Prompt chaining means breaking one big task into smaller steps. Instead of asking the AI to do everything at once, you guide it through each step, using what you learned from the previous step to inform the next one. Think of it like training a junior associate. You wouldn’t say “review this contract.” You’d say “first find the payment terms, then check if they’re standard, then draft alternatives for any problems.”Documentation Index
Fetch the complete documentation index at: https://docs.pincites.com/llms.txt
Use this file to discover all available pages before exploring further.
Why Chaining Works
When you try to do everything in one prompt, the AI gets overwhelmed and misses details. Breaking tasks into steps gives you:- Better accuracy because the AI focuses on one thing at a time
- Visibility into the AI’s thinking at each stage
- The ability to correct course if something goes wrong
- Reusable workflows you can apply to similar documents
The Basic Approach
After each AI response, you make a decision: continue to the next step, refine what you got, or change direction entirely. You’re in control at every stage.Three Core Techniques
-
Sequential Chain
-
Forked Chain
-
Loop and Refine
Common Legal Chains
Clause Review
- Extract the relevant clause
- Summarize what it says
- Identify the risks
- Suggest better language
Risk Analysis
- List all potential risks
- Rank them by importance
- Explain why each matters
- Suggest how to fix them
Negotiation Prep
- Find the problematic terms
- Group by how hard to negotiate
- Create fallback positions
- Format for management review
Building Your Own Chains
The Standard Template
Most legal review chains follow this pattern:Making Chains Work Better
- Keep steps focused. Each prompt should do one clear thing.
- Reference previous steps explicitly. Say “Using the risks you just identified…” not just “Using the risks…”
- Label your steps. Use “Step 1:” or ”## Analysis Phase” to keep things organized.
- Save successful chains. When a chain works well, save it for next time.
- Start fresh if things go wrong. If the AI gets confused, don’t try to fix it. Begin again with clearer instructions.
- Get reasoning before edits. First understand the problems, then work on solutions.
When to Use Chains
Use Chains For:
- Complex documents with multiple issues
- Tasks where you need to see the thinking process
- Situations where you might need to adjust your approach
- Work requiring multiple perspectives
- Anything you’d normally do in stages yourself
Skip Chains For:
- Simple, routine tasks
- Standard playbook applications
- Quick yes/no questions
- When you know exactly what output you need
Common Problems
- The chain drifts off topic: Start over with clearer first step.
- The AI forgets earlier context: Begin each step with “Based on the above…” or similar reminder.
- Quality varies between steps: Add a quick verification between major stages.
- Too many steps: If you need more than 5-6 steps, you might be overcomplicating. Combine related steps.