Write Better Prompts with the ICA Framework

A tactical guide to writing smarter, faster prompts for AI contract review

AI won’t replace lawyers, but lawyers who can prompt effectively will replace those who can’t.

As AI becomes a staple in contract workflows, legal teams face a new challenge: how to ask the right question. Weak prompts lead to vague, generic, or risky output. Great prompts, by contrast, generate commercially sound, enforceable suggestions that save hours of review time.

Enter the ICA Framework: a three-step model for engineering high-impact prompts that get the most out of AI contract review tools.

Whether you’re building a playbook, reviewing a vendor MSA, or pressure-testing boilerplate, the ICA method helps you turn abstract legal judgment into structured machine instruction.

🎯 I = Identify

What are we looking at?

Start by directing the AI’s attention. This is where most prompts go wrong – too vague, too broad. Your job is to zoom in on the relevant clause, concept, or legal issue.

You can Identify by:

• Clause name (e.g., “Termination”)
• Concept (e.g., “payment triggers tied to deliverables”)
• Function (e.g., “any obligations to provide indemnity”)

✅ Example Prompts:

• “This relates to clause(s) dealing with data ownership or rights to customer data.”
• “Locate any terms requiring automatic renewal or notice periods for termination.”
• “Identify the indemnity clause.”

🧠 C = Check

Does it meet our standard?

Once the clause is identified, check it against your preference or commercial position. This is where nuance matters. You’re asking the AI to assess – not just copy.

Make your position explicit: what is acceptable, preferred, or risky?

✅ Example Prompts:

• “Check if the limitation of liability excludes indirect or consequential damages”
• “Assess whether the data processing clause includes subprocessor flow-down obligations.”
• “Ensure the termination clause allows either party to terminate for convenience with 30 days’ notice”

⚡A = Act

Now what?

Tell the AI what to do based on what it finds. This is the execution layer: redline or suggest. You’re turning analysis into action.

Match the action to the type of intervention you want:

• Insert: the clause is missing
• Redline: it exists but doesn’t meet your standard
• Suggest: it’s risky or ambiguous and needs escalation so flag for review 

✅ Example Prompts:

• “If no clause limits liability for delay, insert one using standard SaaS language.”
• “Redline the indemnity clause to cap it at 12 months’ fees and exclude third-party IP claims.”
• “If the clause allows unilateral assignment, flag it for review.”

ICA in Action: End-to-End Example

Let’s walk through a full prompt using the ICA model.

Prompt:

Identify the indemnity clause. Check whether it includes indemnity for third-party IP claims. If it does, redline the clause to exclude IP indemnity.

What the AI does:

1. Identify → Finds the clause titled “Indemnification”
2. Check → Confirms it includes third-party IP indemnity
3. Act → Rewrites the clause to exclude that scope or add a $500K cap

✍️ ICA Cheat Sheet

Why ICA Works

• Clear mental model for you and the AI
• Minimizes hallucination by breaking complex requests into parts
• Standardizes prompting across your team
• Scales with playbooks and clause libraries
• Reduces back-and-forth and manual corrections

Final Word

The ICA Framework is simple by design but powerful in practice.

In a world of AI-enhanced legal work, your value lies in how well you translate legal expertise into structured instruction. ICA gives you the language and logic to do that with precision.

Use it to prompt better, faster, and smarter. And if your tool doesn’t support ICA-style prompting yet… get a better tool.