Workshop 5: Prompt Iteration & Output Refinement for Legal Professionals
A comprehensive guide to refining prompts for legal applications and custom GPTs. Learn systematic improvement techniques over multiple iterations to achieve production-ready outputs.

by Haojun See

On The Ground - AI Adoption for Businesses and Individuals
Why Prompt Refinement Matters
Control Tone
Ensure consistent voice in client communications and legal documents.
Improve Formatting
Create structured outputs that follow legal documentation standards.
Reduce Hallucination
Minimise inaccuracies that could lead to legal or ethical issues.
Anticipate Edge Cases
Plan for uncommon scenarios that might arise in legal contexts.
Common Prompt Failure Points
Too Broad
Vague instructions like "Make this better" lead to unpredictable results.
Too Vague
Ambiguous requests such as "Write this nicely" lack specific direction.
Too Long
Overloaded prompts try to accomplish multiple tasks simultaneously.
Too Assumption-Heavy
Presuming context without verification creates misaligned outputs.
Week's Objective
Start with Working Prompt
Begin with your functional prompt from the previous week.
Improve Over Multiple Rounds
Refine through 3-4 iterations using real-world legal test data.
Understand Impact of Tweaks
Observe how minor adjustments improve reliability and usefulness.
Prepare for Custom GPT
Finalise a refined prompt ready for embedding next week.
Anatomy of a High-Quality Prompt
Clear Role
"You are a legal assistant specialising in contract review..."
Specific Task
"Summarise these clauses..." or "Identify potential liabilities..."
Desired Format
"Provide 3 bullet points" or "Format as a formal legal memo"
Style/Tone Constraint
"Use precise legal terminology whilst remaining accessible"
Fallback Logic
"If details are insufficient, ask for specific clarification"
Original vs Refined Prompt
Version 1: Raw
"Summarise this contract for a client"
Output: Lengthy, unstructured text with mixed legal jargon and inconsistent focus.
Version 2: Tone-Edited
"Summarise this contract in accessible language for a non-lawyer client"
Output: Clearer language but still lacks structure and focuses on wrong elements.
Version 3: Format-Edited
"Summarise this contract in 5 bullet points highlighting key obligations, using plain English"
Output: Concise, focused points that a client can easily understand and act upon.
Clarity Boosters
Tone and Style Prompts
Professional Authority
"Write this as a reply from a senior partner to a client"
Collaborative Tone
"Sound approachable yet knowledgeable without legalese"
Balanced Clarity
"Explain complex legal concepts as if to an intelligent non-lawyer"
Assertive Precision
"Use precise terminology whilst maintaining clear directive statements"
Formatting Tips
Lists
"Return potential legal risks in a numbered list by priority"
Tables
"Create a 2-column table: Contractual Obligation / Deadline"
Checkboxes
"List compliance requirements as tasks with checkboxes"
Headings
"Use H2 format headers before each clause explanation"
Guided Iteration Exercise

Select a Prompt
Choose one legal document prompt from your toolkit
Test with Real Input
Use actual legal text or realistic sample data
Identify and Fix One Aspect
Focus on improving tone, format, or clarity
Iterate Multiple Times
Refine through three complete rounds of improvement
Prompt Comparison Sheet
Original Prompt + Output
"Summarise this contract"
Result: Verbose, unfocused summary missing key obligations.
Revised Prompt + Output
"Identify the 5 most important clauses in this contract for our client"
Result: Better focused but mixed technical/plain language.
Final Prompt + Output
"List 5 key obligations with deadlines in this contract, using plain English"
Result: Clear, actionable points with consistent language.
Spotting Hidden Issues in Output
Client-Ready Tone
Would you forward this directly to a client without edits?
Factual Accuracy
Has it invented details or added filler content?
Usability
Can you copy/paste this into a document without reformatting?
Clarity
Would a colleague understand this immediately without context?
Peer Feedback Rounds
Would You Use This?
Rate 1-10 on whether you'd use this in real client work.
Explain what makes it usable or what's holding it back.
What Still Needs Work?
Identify specific improvements that would make it client-ready.
Focus on both content accuracy and presentation elements.
What's Impressive?
Note strengths that could be applied to other prompts.
Highlight particularly efficient or elegant solutions.
Case Study Walkthrough
Initial Output
A lengthy contract analysis with mixed formatting, unnecessary technical terms, and missing crucial deadline information.
Refined Output
Five clear bullet points highlighting key obligations, deadlines, and potential risks in plain English appropriate for non-lawyer clients.
Metrics
  • Time to refine: 10 minutes
  • Word count: Reduced by 70%
  • Clarity rating: Improved from 3/10 to 9/10
Mini-Sprint Briefing
Choose Real Scenario
Select an actual legal document or task you encounter regularly.
Run 3 Iterations
Systematically improve your prompt through three complete cycles.
Document Progress
Record each version and its output with improvement notes.
Finalise for Week 6
Prepare your best version for embedding in a custom GPT.
Prompt Finalisation Template
Task Description
"Contract clause extraction for risk assessment"
Input Sample
Example contract excerpt used for testing
Prompt Versions + Outputs
Document v1, v2, v3 with their respective results
Final Prompt
The optimised version ready for implementation
Reflection
Analysis of why the final version succeeds
Prompt Deployment Considerations
Team Usability
Will colleagues with different expertise levels be able to use this effectively?
Consistency Control
Should tone or formatting be locked to ensure standardised output?
3
3
Clarification Logic
Does the prompt need built-in questions for handling ambiguous inputs?
Custom GPT Readiness
Is this refined prompt ready for next week's GPT implementation?
Common Red Flags to Fix Now
Overly Complex Prompt
Attempting to accomplish too many tasks in a single prompt.
Fix: Break into focused, single-purpose prompts.
Unclear Output Expectations
Failing to specify format, length, or style requirements.
Fix: Add explicit formatting instructions.
Missing Fallback Rules
No contingency for handling incomplete or ambiguous inputs.
Fix: Include "if/then" instructions for edge cases.
Difficult Testing
Cannot easily validate with new data or scenarios.
Fix: Create a standard test suite of sample inputs.
Building Reliable Prompts: Recap
1
Iteration is Essential
No prompt is perfect initially—improvement comes through testing.
2
Format + Tone
These elements create the most significant improvements in output quality.
3
Strong Foundation
Every effective custom GPT begins with a well-refined prompt.
Check-In & Next Steps
Reflection Questions
Which refinement made the biggest difference in your prompt today?
What's your most reliable prompt so far?
What concerns do you have about building a custom GPT?
Contact for Support
Need feedback or testing help for your final prompt version?
WhatsApp: wa.me/6598419481
Coming Next Week
We'll build Custom GPTs using your refined prompts as the foundation.
Come prepared with your finalised prompt ready for implementation.