How to write a contract playbook your AI can actually use
Most playbooks are written for humans who already know the context. Here is how to turn yours into positions a review engine can apply consistently.
A playbook that lives in a 40-page Word document works because the lawyer reading it already knows what “be reasonable on indemnities” means. A review engine does not. The good news: the work you do to make a playbook machine-readable also makes it better for the junior lawyers on your team.
Start with positions, not prose
For every clause, write three positions:
- Preferred: the language you propose on your own paper.
- Fallback: what you accept without escalation.
- Walk-away: the point at which the deal needs a senior sign-off, or does not happen.
If a clause has no walk-away, it probably does not belong in the playbook.
Make every position testable
“Liability should be capped at a sensible level” cannot be checked. “Aggregate liability capped at no less than 12 months of fees” can. Numbers, durations and named concepts (gross negligence, data breach, IP infringement) give the engine something concrete to match.
Explain the why in one sentence
Each position should carry a single sentence of reasoning. It is what Clausewell shows next to a suggested redline, and it is what convinces a counterparty that your ask is principled rather than arbitrary.
Version it like code
Playbooks drift. Keep a changelog of positions, who approved each change and why. When a contract is reviewed, it should be clear which version of the playbook applied.
A playbook is not finished when it is written. It is finished when two different lawyers reach the same answer with it.