Redline a contract against your playbook (Harvey in M365 / Word Copilot)

The task

If you're in-house counsel, in procurement, or a founder reviewing vendor agreements, AI can read a third-party contract, flag where it deviates from your standard positions (your "playbook"), draft redlines, and produce a negotiation summary — so you start from a marked-up draft instead of a blank one. For firms with Harvey, this now runs inside Word and Outlook via Microsoft 365 Copilot (generally available since June 2026); without Harvey, the same workflow runs on an enterprise LLM tier under counsel's direction. Either way, the lawyer owns every position and every word — this is a first pass, not an autopilot.

This content is for informational purposes only and is not legal advice. Confirm confidentiality, privilege, and jurisdictional rules before using any AI tool with client matters.

Before AI

Reviewing a vendor contract cold means reading every clause, comparing each to your standard position from memory or precedent, and marking up by hand — an hour or more for a first pass on a routine agreement, longer if you're context-switching across deals.

What you'll need

• A privilege-safe tool: Harvey (in M365 Copilot), Spellbook, CoCounsel, Lexis+ AI, or Word Copilot — on an enterprise tier under counsel's direction. Not a consumer chatbot (see Compliance).
• Your clause playbook (standard positions). We scaffold a starter below.
• The third-party contract to review.

The workflow

1. Build your clause playbook (no client data). Use any LLM to scaffold the structure, then fill in your firm's real positions yourself:

Create a contract-review playbook template for vendor SaaS agreements. List the
clauses that matter most — limitation of liability, indemnity, IP ownership,
data protection/security, termination, auto-renewal, governing law — and for
each leave a blank for "our standard position" and "our fallback position."
No legal advice, just the structure.

Fill in your actual standard and fallback positions. This is your leverage — the AI is only as good as this playbook.

2. Pick a privilege-safe surface. Use your enterprise tool. Do not paste a client-confidential or privileged contract into a consumer chatbot (see Compliance).

3. Load the contract + playbook and run the review.

Compare the attached contract against my playbook. For every clause that
deviates from my standard position: quote the contract language, explain the
deviation and its risk, draft a redline that moves it toward my standard
position, and note my fallback. Rank issues High/Medium/Low. Do NOT cite cases
or statutes or invent law — flag anything that needs legal research instead.

4. Review every flag. Verify any standard or authority it references independently (it can fabricate). Edit the redlines to your judgment.

5. Produce the negotiation summary — issues ranked by risk with your asks, for the business owner or the counterparty.

6. Apply the redlines in Word. You, as counsel, own the final markup.

Verify it worked

• Every redline maps to a position in your playbook.
• No invented case law or statutes — verify any citation it offered.
• The High/Medium/Low ranking matches your own read of the risk.
• Nothing confidential went into a non-enterprise tool.

Troubleshooting

• It cited a case or statute. Verify it independently — AI fabricates authority, and fabricated citations have drawn sanctions up to ~$96K (see our AI Incident Tracker). Tell it to flag for research instead of citing.
• It missed a clause. Your playbook didn't cover it — add the position and re-run.
• Redlines feel generic. On an enterprise tier only, feed one or two of your past marked-up contracts as examples of your style.
• Not sure it's privilege-safe. Stop and read Compliance before continuing.

Reality check

Harvey is real and at scale (100,000+ lawyers; M365 GA June 2026) — but "drafts an NDA in seconds" is vendor-stated capability, not a measured benefit. AI redlining saves reading time on the first pass; it does not replace judgment. The danger isn't speed — it's a confident wrong clause or a fabricated citation slipping through to a filing or a signed agreement.

Compliance

• Privilege. Inputting client confidences into a consumer/public AI may waive attorney-client privilege as a third-party disclosure (US v. Heppner, S.D.N.Y. 2026). Work product used at counsel's direction on an enterprise tool is more likely protected (Warner v. Gilbarco, E.D. Mich. 2026), but get informed client consent first — boilerplate engagement-letter language is not adequate (ABA Formal Opinion 512).
• Verify every citation. AI-fabricated citations have produced sanctions and removals; 300+ federal judges now require AI-use disclosure — check the assigned judge's standing order before filing anything.
• This content is for informational purposes only and is not legal advice. Confirm confidentiality, privilege, and jurisdictional rules before using any AI tool with client matters.

Going further

• Turn your reviewed playbook into a reusable AI redline skill (the Day 27 guide).
• Track the moving landscape on our AI Regulation Tracker, and the precedents on the AI Incident Tracker.
• Pair with the privilege-safe AI workspace guide (Day 19).

Your takeaway

Download the clause-playbook starter + the redline prompt. This only works inside a privilege-safe, enterprise tool — the playbook is your leverage, and counsel signs off on every redline.