AI for Lawyers.
AI will not replace your judgment, your strategy, or your license. It will absolutely replace the four hours you spend producing a first draft, summarizing a 90-page brief, or triaging a document dump. This handbook is the practical middle ground: where AI genuinely helps legal work, the three rules that keep it from ending your career, and five workflows you can use this week — no technical background required.
Where AI actually helps — and where it doesn't
The single most useful mental model: treat AI like a brilliant, tireless junior associate with no license, no memory of your client, and occasional confident lying. You'd happily hand that associate a first draft or a summarization job. You would never file their work unread, and you'd never let them talk to a client unsupervised.
| Task | Verdict | Why |
|---|---|---|
| First drafts — letters, memos, standard clauses | Excellent | Blank-page problem solved in seconds; you edit instead of compose |
| Summarizing long documents you provide | Excellent | Grounded in text you gave it — far less room to invent |
| Document review triage & organization | Strong | Sorting, flagging, first-pass relevance — with human sampling |
| Plain-language explanations for clients | Strong | Translation of your correct analysis, which you then check |
| Research starting points — doctrines, arguments, angles | Useful, verify | Good at "what areas of law govern this?"; every authority must be verified |
| Citations of record | Never unverified | Models fabricate convincing citations — see Rule 2 |
| Filings, advice, or anything signed — unreviewed | Never | Your signature, your responsibility — see Rule 3 |
The three non-negotiable rules
Rule 1 — Confidential client information never goes into a consumer AI tool.
Free and consumer AI products may retain what you type and use it to improve future models. Client names, deal terms, medical facts, privileged communications — once pasted, you cannot un-paste them, and you may have breached confidentiality or undermined privilege. The discipline: use enterprise tools with contractual no-training and retention commitments, and even then, redact or abstract identifying facts ("our client, a mid-size logistics company" — not the name). Bar guidance, including ABA Formal Opinion 512 (2024), treats this as a live duty of competence and confidentiality, not a technicality.
Rule 2 — Every citation gets verified in a real database. Every one.
Language models generate plausible text, not verified records. They will produce a perfectly formatted citation — right reporter, right court, right year — to a case that does not exist. This is not hypothetical: in Mata v. Avianca (S.D.N.Y. 2023), attorneys were sanctioned after filing a brief with six fabricated, AI-invented citations, and courts have only grown less patient since. The discipline: a cite is a claim, not a fact, until you have opened it in Westlaw, Lexis, or an official reporter and read that it says what the AI claims it says.
Rule 3 — You sign it, you own it.
Professional responsibility does not delegate to software. Whatever the tool produced, the moment it goes out under your name it is your work product, your representation to the court, your advice to the client. Practically, this means AI output gets the same review you'd give a first-year associate's draft: read every line, check every assertion, rewrite what's off. The time you save is real — it comes from skipping composition, not from skipping review.
Five workflows you can use this week
Each recipe: what to give the AI, what to ask, and what you must verify by hand. The prompts are starting points — adjust the specifics and keep your redaction discipline.
1 · Contract review, first pass
- Redact party names and deal-identifying specifics from the clause or section.
- Paste it with this ask:
2 · Summarize opposing counsel's brief
- Provide the full brief (an enterprise tool with document upload, per Rule 1).
- Ask for structure, not just a summary:
3 · First-draft letters and client emails
- Give facts, audience, and tone — the three things drafts actually turn on:
4 · Research starting point — never endpoint
- Use AI for the map, then do the actual research in a real database:
5 · Plain-language client explanation
- Write (or verify) the correct analysis first. Then translate:
The judgment exercise: spot the danger
Three scenarios from real practice patterns. Pick what you'd do — the point is calibrating when the rules bite.
1. Drafting a motion, the AI offers: "See Whitmore v. Continental Freight Corp., 892 F.3d 1104 (9th Cir. 2018) (holding exactly what you need)." The citation format is perfect. What now?
Perfect formatting is exactly what fabricated citations look like — the model has read a million real cites and reproduces the shape flawlessly. "See generally" on a nonexistent case is still a fabricated citation in a court filing. This is the Mata v. Avianca failure, and it's Rule 2: a cite is a claim until you've opened it.
