How to Use Claude for Legal Work (2026): A Practical Guide for Lawyers

How Do Lawyers Use Claude?

Lawyers use Claude for five core tasks: reviewing and summarizing documents, drafting from templates and prior work product, writing client and opposing-counsel correspondence, supporting legal research (with independent citation verification), and clearing billing and administrative work. Most attorneys start on Claude Pro at $20 per month for non-confidential work; firms handling client documents should use Claude Team, Enterprise, or the API, the commercial tiers where Anthropic does not train on your data by default.

That is the whole answer in one paragraph. The rest of this guide covers how to do each of those five things well, how to pick a plan, what the confidentiality rules actually require, and, just as important, when not to use Claude at all.

AI adoption in law is no longer a fringe experiment. The Clio 2025 Legal Trends Report found that 79% of legal professionals now use AI in some capacity. The question has shifted from whether to use these tools to how to use them competently, which is precisely what ABA Formal Opinion 512 (July 2024) requires of you anyway.

The Five Core Use Cases

1. Document Review and Summarization

This is where Claude delivers the most obvious hours-saved. Current top models handle up to 1 million tokens of context, roughly 750,000 words, so you can upload an entire merger agreement, a full deposition transcript, or a stack of discovery documents in one conversation and work across all of it at once.

Practical examples attorneys run every week:

• Upload a 60-page commercial lease and ask for every obligation triggered by an assignment, organized by section, with the exact quoted language.
• Feed in a 200-page deposition transcript and get a topic-organized summary with page and line references to check against the original.
• Drop in both parties' contract drafts and ask for a clause-by-clause comparison of the material differences.

The output is a starting map, not a finished work product. You still read the sections that matter. But you read them second, with a guide, instead of first, cold. For a full walkthrough, see AI Contract Review: How to Use Claude for Due Diligence.

2. Drafting From Templates and Prior Work

Claude's strongest drafting mode is not writing from nothing. It is adapting something you already trust. Give it your firm's engagement letter, motion shell, or demand letter template plus the facts of the new matter, and ask for a first draft in your house style. Because you supplied the skeleton, the output stays inside language you have already vetted.

This works for complaints, discovery requests and responses, contract clauses, policy documents, and settlement agreements. The pattern is always the same: reference document plus new facts plus explicit instructions about what to change and what to leave alone. Our guide to reference-document prompting covers the technique in detail, and the prompt library has copy-paste starting points by task.

3. Client and Opposing-Counsel Correspondence

Correspondence is the lowest-risk, highest-frequency use case. Claude is genuinely good at register: it can turn your three-line bullet summary into a clear client update in plain English, tighten an emotional draft to opposing counsel into something firm but professional, or translate a dense ruling into a paragraph a client will actually understand.

Two habits make this safe. First, strip client-identifying details when you are on a consumer plan (more on plans below). Second, treat the draft as a draft. Claude does not know what you promised the client last month or what tone this particular relationship needs. You do.

4. Research Support, With a Hard Caveat

Claude is useful for the thinking layer of research: framing issues, generating theories, outlining the elements of a claim, identifying what you would need to prove, and pressure-testing an argument from the other side's perspective. With web search and Research mode on claude.ai, it can also pull and synthesize current sources.

Here is the hard caveat, and it is non-negotiable: every citation Claude produces is unverified until you have read the case yourself. Language models can generate citations that look perfect and do not exist, or that exist but do not say what the summary claims. The sanctions cases that made headlines all share one fact pattern: a lawyer filed AI-generated citations without checking them. Use Claude to find the shape of the argument. Use Westlaw, Lexis, or the reporter itself to confirm the law. Our legal research guide covers a verification workflow step by step.

5. Billing, Intake, and Administrative Work

The unglamorous use case that quietly pays for the subscription. Attorneys use Claude to turn rough time notes into clean billing narratives, draft intake questionnaires and follow-up emails, summarize meeting notes into action items, organize matter chronologies, and produce internal memos and checklists. None of it is billable-hour work, which is exactly why delegating it matters. Browse the workflows library for ready-made processes covering intake, review, and drafting.

Quick Start: Plan, Setup, First Three Tasks

Which Plan Should You Pick?

Plan	Price	Right for	OK for client documents?
Free	$0	Trying Claude on public material	No
Pro	$20/mo ($17/mo annual)	Solo attorneys, non-confidential work, learning	Not recommended
Max	From $100/mo	Heavy individual users who hit Pro limits	Not recommended
Team Standard	$20/seat/mo annual ($25 monthly, 5-seat minimum)	Small and mid-sized firms	Yes
Team Premium	$100/seat/mo	Firms needing higher usage and Claude Code	Yes
Enterprise	Custom	Larger firms needing SSO, audit logs, admin controls	Yes

The dividing line is training policy, not features. On commercial tiers (Team, Enterprise, and the API), Anthropic does not train on your data by default. On consumer tiers (Free, Pro, Max), whether your conversations are used for model training depends on a privacy setting in your account, with a carve-out for safety review, as of August 2026. That is why the practical rule is simple: client-confidential work belongs on a commercial tier. Full breakdown in Claude AI Pricing for Lawyers.

