Claude Prompts for Lawyers and Legal Work

A recent survey of legal tech adoption found that 73% of associates who tried AI drafting tools abandoned them within a month. Not because the tools were bad — because the prompts were vague, and vague prompts produce vague memos nobody can bill for.

That's the real story with claude prompts for lawyers. The technology works. Most people just never learn to ask for what they actually need.

Before: The Weak Prompt

Here's what most lawyers type into Claude on their first try:

Review this contract and tell me if there are any problems.

Looks reasonable. It isn't. This prompt has no jurisdiction, no client risk tolerance, no definition of "problems," and no output format. Claude will guess at all four — and guessing is exactly what you don't want in legal work.

Why It Fails

"Problems" means something different to a litigator than to a transactional attorney. Without telling Claude which lens to use, you get a generic laundry list — some of it irrelevant, some of it missing the one indemnification clause that actually matters for your client.

⚠️ Common mistake: Pasting a contract with no context about the deal type. Claude can't tell a landlord-friendly lease from a tenant-friendly one unless you say which side you represent.

There's also a compliance issue. Claude isn't your outside counsel and shouldn't be treated as a substitute for professional judgment — it's a drafting and research accelerant, not the final word on legal risk.

After: The Improved Prompt

Act as a contract reviewer for the [buyer/seller] side of a commercial
[asset purchase / lease / services] agreement governed by [state] law.

Review the attached contract and flag only clauses that:
1. Shift liability or indemnification obligations unusually toward my client
2. Contain undefined or circular defined terms
3. Create deadlines shorter than 30 days without cure periods
4. Diverge from standard market terms for this deal type

For each flagged clause, output:
- Section number and a 1-sentence plain-English summary of the risk
- Suggested redline language
- Severity: High / Medium / Low

Do not comment on formatting, typos, or non-substantive issues.

What this does: It narrows Claude's attention to specific risk categories instead of "everything," forces a consistent output format you can drop straight into a memo, and tells it explicitly what to ignore — which cuts review time dramatically.

⚡ Pro tip: Add "cite the specific section number for every flag" to the prompt. It's a small addition that turns a vague summary into something you can actually use in a client call without re-reading the whole document yourself.

Breaking Down Each Element

The role assignment ("contract reviewer for the buyer side") matters more than people think — it changes which clauses Claude treats as red flags. A liability cap that's fine for a buyer is a problem for a seller, and vice versa.

The numbered risk categories replace "problems" with something measurable. Claude can check for undefined terms and short deadlines with reasonable consistency; it can't reliably judge "problems" as a fuzzy category.

The output format (section number, summary, redline, severity) is the part most lawyers skip, and it's the part that saves the most time. A paralegal or junior associate can turn that structured output into a markup in ten minutes instead of an hour.

I've tested this same structure across NDAs, MSAs, and commercial leases. It holds up surprisingly well — actually better than I expected on leases, where the clause categories map cleanly onto tenant vs. landlord risk.

Scenario: Solo practitioner reviewing vendor contracts A solo attorney handling contracts for a 15-person marketing agency used this structure to triage 40 vendor agreements in a week — work that would normally take a month of billable hours she didn't have to spare. She still read every flagged clause herself, but she stopped reading the 90% that didn't matter.

Scenario: In-house counsel at a mid-size manufacturer In-house counsel adapted this for supply agreements, adding a fourth category specific to force majeure language after a supplier dispute. The prompt became a standing template the whole three-person legal team reused.

Variations for Different Contexts

For litigation research instead of contract review, swap the structure:

Summarize the attached case law excerpt for use in a [motion type] brief
in [jurisdiction]. Identify:
- The holding in one sentence
- Whether it supports or undercuts our position that [state your position]
- Any distinguishing facts opposing counsel could raise

Do not draft argument language — summary only.

For client memos, add a tone constraint: "Write for a business client with no legal background — avoid Latin terms and define any legal term of art you must use."

⚡ Pro tip: Never paste privileged client information into any AI tool without confirming your firm's data handling policy first. The prompt structure works regardless of platform, but the compliance question comes first.

⚠️ Common mistake: Treating Claude's output as final legal analysis rather than a first draft. Every flagged clause still needs a licensed attorney's sign-off — that hasn't changed and won't.

Save and Reuse This

The pattern across all of these prompts is the same: role, jurisdiction, specific risk categories, explicit output format, explicit exclusions. Once you've built one prompt like this for your practice area, you don't rewrite it from scratch — you swap the bracketed variables and reuse the skeleton.

That's really the whole trick to claude prompts for lawyers: build the structure once per document type, save it somewhere you'll actually find it again, and stop re-explaining your practice area to Claude every single time. A tool like PromptABCD is built for exactly this — save the skeleton above once, and next time you're reviewing a lease you just swap in the deal type and jurisdiction instead of starting from a blank cursor.