Legal and compliance
Review contracts against your playbook with redlines in Word, triage NDAs, extract clauses across a contract set, research with every citation opened, draft policies, track regulatory change, protect privilege and know where the EU AI Act stands.
For In-house counsel, legal operations, contract managers, compliance and privacy officers, paralegals
Review more contracts against your playbook, and never file a citation nobody opened
- A first-pass review against your playbook in about 15 minutes, with redlines as tracked changes
- Incoming NDAs sorted green, yellow or red in under five minutes each
- A clause table across 50 contracts with a quote and section reference for every value
- Zero unverified citations in research memos that leave the team
- A weekly regulatory digest where every item links to its primary source
On this page
- Where AI helps in this job
- Start in 15 minutes
- The tools people in this role use
- Members only: Workflow: playbook review with redlines in Word
- Members only: Workflow: NDA triage in five minutes
- Members only: Workflow: clause extraction across a contract set
- Members only: Workflow: research where every citation is opened
- Members only: Policy drafting
- Members only: Privilege and confidentiality
- Members only: The EU AI Act as it stands today
- Members only: A skill for legal
- Members only: Guardrails and hooks
- Members only: Routines that run without you
- Members only: Prompt ladder for legal
- Members only: Measure it
- Members only: Risks and what not to automate
Legal work is reading, comparing and writing against the clock: third-party paper, NDAs, research questions, policies and a steady stream of new rules. AI is fast at the first pass. It also invents authorities. The AI Hallucination Cases database lists 2,184 decisions, as of 10 October 2026, in which a court or tribunal found that a party relied on hallucinated content, and 864 of them involve lawyers. The working rule on this page: AI reads, extracts and drafts; a lawyer opens every source and signs every output.
Where AI helps in this job#
| Task | What AI does well | What you still own |
|---|---|---|
| Contract review | A first pass against your playbook: issues table, risk rating, proposed redlines | Negotiation strategy, accepting risk, sign-off |
| NDA triage | Sorts standard NDAs from the ones that need a lawyer, in minutes | Approving any deviation |
| Clause extraction | Pulls term, renewal, caps, governing law and change of control into one table across many contracts | Checking samples and interpreting the results |
| Legal research | Frames issues, suggests search terms, summarises authorities you provide | Opening and reading every authority, and the conclusion |
| Policy drafting | First drafts and FAQs from your requirements and house style | Legal accuracy, approvals and consultation |
| Regulatory change | Checks named sources each week and summarises what changed, with links | Assessing the impact on your business |
| Compliance questions | Answers from your own policies with the section quoted | Advice on specific facts |
| Templated responses | Drafts answers to vendor questionnaires and routine requests | The facts and the decision to send |
AI is weak where legal work is most exacting. It cites cases that do not exist and quotes passages that are not in real ones. It misses how clauses interact, such as an indemnity carved out of the liability cap, unless you tell it to read the whole contract first. It blurs jurisdictions and does not know which rules changed last month. In Ayinde v London Borough of Haringey (June 2025), the Divisional Court put it plainly: freely available generative AI tools "are not capable of conducting reliable legal research", and "they may cite sources that do not exist."
Start in 15 minutes#
These work in Claude, ChatGPT, Gemini or Microsoft Copilot (formerly Microsoft 365 Copilot), on your company account.
Prompt
Explain a clause in plain English
Explain this clause in plain English for a [BUSINESS READER, FOR EXAMPLE A SALES DIRECTOR]. List what each party must do, by when, and what happens if they do not. Quote the words that create each obligation. Do not give legal advice and do not refer to any law that is not in the clause. Clause: [PASTE CLAUSE]
Good output quotes the operative words for every obligation. Check each quote against the clause, and treat the explanation as a translation, not advice.
Prompt
Spot issues against three positions
Compare the contract text below with our three positions and list every clause that departs from them. Our positions: 1. Liability is capped at 12 months of fees, and the cap is mutual. 2. No uncapped indemnities, except for IP infringement and data breach. 3. Governing law is [JURISDICTION]. Format each issue like this example: Clause 12.2 (Limitation of liability) | "aggregate liability shall not exceed fees paid in the preceding three months" | Departs from position 1: 3 months, not 12 | Propose 12 months. If a position is not addressed anywhere in the text, say so. Contract text: [PASTE]
Good output quotes the clause for every issue and says when a position is missing entirely. Read the clauses yourself before you rely on the list.
