avoid entering privileged strategy, internal legal advice, or highly sensitive business information into consumer-grade AI tools
• Understand how the provider stores, retains, and uses data
• Keep lawyers responsible for reviewing all outputs before they are used in client work

Purpose-built legal AI platforms are designed to address these concerns.

Harvey supports customer data isolation, encryption at rest and in transit, role-based access controls, audit logs, matter-level governance, and independently audited security standards including SOC 2 Type II and ISO 27001. Harvey does not use client data to train AI models, and organizations can maintain control over permissions, retention policies, and access rules.

Strong prompting practices and strong governance should work together. Legal teams should evaluate both before incorporating AI into client-facing workflows.

How Legal AI Platforms Improve Prompting Beyond One-Off Prompts

Better prompting matters. Using the right platform matters too. General-purpose AI tools can assist with drafting and analysis, but legal work often requires much more than a single prompt and response.

Lawyers frequently need to work across large document sets, internal knowledge repositories, client playbooks, precedent libraries, and matter-specific materials. They need outputs grounded in authoritative sources and workflows that can be repeated across matters.

Purpose-built legal AI platforms help address these challenges. Legal teams can:

• Ground outputs in legal authorities, internal precedent, templates, and matter documents
• Analyze large document collections rather than isolated excerpts
• Build repeatable workflows instead of recreating prompts from scratch
• Work directly within familiar tools such as Word, Outlook, SharePoint, and document management systems
• Maintain permissions, governance controls, and auditability

Harvey helps legal teams move beyond one-off prompting by connecting AI assistance to trusted legal sources, institutional knowledge, and repeatable workflows.

For example, litigation teams can generate early case theories, analyze authorities, draft motions grounded in cited sources, and identify responsive documents across large evidentiary records. Transactional teams can review purchase agreements against letters of intent, generate diligence insights, draft disclosure schedules, and analyze negotiation positions across multiple deal documents.

Workflow Agents further allow organizations to build customized processes that reflect firm-specific playbooks, standards, and review methodologies. Instead of relying on individual prompting expertise, organizations can embed institutional knowledge directly into repeatable workflows.

A Prompt Template Lawyers Can Reuse

The following template can serve as a starting point for many legal tasks:

“You are assisting a [type of lawyer/team] working on a [matter type] in [jurisdiction]. Using [documents/sources], please [task]. Focus on [issues]. Do not [limitations]. Format the output as [table/memo/checklist/outline]. Include [citations/source references/factual assumptions/open questions]. Flag any uncertainty or areas requiring attorney review.”

Better Prompts, Better Legal Work

Lawyers should use AI to accelerate work, surface issues, and reduce blank-page friction — not to outsource legal judgment.

The greatest value comes when strong prompting practices are paired with secure systems, trusted legal sources, and repeatable workflows.

Harvey helps legal teams draft, research, analyze documents, and build repeatable AI workflows grounded in legal sources and institutional knowledge. See how Harvey can help your organization work more efficiently while maintaining the rigor legal work demands.