Not just prompts: recipes, sequences, and a methodology your team can run repeatedly. Everything you need to get the most out of AI on documents. Free. Open-source.
"Review this contract and let me know if there are any issues."
You get a generic 2,000-word essay. Some of it's useful. Most of it isn't. You don't know where to start. You spend 30 minutes reading it and still aren't sure what to do.
| Risk | Clause | Issue (Plain English) | Fix |
|---|---|---|---|
| 🔴 | 8.3 | Unlimited liability — no cap at all | Cap at 100% of fees paid |
| 🔴 | 12.1 | Auto-renews for 3 years — not 1 | Change to 12-month renewal |
| 🟡 | 5.7 | IP assignment too broad — covers pre-existing IP | Carve out pre-existing IP |
| 🟢 | 3.2 | Payment terms 45 days — slightly long but fine | Accept or push to 30 |
Never used AI for document review? Start here. This works on any contract, policy, or proposal. Try it right now.
That took 10 seconds to set up. Now scroll down for the recipes that do the hard stuff.
Three steps. That's it. You'll be running your first recipe in under a minute.
Open any major AI assistant: Claude, ChatGPT, Microsoft Copilot, or Google Gemini. All recipes work on all platforms.
Not sure which? Copilot for Microsoft teams, Claude for deep analysis, Gemini for large docs. Full guide below.
Drag and drop any contract, policy, proposal, or agreement into the chat. Most tools accept PDF, Word, or plain text.
Copy any prompt from below and paste it in. Fill in the bracketed placeholders (your name, deal value, concerns) and hit send.
Pick one. Copy it. You're done.
Click a situation to filter, or search by contract type.
What makes a recipe different from a prompt
Anyone can type a question. A recipe bundles everything the AI needs to give you a consistent, actionable result, not a lucky guess.
The workhorse of the kitchen.
Compare any document against a reference and get a colour-coded risk table with plain-English explanations and ready-to-paste fixes.
See it from every angle.
Review the same document from 8 different stakeholder perspectives in a single conversation. Each angle catches what the others miss.
What they always push back on.
Upload multiple negotiated versions and map what counterparties consistently resist. Stop solving the same problem deal by deal.
The safety net before signature.
Catches typos, broken cross-references, missing blanks, and formatting gremlins. Use after all negotiations are complete.
Get up to speed, fast.
New to a deal or document type? This creates a briefing for anyone who knows the domain generally but not this specific agreement.
What flows down to subcontractors?
Extract all flow-down obligations or gap-check your subcontractor flow-downs against a client contract.
Which clauses just restate the law?
Identify which clauses merely restate what the law already requires versus which create new obligations. Cut the noise from your templates.
Map multiple agreements at once.
Compare a portfolio of contracts to understand which model works best on pricing, termination, liability, structure.
Fix a bloated, broken template.
Comprehensive template surgery: targets verbosity, weak risk allocation, placeholders, and structural problems.
Front-load the variables, fix the boilerplate.
Convert any document into a modular format with variable data in boxes up front and standard clauses behind. Makes templates self-service.
Surface what matters before you start.
Identify negotiable vs. fixed terms, create an intake form, and build a client-facing FAQ explaining why certain terms aren't open for discussion.
Hidden risks in the statement of work.
Review a SOW markup for vague deliverables, hidden time commitments, and scope creep. Traffic-light prioritised.
Is this standard? Or are we the outlier?
Benchmark your document against industry standards and SEC filings. Find what's better, worse, or missing before your counterparty does.
What changed? And should it have?
Clause-by-clause comparison of old vs. new template versions with risk-benefit analysis of every change.
No template? No problem.
Reviewing a document for the first time with no internal template to compare against. Uses industry standards and SEC filings as the benchmark.
One clause. Full market analysis.
When a single clause is the sticking point. "Is this reasonable? What's market? What are the considerations?"
Would this actually hold up?
Comprehensive enforceability analysis of post-termination restrictions under UK/Irish law. Traffic-light assessment with duration benchmarks.
Benchmarked against the OneNDA standard.
Fast NDA assessment against the open OneNDA standard. Risk table with surgical fixes.
BAA review against Common Paper standard.
HIPAA Business Associate Agreement assessment distinguishing mandatory requirements from optional pro-party terms.
Have we gone too far? Or not far enough?
Run this after any detailed analysis. Catches over-engineering, redundancies, and missing practical considerations.
How could I have done that better?
After any AI session, ask it to critique your prompting. The fastest way to get better at this. You'll be surprised what you learn about your own assumptions.
Turn a good session into a reusable prompt.
Extract your prompts from any conversation with AI's clarifications appended. Build your own personal playbook over time.
These recipes were born in legal, but the patterns work wherever documents get negotiated, reviewed, or scrutinised. Procurement, HR, finance, operations, and compliance teams are using the same structures. If you review documents for a living, there's a recipe that fits.
Works on vendor proposals, procurement bids, and any document you didn't draft but need to approve.
Works on procurement RFPs, supplier negotiations, and any template your counterparties consistently push back on.
Works on any document multiple stakeholders will scrutinise, policies, project plans, board papers, audit reports.
Works for anyone joining a new project mid-stream: HR policies, finance agreements, partnership terms, grant conditions.
