PlaybookOS (PbOS) isn’t just a static library of rules — it’s designed to make the most of AI so that your contract review becomes faster, smarter, and more consistent. This article explains how AI fits into your playbooks, why the way you set up search criteria matters, and best practices for balancing automation with human judgment.
AI’s Role in PlaybookOS
AI in PbOS serves two main functions:
- Finding Relevant Clauses – Using your search criteria, AI scans the contract to identify relevant sections.
- Assisting with Redlines – When a clause is found (or missing), AI can generate markups based on your playbook’s guidance, preferred language, and scenarios.
Think of AI as your first pass reviewer — it gathers what matters, then surfaces tasks and suggested edits so your team can focus on decisions that actually need human judgment.
Crafting Search Criteria: Where AI Starts
Your criteria fields teach AI what to look for. The most common and recommended method is to use a PROMPT — natural language instructions like:
“Locate governing law clauses in the agreement.”
When writing criteria:
- Keep it broad if you plan to use positions for deeper analysis. For example, “Find force majeure clauses” rather than “Find force majeure clauses that include pandemics.”
- Use MUST NOT criteria to reduce false positives (e.g., exclude unrelated acceptance clauses).
- Layer multiple criteria only if needed — most topics only require one strong prompt.
The goal is to make sure AI captures all potentially relevant clauses so that your outcomes and positions have the right material to analyze.
Choosing the Right Search Criteria
Search criteria are the first step in teaching PbOS what to look for — they are the foundation of every topic in your playbook. The way you configure these criteria determines how effective your AI-powered analysis will be. PbOS supports three types of criteria, each designed for a different use case.
-
Prompt (Recommended)
This is the default and most common choice. Prompts let you write natural language instructions that AI interprets when scanning the contract.
Example:
“Locate governing law clauses in the agreement.”
Use prompts when you want a clear, human-readable search that’s easy to maintain. This approach works especially well if you plan to use positions later for deeper analysis. Keep prompts broad at the topic level so AI captures everything potentially relevant — then refine at the position level.
-
Terms & Connectors (Advanced Precision)
For users who need exact control, Terms & Connectors provide a way to craft highly targeted searches using Boolean-style operators. This is the same logic older playbooks used, and it’s still supported for power users.
Examples:
- exclusive & jurisdiction → finds both words in the same paragraph
- law* of /s% texas → finds governing law clauses that don’t mention Texas
- indemn! & hasredline() → finds indemnification clauses that have been redlined
Use this approach when you want to reduce false positives or capture very specific drafting patterns. It’s also helpful when migrating a legacy playbook to PbOS and you want to preserve exact term logic.
-
Edited (Special Case)
This criterion filters for clauses that were actually changed in an incoming markup — letting you focus only on sections that have been redlined.
Use this only when you are analyzing your own template with third-party edits applied. Avoid using it when reviewing third-party paper (since there’s no baseline template to compare against, nothing would show as “edited”).
Why This Matters
The type of search criteria you choose sets the tone for the entire playbook. Prompts keep your playbook light and readable, while Terms & Connectors give you granular control. Edited criteria are useful for playbooks focused on tracking changes to your own paper. Choosing the right one ensures that PbOS finds what you care about — and nothing more — so the tasks and AI suggestions that follow are meaningful and actionable.
AI in Outcomes: Turning Findings into Actions
Once clauses are found, outcomes define what happens next. AI uses your instructions to generate tasks like Accept, Edit, or Add. For example:
FOUND + EDIT might trigger an AI-generated redline that adjusts language to align with your standard.
NOT FOUND + ADD could insert your preferred fallback clause.
These actions are color-coded in the task list so reviewers know exactly what to do.
AI in Positions: Going Deeper When It Matters
When you choose Proceed to Positions as an outcome, AI performs a second, nested search within the clauses it already found.
Example:
- Topic: Limitation of Liability (broad search)
- Position: Search within those clauses for specific carve-outs (IP, confidentiality, data breach).
- Outcome: Add carve-outs if missing, or flag for review if they exist but are incomplete.
This layered approach keeps your playbook manageable (fewer topics) while still handling nuanced requirements.
Best Practices for Using AI in Playbooks
- Broad at the Top, Specific at the Bottom: Let your topic criteria catch everything relevant, then refine with positions.
- Write Clear Prompts: AI performs best when your language is natural and specific about what to find.
- Explain the Why: Include strong explanations in your outcomes so reviewers understand the reasoning behind each action.
- Use Scenarios for Context: AI can’t know external business context — provide multiple scenarios when human choice is required.
Putting It All Together
When properly set up, AI in PbOS does more than flag language — it guides your team through a structured, context-aware review process. By balancing broad search criteria with targeted positions and clear outcomes, you give AI the tools it needs to make your contract review faster, more accurate, and easier to scale.
Example of the relationship:
Criteria → Outcomes → AI Tasks → Positions (optional) → Final Decision