Justin McKelvey

Justin McKelvey

Fractional CTO · 15 years, 50+ products shipped

AI for Business 7 min read

Claude for Lawyers: What It's Actually Good At, What to Never Paste In, and What It Costs (2026)

Quick Answer

Is Claude good for lawyers? Yes, for the drafting-and-review half of the job, and as of 2026 Anthropic sells it to legal teams directly: Cowork plugins for litigation, corporate, commercial, and IP work, add-ins for Word and Outlook, connectors into iManage and NetDocuments, and no training on your data by default on Team and Enterprise plans. A solo practitioner needs Pro at $20 a month. A firm of two or more needs Team from $20 per seat. And every citation it produces gets verified before it leaves the building, because the model is a drafter, not a database.

Verified September 2026 against Anthropic's legal and pricing pages · Author: Justin McKelvey, fractional CTO & AI consultant, 15 years in software, 50+ products shipped

A year ago "Claude for lawyers" meant a lawyer with a Pro account and a guilty conscience. As of September 2026 it means a product line. Anthropic's legal page lists Cowork plugins configured by practice area, Claude inside Word and Outlook, connectors into the document management systems firms already run, and partner integrations with Harvey, Docusign, Ironclad, and Thomson Reuters' CoCounsel. The marketing is aimed at the Freshfields and Accenture tier. This page is the translation for the solo, the three-lawyer shop, and the in-house team of one: which plan, what the data terms actually say, the four things you never paste in, and the workflows where it earns its subscription in the first week. I've done enough AI installs in law firms to know the tool is rarely the problem. The setup and the policy are.

Read straight off Anthropic's legal solutions page, without the adjectives:

  • Cowork plugins by practice area: commercial legal, corporate legal, IP, litigation. "Configured for your practice area, extensible with your team's playbook and house style." In plain terms: pre-built skills for redlining against a playbook, diligence review, and matter summaries, that you extend with your own templates.
  • Claude for Word and Outlook, plus connectors into the document management systems named in their examples (iManage, NetDocuments) and data rooms (Box, Intralinks). The point is that the model reads the matter file where it lives instead of you pasting it.
  • Partner integrations: the Harvey for Claude connector, Docusign, Ironclad for contract lifecycle, and CoCounsel Legal. If your firm already pays for one of those, Claude is increasingly the thing running underneath it.
  • The data sentence that matters: "no training on your data by default on Team and Enterprise plans," with audit trails. That sentence is the reason a firm should be on a business plan and not five individual consumer accounts.
  • Legal aid: qualifying clinics and public defenders get discounted pricing through the Claude for Nonprofits program.

The examples on the page are telling about who it's built for: diligence review across a data room, outside-counsel billing review across iManage, a GDPR audit tracker. Big-firm workflows. The good news is that the same capabilities run fine at three lawyers; you just don't need the plugins to start.

What Claude is actually good at in a law practice

Four workflows, in the order I'd install them, because each one is lower-risk than the next:

  1. Intake and client communication. Turning a messy inquiry into a structured intake summary, drafting the engagement letter from your template, the status update, the fee explanation a client will actually understand. No privileged strategy involved. This is where a solo gets hours back in week one.
  2. Drafting from your own precedents. Load your templates and house style into a project. Ask for the first draft of the NDA, the demand letter, the motion outline. The output is your precedent with the facts filled in, which is what your associate would produce, minus the associate.
  3. Review and redline against a playbook. "Here are the counterparty's changes to our SOW. Flag every deviation from our standard positions." This is the workflow Anthropic's litigation and commercial plugins are literally built around, and it's the one where lawyers in their quoted blind reviews said the redlines matched or beat their existing tools.
  4. Summarizing long records. Deposition transcripts, discovery productions, a year of correspondence. Themes, timeline, the three things nobody escalated. You still read the source. You read it knowing where to look.

Notice what's not on the list: legal research as a source of authority. Claude will find and frame the issues beautifully. Every case, statute, and citation it names then gets verified in Westlaw, Lexis, or the source document, because the model is a drafter, not a database. The Avianca sanctions in 2023 were about fabricated citations from a general chatbot, and they're the reason every bar now has guidance. Claude is not immune to inventing a plausible cite. The fix is a rule in your policy, not a feature in the tool.