Not a single prompt. A strategy. Four stages that turn AI from a parlour trick into a reliable workflow.
Get the lay of the land. What's standard, what's unusual, what matters? One prompt gives you a map of where to spend your time.
Review from multiple angles. Senior counsel, counterparty, CEO, contract admin. Each perspective catches what the others miss.
Consolidate into actionable output: a checklist, a comparison table, a markup with surgical edits. Not a 40-page memo.
Turn insights into reusable assets: templates, playbooks, intake forms. The goal isn't a one-off answer. It's a system.
Every kitchen needs good knives. These recipes work on any major AI platform, but some tools are better suited to specific jobs. Here's what we've found after hundreds of document reviews.
Why we recommend it: If your team lives in Microsoft 365, Copilot's integration is unmatched as a daily driver to review a contract, draft an email, update a tracker, all without switching tools. Simpler setup, less friction. However, being currently reliant on OpenAI, same issues as ChaGPT.
A word of caution: It seems to have the best style and feel, but can feel brittle at times. Not uncommon for it to stop generating or responding for no apparent reason.
The taste of slop: Success breeds its own problems - ChatGPT can smell a bit like slop since it is so ubiquitous. It also seems the most sycophantic of the models, so important to prompt it adversarially. Eg don't say 'I've drafted the contract' but say 'a new lawyer sent me this contract and I am concerned it is full of holes' so it's a bit more critical.
Watch out While known for being able to find a needle in the haystack, it can be prone to dropping large parts of text or bits without telling you. Requires constant vigilance and asking "have you left anything out?"
AI models improve (and change) constantly. We recommend regularly testing your prompts side-by-side on LMSys Chatbot Arena to make sure you're using the best tool for the job. What was best three months ago may not be best today.
Recipes are ingredients. These are full meals: chained sequences for complete workflows.
Run the traffic-light risk table. Get your executive summary and clause-by-clause breakdown.
Catch over-engineering, redundancies, and anything that doesn't pass the "would a judge care?" test. Done.
Surface what's negotiable vs. fixed. Build your intake form before touching the drafting.
Attack the bloat. Tighten risk allocation, fix inconsistencies, remove placeholders.
Front-load the variables into boxes. Separate deal terms from boilerplate.
Benchmark against SEC filings and industry standards. Find gaps before your counterparty does.
"Are we missing anything salient? What would be a fantastic idea?"
Scope, liability, and top 10 watch-outs. Get oriented fast.
Position the document against industry comparators.
The full gauntlet, all 8 perspectives. This is where the deep insights live.
Final calibration before you ship it.
Map every deviation from your template with risk levels and proposed fixes.
Check the SOW doesn't contradict or weaken your MSA protections.
Deep-dive on specific friction points. "Is this reasonable? What's market?"
If you have multiple deals, map the patterns. Find what they always push back on.
"How could I have prompted you better?" - the single most underused prompt in AI.
Extract and save your prompts. Build your personal playbook over time.
Open-source reference standards we use as benchmarks. You don't need all of these as each recipe tells you which ones it uses.
Open-source building blocks for technology transactions: modular, mix-and-match terms for cloud, SaaS, and data agreements.
Used in: Template development recipesA free, standardised NDA that anyone can adopt. Used as the benchmark in The NDA Quick-Check so you can instantly see how a proposed NDA deviates from a known standard.
Used in: The NDA Quick-CheckOpen-source contract standards for cloud, SaaS, and HIPAA agreements. The HIPAA Check uses their BAA v1.0 as the reference for what "market" looks like.
Used in: The HIPAA CheckGlobal benchmarks for commercial contracting terms. Useful when you need to argue that a position is (or isn't) market standard across industries.
Used in: The Market Check, The Sticking PointsPublic filings contain real, negotiated contracts between real companies. The best free source of "what does market actually look like?" for specific industries.
Used in: The Market Check, The Cold Read, The Eight-Eye ReviewThe concept and community treating law as code. Collaborative, version-controlled legal documents applying software engineering principles to contract drafting.
Used in: Contract draftingHard-won principles from hundreds of AI-assisted document reviews.
Always tell the AI who you are, what the deal is worth, and what your concerns are. "Review this contract" is a lottery ticket. Adding three lines of context turns it into a strategy.
Request tables with clear columns and traffic-light risk levels. Ask for "ELI5" plain language. Your stakeholders will thank you and you'll actually use the output.
Break complex tasks into steps. SOW review → MSA reconciliation → market benchmark. Each step builds on the last. The AI carries context forward within a conversation.
When you need clean output for a markup or email, explicitly say: no markdown, no HTML, no brackets, no footnotes. Otherwise you'll get citation noise you have to strip out.
After any detailed analysis, run The Sanity Check. AI can over-engineer. "Have you gone too far on anything?" is surprisingly effective at catching it.
End with The Replay and save your prompts with The Recipe Saver. You're not just doing the work, you're building a system that compounds!
Common questions about using AI for contract and document review.
Built from 15 years of cross-border legal practice across pharma, tech, and maritime and hundreds of AI-assisted document reviews. These recipes exist because structured prompting shouldn't be a secret.
Curated by Sergey Kinchin, PhD (AIGP, GSEC) · Dublin
Need governed AI workflows built into your environment? That's what I do →
