tellaprompt

AI prompts for lawyers

Last updated

AI belongs in legal work as a first-pass reader and a translator — never as an authority. It scaffolds contract reviews against your checklist, compares clause versions precisely, turns legalese into client-readable memos, and preps negotiations — all on documents you provide. It cites cases it invents; nothing below asks it for law.

Confidentiality first: anonymize parties and identifying facts before pasting anything into consumer tools, or work in a vetted enterprise instance. Every template here works on redacted text.

How to prompt as a lawyer

  • Never ask for case law or statutes from memory — models fabricate authorities with confident citations. Research stays in your databases; AI processes what you paste.
  • Anchor to jurisdiction and role: "under [JURISDICTION] law, acting for [BUYER/SELLER/EMPLOYEE]" changes every answer.
  • Review against YOUR checklist, pasted in — not against the model's idea of what matters.
  • Everything is a draft for professional review; the value is speed to first pass, not delegation of judgment.

The five templates

Contract review scaffold (your checklist, its speed)

A 40-page agreement lands at 17:00. You want a structured first pass against your firm's checklist before you read line by line.

You are my [contract review assistant]. I act for [PARTY/ROLE] under [JURISDICTION] law. Review the agreement below strictly against my checklist.

For each checklist item: status (present / absent / deviates), the exact clause quoted with its number, deviation from my standard position, and risk flag (high/medium/low) with one-line reason. Then: anything unusual in the document that my checklist does NOT cover — as questions, not conclusions. No legal advice; findings only.

My checklist and standard positions:
[PASTE CHECKLIST]

Agreement (anonymized):
[PASTE CONTRACT]
Clause comparison across versions

Redline round four. What actually changed in the indemnity, and does it matter?

Compare these two versions of the [INDEMNITY/LIABILITY/TERMINATION] clause. I act for [PARTY].

Deliver: 1) a difference table — element, version A wording (quoted), version B wording (quoted), substantive effect of the change; 2) which changes shift risk toward my client, ranked by impact; 3) which changes are cosmetic — say so explicitly; 4) the 3 questions I should put to opposing counsel about their intent.

Version A:
[PASTE]
Version B:
[PASTE]

Why this works: Quoted-wording tables prevent the model from summarizing differences into mush, and the cosmetic-vs-substantive split is where hours of redline review actually go.

Output formatFact densityContext first

Plain-language client memo

The client needs to understand the ruling/contract/risk — in their language, at their altitude, without losing precision where it counts.

Turn my notes below into a client memo. Client profile: [SOPHISTICATION, E.G. FIRST-TIME FOUNDER / EXPERIENCED GC]. 

Format: 1) the answer in two sentences up front; 2) what this means for the client practically — max 3 bullets, each starting with a verb; 3) key terms explained in one clause each (no term used before it's explained); 4) options with my recommendation clearly marked; 5) what I need from them, with deadline.
Max [350] words. Tone: confident, warm, zero hedging-stacking ("it may possibly be arguable") — where uncertainty is real, state it once, precisely, with the reason.

My notes and conclusions:
[PASTE — INCLUDING YOUR ACTUAL RECOMMENDATION]

Why this works: Answer-first structure and the one-precise-uncertainty rule are what clients experience as clarity; the memo carries your conclusions — the AI only rebuilds the language around them.

Output formatContext firstAnti-AI style

Negotiation prep: the other side's brief

Before the call, you want to have already heard their best arguments — and drafted your answers.

Prepare me for a negotiation on the attached draft. I act for [PARTY]; counterparty is [WHO, THEIR APPARENT PRIORITIES].

1) Write the brief THEIR counsel plausibly wrote: their 5 strongest positions on the open points, each argued as they would argue it.
2) For each: my strongest response, the evidence/clause it rests on, and my fallback position.
3) Rank the open points by what's actually valuable to my client [CLIENT PRIORITIES] vs. what's tradeable.
4) The 2 questions I should ask early to reveal their constraints.

Open points and current draft positions:
[PASTE]

Why this works: Arguing the counterparty's side first is the oldest prep discipline in the profession — the prompt automates the first draft of it, so your energy goes into the responses.

Role definitionBrutal criticContext first

Grounded document summary (cite the paragraph or say nothing)

300 pages of discovery, due diligence or a judgment — and a summary you can rely on because every claim points home.

Summarize the document below for [PURPOSE, E.G. DD RED-FLAG REVIEW / PARTNER BRIEFING].

Rules: every statement in the summary carries a pinpoint reference (section/page/¶) to the source text; anything you infer rather than read is marked [INFERENCE]; anything the document leaves open is listed under "Open points" — do not resolve them. Structure: 1) three-sentence overview, 2) findings by topic with references, 3) red flags ranked, 4) open points.

Document (anonymized):
[PASTE — IN CHUNKS IF LONG; SAY "NEXT CHUNK" AND KEEP NUMBERING CONSISTENT]

Why this works: Pinpoint-or-silence is the difference between a summary you can cite in a memo and one you have to re-verify line by line; [INFERENCE] tags keep reading and reasoning separate.

Fact densityOutput formatContext first

Frequently asked questions

Can I rely on AI legal research?

Not from general-purpose models — fabricated citations have already produced sanctions in real cases. Research belongs in your legal databases (some now with grounded AI layers); general AI processes documents you feed it. That split is absolute in every template here.

What about client confidentiality?

Treat consumer AI tools as third parties: no client-identifying information without a basis. Practical path: anonymize (parties, amounts, unique facts) before pasting — the templates work on redacted text — or use firm-approved enterprise instances with contractual confidentiality and no-training terms.

Which legal tasks does AI genuinely accelerate?

First-pass contract review against your checklist, version comparison, summarization with pinpoint references, client-facing translation, and negotiation prep. Time studies consistently show the biggest gains in review and drafting-support tasks — with lawyer judgment unchanged as the bottleneck that matters.

Does AI output create malpractice risk?

Unreviewed, yes — like unreviewed work from any junior. The mitigation is structural: templates that quote and cite the source text (verifiable in seconds), findings-not-advice output, and your review before anything leaves the firm. AI speed is safe exactly to the degree verification stays cheap.