Which Claude plan a law firm needs

Plan Price (September 2026) Who it fits The legal-relevant part
Pro $20/month ($17 annual) Solo practitioner Includes Cowork and projects; a consumer plan, so read the data settings before matter material goes in
Team $20/seat annual ($25 monthly), teams of 2 to 150 Any firm with two or more people No training on your data by default, admin controls, shared projects and skills, audit trails
Team premium seat $100/seat annual ($125 monthly) The one person who also builds or runs heavy agentic work Adds Claude Code and 5x usage; most firms need zero of these
Enterprise Custom Firms with SSO, procurement, and larger-context needs Same no-training default plus enterprise security and admin

The decision rule is simpler than the table: one lawyer, Pro; two or more, Team. The five-seat floor that made Team awkward for small firms is gone as of September 2026 — Anthropic lists it for teams of two to 150. A three-lawyer firm is $60 a month on annual billing. I've seen firms pay more than that for a single month of a case-management add-on they never opened.

If your practice touches protected health information — personal injury, medical malpractice, healthcare clients — the plan question has a second layer, which is whether a business associate agreement is in place. I wrote that up separately in is Claude HIPAA compliant, and the short version is that no model is compliant; a contract is.

The never-paste list

The section vendors mumble through. Four things, on any plan, until your policy says otherwise:

  • Client identifiers together with matter facts on any account whose data terms you have not personally read. A name plus a diagnosis plus a claim is a confidentiality problem the moment it's in a system you don't control.
  • Anything under a protective order or a confidentiality agreement that names the systems permitted to hold it. Your vendor list is now part of the agreement.
  • Privileged strategy you would not want produced from a vendor's logs. Privilege depends on how the material was handled. Handle it like it matters.
  • Protected health information without a signed BAA. Not on Pro, not on Team, not until the paper exists.

The practical move is the one from the law-firm install order: run the first workflows on material that isn't privileged at all — intake, scheduling, billing narratives, public filings — and add matter work once the account is a business plan and the one-page policy exists. The uncomfortable truth I tell every firm owner: the real confidentiality risk in most practices today isn't a deliberate Claude Team install with vetted terms. It's the paralegal pasting matter details into a free consumer chatbot because nobody gave them a sanctioned alternative.

What the bar rules mean in practice

The ABA's 2024 formal opinion on generative AI, and the state bar guidance that followed it, comes down to four duties you already have: competence (understand the tool well enough to supervise it), confidentiality (know where the data goes), candor (verify what you file), and fees (don't bill a client an hour for a draft that took four minutes without a conversation about value). None of that is new law. It's the same supervision standard you apply to a first-year associate, applied to a tool that never sleeps and occasionally lies with confidence. Read your own state's guidance before a client asks whether you have.

A first week that pays for the subscription

Day one: buy the plan, create one shared project, load your engagement letter template, your intake questions, and a style note ("we write short sentences, we never say 'herein'"). Day two: run every new inquiry through the intake summary. Day three: draft the next routine document from your precedent and time it. Day four: take a counterparty's redline and ask for every deviation from your standard positions. Day five: write the one-page policy — approved account, never-paste list, verification rule for anything with a citation, who owns exceptions. That's the whole install for a small firm. The Claude for small business setup guide has the click-by-click version, and the Cowork skills post covers turning a repeated task into a reusable one.

The micro-action for today

Open Anthropic's legal page and read the data-terms sentence yourself, then open your own bar's AI guidance and read that. Twenty minutes, and you'll know more than most of the vendors calling you. If you'd rather have the install done with the policy written and the templates loaded, that's what the Claude for Small Business work is, and the 30-minute strategy call is free with no pitch. Firms that want an AI consultant to look at the wider practice first should start there.

Free Resource Justin McKelvey

Get the Free AI Content Toolkit

The exact system I use to turn one idea into a month of content — atomization framework, voice template, prompt library, weekly system.

