A Playbook is only as good as its rules. Once you've generated a Playbook, the real value comes from fine-tuning those rules so the AI redlines the way your team actually negotiates — catching what matters, leaving titles and structure intact, and avoiding unnecessary changes.
This guide covers how to write, test, and maintain strong Playbook rules. If you haven't built a Playbook yet, start with Building AI Playbooks.
Overview
In this article you'll:
Learn how a rule is structured — a Starting Position with ordered fallbacks beneath it
Stop the AI inserting clauses your agreement never had, using if/then rules
Keep redlines surgical instead of verbose
Merge duplicate and overlapping rules that produce conflicting redlines
Protect section titles and document structure
Enforce exact wording with the verbatim keyword
Read the starting AI notes before you work through a review
Test against saved agreements and maintain the Playbook over time
Anatomy of a Rule
A rule is a title plus an ordered list of positions. The top entry is the Starting Position — what you ask for first. Everything below it is a fallback — what you will accept if the first ask is refused.
Drag a fallback above the Starting Position to promote it; add new fallbacks at the bottom.
The higher the fallback number, the broader and more general the position. Order them from your preferred compromise down to your last acceptable one.
When a rule needs review, accepting one position greys out its siblings so you cannot accept two.
💡 The pattern that works: set a Starting Position, then build fallbacks out of what your team has genuinely accepted before. That way a document that does not match your standard can still be resolved without a redline — you simply accept the fallback it does meet.
The clearest rules:
State a single, specific position — one concept per rule
Say what the acceptable language is, and what the fallback is if the counterparty won't agree
Use precise wording — the AI follows your phrasing closely
Write Rules on an If/Then Basis
This is the single most important technique on this page, and the cause of most complaints that a Playbook "over-redlines".
A Playbook does nothing except follow the rules you give it. When you write a rule such as subcontractors must not receive confidential data, the AI reads that as a provision your contracts need — so if the agreement in front of it is silent on subcontractors, it will insert a subcontractor clause rather than leave the topic alone.
This happens most often with subcontractor, waiver and assignment provisions — topics many agreements simply do not address.
The fix is to make the rule conditional:
❌ Subcontractors must not receive confidential data.
✅ If the agreement addresses subcontractors, require that subcontractors do not receive confidential data. If the agreement is silent on subcontractors, do not add a provision.
💡 You do not have to rewrite them one by one. Ask Aline AI to change all of the rules that are applicable to an if/then basis — only apply certain provisions if that topic is already present in the document, then accept the tracked changes. See Building AI Playbooks.
Other conditional positions worth writing this way:
Requiring a waiver of subrogation only when workers' compensation coverage is in play
Limiting additional-insured language to the counterparty and no third parties
Keep Redlines Surgical
A rule that does not say how to edit tends to produce a wholesale rewrite of the clause. Tell your rules to apply redlines surgically rather than in verbose language — change the words that need changing and leave the rest of the clause alone.
Surgical redlines are easier for a counterparty to accept, easier for your own team to review, and much less likely to disturb the document's structure.
One Playbook per Agreement Variant
Do not try to make a single Playbook cover every version of an agreement type. A mutual NDA and a unilateral NDA need different positions, and a Playbook written to handle both will apply the wrong one.
Build a mutual NDA playbook and a unilateral NDA playbook separately. The same applies anywhere your position genuinely changes with the shape of the deal.
📌 A Playbook already knows which party it represents — you set that when you generate it — so you do not need rules restating your side.
Step 1: Generate, Then Edit
Let Aline generate a first set of rules from a template, a rulebook, or a set of previously redlined agreements — then review and edit every rule before you rely on it. The generated rules are a strong starting point, not a finished Playbook.
💡 Best Practice: Attach your standard document and a handful of redlined agreements — five or more, and ideally 10 to 20. Generating from a single clean template usually produces starting positions only; it is the redlined deals that teach it your fallbacks.
Step 2: Merge Duplicate and Overlapping Rules
When rules are generated from multiple documents, you'll often get several that cover the same ground — for example, three variations on limitation of liability. Overlapping rules can produce conflicting redlines.
Combine them into a single, clear rule so the AI has one unambiguous instruction per concept.
Step 3: Preserve Section Titles and Structure
A common problem: the AI rewrites a clause so aggressively that it removes the section heading.
Add a rule that explicitly preserves section titles — for example, an instruction that "Limitation of Liability" and similar headings must never be deleted when the clause is edited. Fold this into the relevant liability or structure rules rather than adding it as a loose one-off.
Step 4: Enforce Exact Language With "Verbatim"
When a clause must match specific wording exactly, include the keyword "verbatim" in the rule. This tells the AI to insert the language word-for-word rather than paraphrasing — essential for indemnification, governing law, or any clause your legal team has settled on.
What a Rule Cannot Check
Rules can check the document against itself and against anything you have written into them. They cannot check the document against facts you have not supplied.
✅ The AI can match a currency to a jurisdiction — it knows that a UK-governed agreement should not be priced in dollars.
❌ The AI cannot tell you whether an address matches the right entity, unless you give it the list of entities and addresses to check against.
If a rule depends on reference data your team holds, put that data in the rule.
Read the Starting AI Notes First
A Playbook run opens with starting AI notes. These are not findings against a rule — they are the things the AI thinks you should know that the Playbook is silent on, including points that may need escalation.
📌 Read the starting notes before you work through the rule-by-rule review. They are the fastest way to spot a gap in the Playbook itself, and they often surface the issue that matters most on an unusual agreement.
Step 6: Test Against Saved Agreements
Before rolling a Playbook out, run it against real, previously redlined contracts you already know the answers to. Compare the AI's redlines to how your team actually marked them up, and adjust any rule that over-reaches or misses.
📌 Use Suggesting Mode so changes come through as tracked suggestions you can accept or reject, rather than direct edits.
Step 7: Maintain and Extend
Playbooks are living documents. When a master agreement gains a new exhibit or a materially new term, add a new rule or category for it rather than stretching an existing rule.
💡 The quickest way to extend a Playbook is to hand Aline AI a redline from a closed deal and ask it to fold in the positions you accepted. It proposes the rule changes as tracked changes for you to accept — see Building AI Playbooks.
⚠️ Accepting or rejecting a redline in a document does not teach the Playbook anything. A Playbook only changes when its rules change.
💡 Best Practices
Write every optional provision as if/then — this is what stops a Playbook adding clauses your agreement never had.
Tell rules to redline surgically, not verbosely.
One concept per rule. Two positions in one rule is how conflicting redlines start.
One Playbook per agreement variant — mutual and unilateral NDAs are not the same Playbook.
Never ship generated rules unedited — treat them as a first draft.
Always pair a position with a fallback so the AI has somewhere to go when the first ask is refused.
Use "verbatim" for any clause where the exact words matter.
Read the starting AI notes before the rule-by-rule review.
Test on contracts you already know before running a Playbook across live deals.
📚 Related Articles
❓ Need Help?
Tuning a Playbook that isn't redlining the way your team does? Click the Support button in Aline — send us a contract you've already marked up and we'll help you close the gap rule by rule.