The Claude prompt library for lawyers

Review the attached [NDA / MSA / lease] against these firm positions: [list your
standard positions, e.g. "mutual indemnification only," "no unlimited liability,"
"30-day termination for convenience"]. Flag any clause that deviates and explain
the practical risk in one sentence per flag.

Turns a first-pass review into a checklist against your own standards, instead of a generic "what's wrong with this contract" summary.

Compare this contract draft to the prior executed version [attach both]. Produce a
redline summary: every substantive change, who it favors, and whether it reads as
a market-standard term or an outlier for a [deal type] agreement.

Useful for the "what actually changed" pass on a redline that's gone through three rounds of opposing-counsel edits.

Read this indemnification clause: [paste clause]. Explain, in plain English a
non-lawyer client could understand, what risk it shifts, to whom, and under what
trigger. Keep it to four sentences.

Drafts the plain-language explanation you'd otherwise write from scratch for a client call.

I need to understand [legal question] under [jurisdiction] law. Draft a research
plan: the specific sub-questions to answer, the primary sources to check first
(statute, regulation, leading cases), and where I should NOT rely on an
AI-generated summary without independently verifying against the primary source.

Uses the model to structure the research, not to answer the legal question itself -- the plan is the draft, the verification is yours.

Summarize the holding and reasoning of [case citation] in three sentences. Then
list the specific facts of my matter [describe facts] that could distinguish it
from that case. I will verify all of this against the actual opinion before
relying on it.

A fast first pass on whether a case is worth pulling and reading in full -- not a substitute for reading it.

Turn this internal case-status memo into a client update email: [paste memo].
Plain language, no jargon, tone should be [reassuring / direct / neutral]. Flag
anywhere I've stated a prediction or outcome estimate that needs my own review
before this goes out.

Keeps the status-update email from becoming a 20-minute writing task every time a matter moves.

Draft three versions of a fee-increase email to existing clients: one direct, one
softened, one bundled with a value explanation. Keep each under 150 words. Client
context: [relationship length, matter type, any relevant history].

Gives you real options to react to instead of a blank page for a conversation nobody enjoys writing.

Write a script for a difficult conversation with a client about [scope creep /
a setback in their case / a fee dispute]. Include the likely pushback and a
suggested response for each point. Tone: firm but not defensive.

Rehearsal material for the calls that are hardest to prepare for on the fly.

Summarize this deposition transcript into a chronological fact timeline, with
page:line citations for every entry. Flag any statement that appears inconsistent
with the witness's prior testimony or declaration [attach the prior statement if
comparing].

The citation requirement is what makes this usable -- an uncited summary isn't worth the time it takes to fact-check against the transcript.

Draft a first-pass motion for [motion type] based on these facts: [facts] and this
jurisdiction's local rules: [paste or describe rules]. Mark every place I need to
insert a citation to controlling authority rather than inventing one.

Explicitly instructing the model not to invent citations is the single most important line in this prompt -- verify every citation it does include, too.