On models: Claude currently offers Fable 5 (the flagship), Opus 5, Sonnet 5, and Haiku 4.5. For most legal work the default model is the right answer, and you should reach for the strongest model on complex analysis and the faster ones on routine summarization. See which Claude model to use for legal work for a task-by-task matrix.

Set Up Your Workspace

Three claude.ai features do most of the work of turning a chatbot into a practice tool:

• Projects. Create one per matter or per task type. Load the project with your templates, style notes, and standing instructions ("always cite the section number," "never invent facts not in the documents"). Every conversation in that project inherits the context.
• File uploads and Artifacts. Upload contracts, transcripts, and exhibits directly; Artifacts gives you clean, editable documents as output instead of chat text.
• Memory and connectors. Memory carries your preferences across conversations. Connectors, built on the open MCP standard, link Claude to the tools where your documents and matter data already live.

Also worth knowing: Claude for Legal, Anthropic's open-source legal toolkit released in May 2026, bundles practice-area plugins, workflow agents, and connectors that run through Claude's Cowork desktop agent, free on any paid plan. If you want a pre-built legal configuration instead of assembling your own, start there. Our Claude for Legal setup guide gets you running in about ten minutes.

Your First Three Tasks

1. Summarize a document you already know. Pick a contract or opinion you have read closely and ask Claude to summarize it. Because you know the source, you can immediately calibrate what it catches and what it misses.
2. Redraft a template with new facts. Take a standard letter or agreement, give Claude the new matter's facts, and compare its draft to what you would have written.
3. Stress-test an argument. Describe a live issue and ask Claude to argue the other side. This uses the model's reasoning where hallucination risk is lowest and the value of a second perspective is highest.

Each of these takes under fifteen minutes and teaches you the tool's actual strengths and limits on your own work. When you are ready to go deeper on technique, the prompt guide for lawyers and the CRAFT prompting framework are the next stops.

Confidentiality: The Rules in Plain Terms

Model Rule 1.6 does not bend for convenient software. ABA Formal Opinion 512 (July 2024) addresses generative AI directly: lawyers must understand how a tool handles client information before putting client information into it, and must evaluate the risk of disclosure.

Translated into practice:

• Use a commercial tier for client work. Team, Enterprise, or the API. No training on your data by default, with agreements and retention controls available.
• On consumer plans, anonymize. If you use Pro for legal work, strip names, identifying facts, and matter-specific details, and check your privacy settings. Better yet, keep client documents off consumer plans entirely.
• Write it down. A one-page firm AI policy covering approved plans, what data may be uploaded, and who verifies output satisfies most of the supervision concerns in Rules 5.1 and 5.3. Our guide to ABA rules for AI-generated legal work covers the full framework, including the state bar landscape.

Verification Discipline: The One Habit That Prevents Disasters

Every AI failure that has made a lawyer famous for the wrong reasons traces back to skipped verification. The discipline is a single rule applied without exception:

Treat every factual claim, quotation, and citation in Claude's output as unverified until you have checked it against the source.

Concretely:

• Every cited case gets pulled and read before it appears in anything you file or send. No exceptions for citations that "look right."
• Every quotation from an uploaded document gets checked against the document. Ask Claude to include section or page references in its output to make this fast.
• Every summary of a document you have not read is a hypothesis about that document, not knowledge of it.

The mental model that works: Claude is a talented junior colleague with unlimited stamina and no license. You would never sign a filing a first-week associate drafted without reading it. Same standard here. This is also the standard Opinion 512 sets: the duties of competence and candor apply to AI-assisted work exactly as they apply to any delegated work.

When Not to Use Claude

A tool this useful needs a defined edge. Do not use Claude for:

• Final legal judgment. Whether to settle, what to plead, how a judge will react: these are your calls. Claude can inform them; it cannot make them.
• Unverified citations in filings. Covered above, but it bears repeating as a prohibition, not just a caution.
• Client documents on consumer plans. If your firm has not put a commercial tier in place, that is the prerequisite, not an inconvenience to work around.
• Credibility and strategy reads. Claude has never watched your witness testify or sat across from opposing counsel. Assessments that depend on human judgment about people stay with you.
• Anything you would not be comfortable explaining to a disciplinary panel. A crude test, but a reliable one.

Where to Go From Here

Start small and specific: one plan, one Project, the three starter tasks above. Within a week you will know which of the five use cases pays off fastest in your practice. From there, the deep-dive guides on this site map to each next step: Claude vs ChatGPT for lawyers if you are choosing between tools, Claude vs Harvey AI if your firm is weighing an enterprise legal platform, and the practice-area guides for litigation and solo practice for workflows tuned to your docket.