Frequently Asked Questions

Is Claude good for lawyers?
Yes, for the drafting-and-review half of legal work, and as of 2026 Anthropic sells it that way: Claude Cowork ships plugins for commercial legal, corporate legal, IP, and litigation; there are Claude add-ins for Word and Outlook; and connectors reach document systems like iManage and NetDocuments. What it's good at: first drafts, redlining against a playbook, summarizing long records, clause comparison, and turning notes into a memo. What it is not: a citation database. Every authority it produces gets verified in Westlaw, Lexis, or the source, the same way you'd check an associate's first draft.
Can lawyers use Claude with client data?
On the business plans, yes, with the terms read first. Anthropic's legal page states there is no training on your data by default on Team and Enterprise plans, with audit trails, and those are the plans a firm should be on. Consumer plans are a different contract: check the privacy settings and the data terms before any matter material goes in. Separately, your state bar's guidance on generative AI (the ABA issued a formal opinion in 2024) governs competence, confidentiality, and disclosure, and privilege depends on how you handle the material, not on which tool you used.
Which Claude plan does a law firm need?
A solo practitioner can run on Claude Pro at $20 a month ($17 on annual billing), which includes Cowork. A firm with two or more people should be on Claude Team, $20 per seat billed annually or $25 monthly, listed for teams of two to 150 as of September 2026, because that's where the no-training-by-default data terms, shared projects, and admin controls live. Premium seats at $100 annual or $125 monthly add Claude Code and more usage, which most firms won't need. Enterprise is custom and is for firms that need SSO, larger context, and procurement-grade paperwork.
What should a lawyer never paste into Claude?
Four things, regardless of plan: client identifiers together with matter facts on any account whose data terms you haven't read; anything under a protective order or a confidentiality agreement that names permitted systems; privileged strategy you would not want produced in discovery from a vendor's logs; and anyone's protected health information unless a business associate agreement is in place. The practical fix is to run the first workflows on material that isn't privileged at all — intake, scheduling, billing narratives, public filings — and add matter work once the account and the policy are set.
Will Claude make up case citations?
Any general model can produce a plausible-looking citation that doesn't exist; the 2023 Avianca sanctions are the case every bar now cites. Claude is not immune. The rule that keeps you safe is procedural, not technological: Claude drafts, a lawyer verifies every authority in a real database before it goes into a filing, and nothing with a citation leaves the firm unverified. Anthropic's own legal page frames Claude as drafting and assembly so lawyers can focus on the calls only they can make, which is the right division of labor.
How much does Claude cost for a small law firm?
As of September 2026: one lawyer on Pro is $20 a month; a three-person firm on Team is $60 a month billed annually ($75 monthly). That's the license. The real cost is the two or three hours it takes to load your templates, house style, and intake questions into a shared project, plus writing the one-page policy on what may and may not go in. Firms that skip the setup hours get a chatbot. Firms that do them get a first-draft associate that never bills.

More on AI for Business

AI for Consultants: How a One-Person Practice Runs on Claude Without Losing the Judgment Clients Pay For (2026)

I'm a consultant. Here's how my practice actually runs on AI as of 2026 — the three jobs it does every day, the one job it's banned from, what it costs, and the pricing mistake most consultants make the week they get faster.

7 min

AI for Startups: The Stack, the Budget, and the Three Mistakes I Get Paid to Fix Later (2026)

Most "AI for startups" advice is a tool list written by someone who has never had to make payroll. Here's the version from someone who fixes the aftermath: the three separate questions hiding inside the phrase, the stack that costs under $100 a month, the API math, and the three mistakes that turn a fast MVP into a $40K rescue.

8 min

AI for CEOs: What to Do Yourself, What to Delegate, and What to Refuse (2026)

The CEO's job with AI is not to become the best prompt writer in the building. It's to use it personally on the six things only you do, delegate the three things that scale, refuse the three that don't, and write a one-page policy so everyone else can move. Here's the weekly routine I run and install.

7 min

AI for Executives: The Five Decisions That Are Yours, and the Twenty That Aren't (2026)

Executives don't need to learn AI. They need to make five decisions about it: what to fund, what to refuse, who owns it, what the board gets told, and what 'we use AI' means in the P&L. Everything else belongs to the people who run the workflows. A fractional CTO's version of the executive AI briefing, without the slide deck.

8 min
Justin McKelvey, Fractional CTO and AI consultant in Austin, TX

Written by

Justin McKelvey

Fractional CTO & AI consultant in Austin, TX. 15 years building software, 50+ products shipped, $53M+ in client revenue generated. I help $1M–$50M founders ship production software and automate operations with AI — without hiring a full-time executive team.

Work with me

If this was useful, here are two ways I can help: