Chapter 05 · Section II · 16 min read
Briefs and legal memos
The legal memo is the workhorse of every law office — and the document where a clean structural pattern plus a strict authorities-only prompt produces the biggest honest gain from AI.
If pleadings are the documents that go to court, memos are the documents that decide what goes to court. A legal memo is the firm thinking on paper — the structured answer to a discrete question that the partner can read in three minutes, the associate can build the next step on, and the client can be advised from. The form has been around long before AI; what AI changes is the time from “we need a memo on X” to “here is a clean first draft of a memo on X.” That gain is real, and it is also the place where a quiet bad habit can take root if the verification discipline relaxes.
The structured-memo pattern
Every useful legal memo answers, in order, the same six questions. A model is excellent at producing a first draft that fits this shape if you give it the inputs in the same shape.
1. Issue. Stated as one precise question, with the relevant statute or doctrine named. “Whether a non-compete clause of two years and a fifty-kilometre radius, in an employment contract governed by the Labour Act 2074, is enforceable against a former employee who has joined a competitor in Lalitpur.” Not three issues smuggled into one sentence. Not the client’s framing.
2. Short answer. Two or three sentences. The conclusion, with the most important qualifying condition. This is what the partner reads first and sometimes only.
3. Applicable law. The Act, the section, any rules and circulars, and the leading cases — every one verified against its source before it enters the memo. This section is short. It is not a literature review.
4. Analysis. The lawyer’s reasoning, applying the law to the facts. This is the part that earns the firm’s fee. The model can produce a competent first draft if you have given it the law and the facts; the senior rewrites this section to add the judgment the model cannot supply.
5. Conclusion. A clean restatement of the short answer, now with the analytical chain visible.
6. Recommendations. What the firm advises the client to do. Operational, dated, with owners.
Give a model the issue, the verified law, and the facts, and ask for a draft structured to those six headings. The output is something a senior can edit in twenty minutes rather than write from a blank page in two hours. That is the honest productivity gain. It is large; you do not need to overstate it.
Internal does not mean unverified
This is the failure mode that has caused most of the AI-related embarrassments in the legal profession to date, and it is worth understanding the mechanism. The junior, under deadline, drafts a strategy memo with a model. The memo cites Singh v. Government of Nepal (NKP 2078, page 1234) for the proposition that the limitation period for a particular cause of action is extendable on equitable grounds. The case looks plausible — the model has produced a real-sounding name, a citation in the right format, and a holding that fits the argument. The junior does not open NKP volume 2078. The senior reads the memo, accepts the proposition, and writes the client advice letter relying on it. The partner reads the advice letter and signs. Three months later the pleading is filed with the same citation. The other side’s counsel cannot find the case because the case does not exist. The court asks.
Every step in that chain assumed verification had happened earlier. None of them did it. The defence — “it was only an internal memo” — does not survive the moment the citation enters a filed document, and by then it is too late to retract gracefully.
Brief writing: argue from a known set
The court brief is the persuasive cousin of the memo. Where the memo asks “what is the law”, the brief argues “the law, applied to these facts, requires this result.” The structural skills the model brings to memos transfer; the failure modes are sharper because the brief is going to be read by a bench, not a partner.
The safer prompt pattern for brief drafting is narrow and specific. Do not ask the model to find authorities for an argument; ask it to organise a set of authorities you have already verified.
Use only the following authorities — do not introduce others. The facts of the case are as set out below. The argument I want to advance is X. Draft a structured argument with: (1) the proposition; (2) application of each authority to the facts, in the order I have listed them; (3) the conclusion. Where an authority does not bear on a point, say so rather than stretching it. Do not paraphrase the authorities’ holdings beyond what I have provided.
The model is genuinely good at this. It will produce a clean argumentative structure around the authorities you give it. It is bad at, and must not be allowed to, hunt for “supporting” authorities on its own — that is where the fabrications enter. The lawyer remains the one who decides which authorities matter; the model arranges them.
For Nepali practice this pattern is especially useful because most working briefs are arguing from a small, well-defined set — three or four NKP decisions, the relevant sections of an Act, occasionally a directive or circular. The set is finite. The lawyer can verify it in an hour. The model then assembles a structured argument around it in minutes. That is the right division of labour.
The signature carries the work
Every memo and every brief that leaves the office goes out under someone’s name. That name carries the firm’s reputation and the lawyer’s professional standing, and the question of which typing instrument produced the prose — a junior at midnight, a senior partner at a weekend, a language model in twelve seconds — is professionally irrelevant. The duty attaches to the signature, not to the keystroke.
This sounds obvious; it is the principle most often quietly forgotten when AI enters the workflow. The right framing for a firm is: we use AI to draft faster, and we hold every output to exactly the same standard we would hold a draft from a first-year associate. That standard is not “good enough to send”; it is “I am willing to sign my name to this.” If the answer is no, more work is required. If the answer is yes, the draft is ready — and the question of how it was produced is between the lawyer and her time-sheet, not between the lawyer and the client.
Check your understanding
Quick check
—A junior drafts an internal strategy memo using AI and tells you she did not verify the cited authorities because the memo will not leave the office. What is the right response?
What comes next
Memos and briefs are the firm’s internal and persuasive writing. The next section is on the document that meets the court directly — the formal submission, the bench memo, the written argument — and on the lawyer’s duty of candor to the court, which is the principle that determines whether AI use becomes a professional asset or a disciplinary problem.