We respect attorney-client confidentiality. No tracking pixels in our emails.
A structured guide for associates on AI-assisted legal memo drafting—IRAC prompting, citation verification, maintaining attorney voice, supervision workflow, and billing ethics.
2026/08/04
A first-year associate at a regional firm was assigned a research memo on personal jurisdiction in a products liability case with a two-day deadline. She used an AI assistant to generate an initial draft, reviewed it, corrected two citations, and submitted the memo to the supervising partner. The partner sent it back with a note: the memo had analyzed the wrong standard—it addressed general jurisdiction, not specific jurisdiction, which was the actual question. The associate had accepted the AI's framing of the issue without independently reading the complaint.
That failure was not an AI failure. It was a supervision failure—the associate delegated issue identification to the tool without verifying that the tool had correctly understood the question. The AI wrote a technically competent memo on the wrong subject.
For associates learning to use AI in legal memo writing, the most important skill is not knowing which button to press. It is knowing which steps require independent attorney judgment and which are appropriately assisted by AI. This guide walks through every stage of memo drafting with that distinction at the center.
The legal memo is the foundational work product of associate training. Writing memos forces the development of skills that define effective legal analysis: framing the issue precisely, identifying the applicable rule, applying the rule to specific facts, and reaching a defensible conclusion. Partners and supervising attorneys use memo feedback to evaluate associate judgment—not just research competence.
The arrival of AI writing assistants created an immediate tension in associate development. Tools like Harvey AI and CoCounsel can produce a plausible-looking legal memo in minutes. Associates under deadline pressure face the temptation to accept AI output with minimal review. Some do. The consequences—sending a memo that analyzes the wrong issue, cites inapplicable cases, or reaches a conclusion that ignores the client's actual facts—are predictable.
But the answer is not to prohibit AI assistance. Associates who learn to use AI tools effectively produce better research, more comprehensive issue coverage, and higher-quality first drafts. The skills required—precise prompting, critical evaluation of AI output, independent verification of citations—are themselves valuable legal skills that develop through practice.
The key is understanding where AI adds value in the memo workflow and where it introduces risk. AI hallucination risk is highest when AI is asked to do legal analysis without being given the relevant authorities. It is substantially lower when AI is asked to synthesize authorities you have already identified and verified.
Prompt construction is where most associates go wrong. Prompt engineering for legal memos is not a technical skill—it is the application of legal analysis skills to the problem of communicating with an AI model. The better you understand IRAC structure, the better your prompts will be.
Firm policies on AI use in associate work are still evolving. Some firms have deployed approved AI tools with training; others rely on general guidance without specific tools. Associates should understand their firm's current policy, including which tools are approved for use with client matters and what disclosure obligations apply.
The most reliable approach to AI-assisted memo drafting is to use IRAC structure as a prompting sequence—separate prompts for each component rather than a single prompt asking for a complete memo. This approach produces better output and maintains clearer attorney control over each stage.
Issue prompt: "I am writing a legal memo on the following question: [paste the precise legal question as framed by the supervising attorney]. Based on the following case facts: [paste relevant facts], identify the specific legal standard at issue and whether there are any threshold questions that must be resolved before reaching the main question."
This prompt forces you to think about whether AI has correctly framed the issue before you proceed to research. If the AI identifies a different standard than you expected, investigate the discrepancy before proceeding—it may mean the question was ambiguous, or AI may have misread the facts.
Rule prompt: "Summarize the controlling rule in [jurisdiction] for [legal standard]. Identify the key elements, the leading cases, and any circuit splits or evolving areas. Use only cases from [citation database] results I provide: [paste verified research results]."
By feeding AI your own verified research rather than asking it to self-generate, you eliminate hallucination risk in the rule synthesis. AI then synthesizes the cases you have already verified—a task it performs well and accurately.
Application prompt: "Apply the rule as stated above to the following specific facts: [paste key facts]. For each element of the rule, analyze whether the facts satisfy it, citing the specific cases that are most on point. Do not introduce cases not included in the research I have provided."
The application section is where AI errors are hardest to detect because they look like analysis. AI may reach plausible-sounding conclusions that are analytically wrong. This section requires the most careful attorney review.
Conclusion prompt: "Based on the rule and application above, draft a conclusion section that states the likely outcome and identifies any significant uncertainty or risk. Be direct rather than hedging every statement."
Every citation AI includes in a memo draft must be verified before the memo is submitted. This is not optional. The verification workflow:
CoCounsel includes citation verification as part of its research workflow, which reduces but does not eliminate this step. Even when using verified-database tools, read the cases before relying on them. AI characterizations of holdings are frequently accurate at the general level but imprecise on the specific point for which they are being cited.
AI-generated legal prose tends toward a particular style: structured, complete, hedged. It covers issues thoroughly but often lacks the precision and directness that partners expect in a memo. Every AI draft requires substantive revision, not just editing.
Specific revision targets: (1) Conclusions—AI defaults to "it is likely that" phrasing for nearly every conclusion regardless of how clear the law is. If the law is clear, say so directly. (2) Issue framing—AI often frames issues more broadly than the memo requires. Narrow to the specific question the client needs answered. (3) Factual application—AI application sections often apply the rule to generic facts rather than the client's specific facts. Ensure every analytical step is grounded in the record.
Maintaining attorney voice is not about stylistic preference. It is about ensuring that the memo reflects attorney judgment, not AI pattern-matching. Partners reviewing a memo are evaluating the associate's analytical judgment. A memo that reads as AI-generated, even if accurate, signals that the associate has not done the analytical work.
AI drafts are good starting points when the legal question is relatively standard—breach of contract elements, qualified immunity analysis, statutory interpretation of a well-litigated provision. They are poor starting points when:
In those situations, use AI for research summarization—feeding it verified case law and asking for synthesis—but draft the analysis yourself.
Supervising attorneys reviewing AI-assisted memos should expect a different error profile than manually drafted memos. Manual drafts are most likely to miss cases or misjudge the strength of authority. AI-assisted drafts are most likely to accurately cite the law but misapply it to the specific facts, or to include cases that are on-point for the general rule but not for the specific issue. Review with that error profile in mind.
Billing for AI-assisted memo work: bar ethics opinions consistently hold that attorneys may bill only for time actually spent. If an AI tool reduced a 10-hour research and drafting task to 4 hours, the client is billed for 4 hours. Some firms have developed flat-fee or value-based arrangements for memo work that address AI efficiency directly—a reasonable approach if the client is informed.
A second-year associate received an assignment: memo on whether a non-compete agreement signed by a California-based employee is enforceable after the employee transfers to a Texas office.
Step 1 – Issue identification (associate-led): The associate read the non-compete agreement and the relevant fact pattern before prompting any AI tool. Identified the core issue: choice-of-law (agreement specified Delaware law; employee is now in Texas; conduct occurred in both states).
Step 2 – Research (AI-assisted): Used CoCounsel to pull controlling cases on non-compete enforceability choice-of-law analysis in Texas and California. Verified each cited case independently in Westlaw.
Step 3 – Rule synthesis (AI-assisted): Fed verified case list into Harvey AI with the rule prompt above. AI synthesized the three-factor Texas choice-of-law analysis and the California public policy exception accurately.
Step 4 – Application draft (AI-assisted, then revised): Used the application prompt with specific facts. AI produced a draft that applied the factors correctly to Delaware/Texas but did not address the California public policy issue—a significant omission the associate caught on review. Application section revised to address it.
Step 5 – Partner review: Partner flagged one citation that was persuasive authority in the wrong direction—associate had included it to show the full picture but had not adequately signaled that it cut against the client's position. Revised.
Total time: approximately 5 hours. Estimated manual time: 9–12 hours for a comparable memo.
CoCounsel – Best for research and citation verification within a verified database. The research memo feature structures output in IRAC-compatible format. See CoCounsel vs Casetext for research comparison.
Harvey AI – Strong for rule synthesis and drafting when given verified research as input. Firm-level deployment with appropriate data handling.
Westlaw Precision – Essential for citation verification regardless of which AI drafting tool is used. Do not skip this step.
Casetext – Research and drafting assistance; strong natural-language research interface for issue-spotting.
Paxton AI – Useful for state-specific research and regulatory memo work in public agency contexts.
Q: If I use AI to draft a memo, do I need to disclose that to the supervising partner?
A: Firm policies vary. Many firms now require disclosure of AI tool use in work product. Even absent a formal policy, transparency with supervising attorneys about how a memo was prepared supports the review and feedback process that is core to associate development. When in doubt, disclose.
Q: How do I handle it when the AI draft is so good that revising it feels unnecessary?
A: Read it critically anyway. AI drafts that feel complete are often the ones that contain the most subtle analytical errors—they are written with enough confidence that problems are easy to miss. Print the draft and mark it up with a pen before editing on screen.
Q: Should I use the same AI tool for research and for drafting?
A: Using separate tools for research (verified-database tools like Westlaw Precision or CoCounsel) and for drafting (Harvey AI or CoCounsel drafting features) is a reasonable separation that reduces the risk of a single tool's errors propagating through the entire memo.
Q: What do I do when my supervising partner disagrees with the analysis in the AI-assisted memo?
A: The analysis in the memo is yours, not the AI's. Engage substantively with the partner's feedback. If the partner identifies a case or argument the memo missed, investigate whether AI failed to surface it or whether it was missed during your review. Use the feedback to improve your prompting and review process.
Q: Is it acceptable to use AI for confidential client matter memos at my firm?
A: Only if your firm has approved the specific tool for use with client data. Using consumer AI tools (ChatGPT, Claude.ai) without enterprise agreements for client matter work likely violates confidentiality obligations. Use only firm-approved tools with appropriate data handling agreements.
AI-assisted memo drafting saves time and can improve research coverage—but only when the associate uses AI as a tool for specific tasks rather than a substitute for analytical judgment. The IRAC prompting framework keeps attorney judgment central at every stage by requiring the associate to evaluate AI output at each step before proceeding to the next.
Citation verification is non-negotiable. The few cases where AI-generated memos have created professional responsibility problems almost always involve citations the attorney did not independently verify. Build verification into your personal workflow as a fixed step, not an optional one.
The long-term skill development question matters too. Associates who rely on AI drafts without engaging in the underlying analysis develop more slowly as lawyers. The goal is to use AI to handle the time-consuming mechanics—research synthesis, structural drafting—while preserving the analytical engagement that builds judgment. That balance is worth being intentional about early in a career.
The billable hour implications are real: efficiency gains from AI reduce the hours billed on memo work. Firms and associates who address this directly—through transparent conversations about value-based billing or adjusted expectations for associate billing targets—will navigate this transition more smoothly than those who ignore it.
This article reflects independent editorial analysis. LawyerAI does not accept payment for editorial coverage. Tool scores are based on methodology described in Our 5-Dimension Methodology. Last reviewed: 2026-08-04.