Prompt
Summarise a new regulation for the business
Summarise the attached regulation for [AUDIENCE, FOR EXAMPLE THE PRODUCT TEAM]. Use only the attached text. 1. Who it applies to, quoting the article that says so. 2. The obligations, each with its article number and the date it applies from. 3. What is unclear or left to guidance. Example of the format for point 2: "Article 12: keep records of each processing activity. Applies from 1 March 2027." Do not cite anything that is not in the attached text.
Open each article number it gives you and check the dates. Good output separates what the text says from what is still unclear.
Warning
Client documents, privileged advice, deal information and material under a protective order belong only in tools your firm or company approved for them. Use an enterprise plan with confidentiality terms: Anthropic, for example, does not train on inputs or outputs from its commercial products (Claude Team, Enterprise and the API) by default. On a consumer plan, the Help improve Claude setting decides whether chats are used for training. Check your professional rules before you put client material into any AI tool.
The tools people in this role use#
- Claude on a company plan, with Claude for Word. Claude for Word, Excel and PowerPoint are generally available on all paid Claude plans, and Claude for Outlook is in beta. In Word, Claude edits with tracked changes, responds to comment threads, and lets you save a workflow as a skill your team reuses.
- The Word skill in Claude apps. Upload a .docx and Claude can return a copy with tracked changes and anchored comments, using Anthropic's built-in Word skill.
- Anthropic's legal plugin. Free and open source, for Claude Cowork and Claude Code. Its skills include review-contract, triage-nda, vendor-check, compliance-check, legal-risk-assessment, legal-response, meeting-briefing, brief and signature-request. It reads your positions from a
legal.local.mdplaybook file. Its README warns that the default examples reflect US positions (Delaware, New York and California), so rewrite them for your jurisdiction first. The plugins guide shows how to install and adapt one. - Microsoft Copilot in Word. Microsoft announced support for tracked changes and comments in Copilot in Word, rolling out between June and early July 2026 for Word on Windows and Mac with a Microsoft Copilot licence (Message Center MC1384426).
- AI features in your research database. If your research platform offers them, treat their output like any other draft: you still open every authority they cite.
Workflow: playbook review with redlines in Word
Members only below this line
Get the full Legal and compliance playbook free
What you get on this page:
- 4 step-by-step workflows
- 13 prompts
- 1 skill
- 2 hooks
- 2 routines
Also in the members part
- Workflow: NDA triage in five minutes
- Workflow: clause extraction across a contract set
- Workflow: research where every citation is opened
- Policy drafting
- Privilege and confidentiality
- The EU AI Act as it stands today
- A skill for legal
- Guardrails and hooks
- Routines that run without you
- Prompt ladder for legal
- Measure it
- Risks and what not to automate
Sources
Checked against these on 13 Oct 2026.
- Anthropic knowledge-work-plugins: legal plugin (opens in a new tab)github.com
- Anthropic skills: the Word (docx) skill, with tracked changes and comments (opens in a new tab)github.com
- Claude for Word (opens in a new tab)claude.com
- AI Hallucination Cases database (Damien Charlotin) (opens in a new tab)damiencharlotin.com
- Ayinde v London Borough of Haringey [2025] EWHC 1383 (Admin) (opens in a new tab)judiciary.uk
- European Commission: AI Act application timeline (opens in a new tab)digital-strategy.ec.europa.eu
- EUR-Lex: Regulation (EU) 2024/1689, the AI Act (opens in a new tab)eur-lex.europa.eu
- EUR-Lex: Regulation (EU) 2026/1744, the Digital Omnibus on AI (opens in a new tab)eur-lex.europa.eu
- Microsoft 365 Message Center MC1384426: Copilot in Word support for complex edits (mirror) (opens in a new tab)mc.merill.net
- Anthropic Privacy Center: commercial products are not used for training by default (opens in a new tab)privacy.claude.com
- Anthropic Privacy Center: consumer accounts and the Help improve Claude setting (opens in a new tab)privacy.claude.com
- Claude Help Center: Use skills in Claude (opens in a new tab)support.claude.com
- Claude Help Center: Use connectors (tool permissions) (opens in a new tab)support.claude.com
- Claude Help Center: Schedule recurring tasks in Claude Cowork (opens in a new tab)support.claude.com
- Claude Code docs: Hooks reference (opens in a new tab)code.claude.com