2. A client emails you their medical records and asks what their injury claim might be worth. A free AI chatbot could summarize the records in seconds. Do you paste them in?
Rule 1. A consumer tool may retain and train on the records; deleting the chat doesn't undo retention on the provider's side. Medical records are among the most sensitive material you hold. Enterprise tools with contractual no-training/retention terms — or the old-fashioned way.
3. The AI drafts a limitation-of-liability clause for a services agreement. It reads clean and professional. The client is waiting. Ship it?
Rule 3. "Reads clean" is the model's specialty and says nothing about whether the cap fits the contract value, whether carve-outs (IP, confidentiality, indemnity) match the deal, or whether mutuality favors your client. Marking it "draft" doesn't move professional responsibility an inch. Review is where your value lives.
Choosing tools: the questions that matter
You don't need to understand the technology to procure it well. You need answers, in writing, to five questions:
| Question | Answer you want |
|---|---|
| Is our data used to train your models? | No, contractually — not "you can opt out somewhere in settings" |
| How long are prompts and documents retained? | Defined, short, and deletable on request |
| Where is data processed and stored? | A stated jurisdiction you can live with |
| Security posture? | SOC 2 Type II or equivalent, encryption in transit and at rest |
| Can we get admin controls and audit logs? | Yes — you'll want to know who used it for what |
General-purpose enterprise AI (Claude, ChatGPT's business tiers, Copilot) covers the recipes in this handbook. Legal-specific platforms layer on citation-checking against real databases, matter management, and firm-wide policy controls — worth evaluating once individual use becomes team use. And if your firm has an AI policy, it precedes everything here; if it doesn't, this handbook's three rules are a reasonable seed for one.
Quick answers
Is using AI for legal work even allowed?
Broadly yes, subject to your duties of competence, confidentiality, and candor. Bar associations have moved from silence to guidance — ABA Formal Opinion 512 (2024) is the landmark — and courts increasingly have standing orders on AI-assisted filings. Know your jurisdiction's and judge's rules.
Will clients pay for AI-assisted hours?
The honest trend: clients increasingly expect the savings. Guidance suggests billing for the value and time actually spent, not the time a task used to take. The upside is capacity — more matters handled well — not padded hours.
What about AI detecting privilege or doing full document review?
AI-assisted review is established in e-discovery (predictive coding predates chatbots by a decade). The new generative tools extend triage and summarization, but sampling and QC by humans remains the professional standard.
Where should a skeptical lawyer start?
Recipe 5 — plain-language translation of your own verified analysis. Zero citation risk, zero confidentiality risk if redacted, and it demonstrates the genuine strength (language transformation) without touching the failure modes.
Related on Vibe Engines
Explore the topic
See this alongside everything else on the same subject — handbooks, system designs, challenges and tools, in one place.
More Handbooks
- The Prompting HandbookA friendly, hands-on field guide for everyday humans — learn the CRISP framework, spot bad prompts, practice with real recipes, play a drag-and-drop game, and test yourself with a quiz. No code required.Read →
- The Agentic AI Interview HandbookTwenty topics every senior AI engineer should be able to reason about live — from eval pipelines to reliability patterns for generative systems.Read →
- The Senior AI Engineer Interview Handbook60 questions across architecture, production incidents, agentic systems, RAG, evals, cost, safety, and leadership — what staff-level AI interviewers actually probe for.Read →
- 50 Angular Interview QuestionsA visual handbook covering components, change detection, RxJS, signals, routing, forms, performance, and testing — what interviewers actually probe for in senior Angular roles.Read →
- 50 Python Interview QuestionsFundamentals to advanced: data structures, OOP, iterators & generators, the GIL, asyncio, memory, testing, and the standard library — a visual walk through everything a Python interview touches.Read →
- 51 LLM Evals Interview QuestionsGolden sets, LLM-as-judge, regression testing, offline vs online evals, RAG evals, agent evals, red-teaming, and observability — demystified for interviews and production.Read →
Explore more from Vibe Engines
Get the next one in your inbox.
New handbooks, system-design walkthroughs, and tools — straight to your inbox. No spam, unsubscribe anytime.