Give Claude a legal workflow, not a vague assignment
A broad request like “Summarize this contract” leaves Claude guessing—and leaves you checking whether it guessed wrong. You’ll get a more useful first pass when you give it a defined legal task, your client’s position, and a clear format for the result.
Use Claude to review a lease against your negotiating positions, extract terms from vendor agreements, or prepare a due diligence issue list. Your judgment stays in charge; Claude helps organize and draft work you can verify and revise.
Here are five practical ways to build that workflow: structured prompts, Projects, custom Skills, Cowork, and Anthropic’s legal plugin. Most are general-purpose capabilities you can adapt to legal work, not tools built exclusively for lawyers.

Give legal prompts five clear instructions
“Summarize this contract” leaves too much open. Claude has to infer your client’s position, the review standard, and which issues matter.
Instead, include five elements:
- Context: Who is your client, and what is the matter?
- Task: What do you want Claude to do?
- Standards: Which playbook or criteria should it apply?
- Tone: Who will read the result?
- Format: How should it organize the findings?
For a commercial lease review, you might write:
You are assisting me with reviewing a commercial lease for a midsized retail tenant in California. Identify provisions that depart from the attached tenant-side negotiating playbook. Prioritize financial exposure, indemnification, termination rights, and operational restrictions.
Write for a senior partner preparing to advise the client. Present a table with clause reference, current provision, preferred position, business impact, proposed revision, and red/yellow/green risk rating.
Quote the relevant contract language. Distinguish playbook deviations from legal conclusions. Do not invent market benchmarks; identify any issue requiring external research.
That last instruction matters. “Market standard” is not a universal fact. If you want Claude to compare a clause with market practice, provide an approved benchmark or verify the comparison independently.

Keep matter materials together in Projects
Projects let you keep matter documents and standing instructions together, reducing repeated setup across conversations.
For an illustrative $45 million acquisition, create a Project called “Acme Corp Acquisition” and add:
- The letter of intent and purchase agreement
- Your due diligence checklist
- Relevant schedules and supporting documents
- Instructions identifying your client as the buyer
- Priorities around purchase price, earnout mechanics, and indemnification
Then ask:
Identify the key buyer-side risks in the purchase agreement. Prioritize provisions affecting purchase price or earnout payments. Cite document names and clause references, and separate confirmed findings from missing information.
A Project gives Claude a consistent starting point. It does not mean every conversation automatically knows everything discussed elsewhere. Put essential facts in Project instructions or reference documents rather than relying on earlier chats.
If your plan supports sharing, colleagues can work from the same approved materials. Check permissions carefully: matter organization should reinforce confidentiality, not weaken it.

Turn repeatable reviews into custom Skills
A custom Skill packages instructions for a recurring task. Think of it as a reusable review playbook, with a Markdown instruction file and, where appropriate, supporting resources.
An NDA review Skill might tell Claude to:
- Flag non-solicitation provisions for escalation
- Prefer California governing law
- Flag confidentiality terms longer than three years
- Review permitted disclosures and return-or-destruction obligations
- Produce clause-specific findings and proposed revisions
These are illustrative firm positions, not universal legal requirements.
In an illustrative NDA demonstration, Claude applied those instructions to a fictional agreement and flagged a non-solicitation clause, a five-year term, and Texas governing law. The useful result was consistency with the supplied playbook—not proof that those provisions were unlawful.
You can also ask Claude to help draft a Skill:
Create a reusable engagement-letter review Skill covering scope clarity, fee structure, conflict disclosures, and retainer terms. Include a checklist, escalation criteria, and a findings table. Leave placeholders for jurisdiction-specific requirements and firm policies. Do not invent professional-conduct rules.
Review and test the Skill before using it on live matters. Keep an approved version and update it when your standards change. For important reviews, explicitly ask Claude to use the named Skill and confirm that it did.

Use Cowork to assemble multi-step deliverables
Cowork supports more involved tasks, such as reviewing several documents, planning the work, and producing files like spreadsheets or memoranda.
For a comparison of three vendor agreements, request a spreadsheet covering:
- Vendor name, term, monthly fee, and annual cost
- Mutual or one-sided indemnification and applicable caps
- Termination rights
- Service-level and uptime commitments
- Post-termination data handling
- IP ownership and governing law
- Overall risk rating, with reasons
Add clear instructions:
Cite the supporting clause for each extracted term. Mark absent provisions as “not found.” Explain annual-cost calculations and distinguish fixed fees from variable charges. Show your proposed approach before proceeding.
In the demonstration, Claude completed a three-contract comparison in approximately three to four minutes. The suggestion that a junior associate might need half a day was an informal estimate, not a controlled productivity study. Neither figure included a measured lawyer-verification stage.
Cowork was described at that time as a research preview with platform, Project-access, and audit-logging limitations. Availability and controls can change. Confirm current capabilities and your firm’s approval before using it with client information.

Apply your playbook when using the legal plugin
Anthropic’s legal plugin provides workflows for tasks such as contract review, NDA triage, and briefing preparation. Its practical value comes from pairing those workflows with your own approved standards.
A generic review might identify “indemnification risk.” A playbook-based review can explain that a clause departs from your preferred mutual indemnification position or exceeds your approved liability cap.
Supply the agreement and playbook explicitly:
Review this contract using the attached negotiation playbook. For each deviation, identify the clause, explain the commercial impact, propose a revision, and state whether escalation is required. Separate legal requirements from negotiating preferences.
Don’t assume that installing the plugin gives it knowledge of your firm’s policies. Confirm which instructions and documents it can access. Commands involving email, calendars, or existing agreements also require the relevant connections and permissions.
Treat NDA triage as an initial recommendation, not an autonomous approval decision.

Build verification into every legal workflow
Claude can produce persuasive errors, including nonexistent or misattributed authorities. Estimates of 5–10% citation errors for Claude and 15–20% for ChatGPT were cited without a study, methodology, or task definition. Treat them as unverified figures, not reliable comparative benchmarks.
Verify every authority through an authoritative source. Don’t assume Claude can access Westlaw, LexisNexis, or another subscription database unless an approved integration actually provides access.
The cited one-million-token context window, described as roughly 2,500 pages, also depends on the model and how you access it. Page capacity varies substantially. A large context window doesn’t remove the need for organized documents and focused questions.
Do this next: Choose one approved workflow and test it on a fictional or appropriately sanitized matter. Compare Claude’s output with your own review, improve the instructions, and document recurring errors.
The goal isn’t just a faster draft; it’s a traceable, consistent work product that supports your professional judgment. Build practical AI skills for your legal work on Tixu.ai, a beginner-friendly AI learning platform.



Leave a Reply