Overview
Before any AI-generated legal content reaches a client, it should pass through systematic validation. This checklist provides a structured approach to quality assurance that catches common AI failures before they become complaints, negligence claims or regulatory problems.
Use this checklist when:
- Deploying a new prompt or AI workflow
- Training junior lawyers on AI validation
- Creating quality gates for AI-assisted work
- Responding to "how do we know this is safe?" questions
The Core Checklist
1. Accuracy Validation
Substantive Correctness
- Legal principles are accurately stated
- Case citations are real and correctly described
- Statutory references are current and accurate
- Jurisdictional requirements are correct
Factual Accuracy
- Dates, names, and figures match source documents
- Mathematical calculations are correct (damages, interest, percentages)
- Cross-references within the document are consistent
- Defined terms are used consistently throughout
2. Hallucination Detection
Citation Verification
- Every case citation has been checked against an authoritative source (Find Case Law, BAILII, the law reports)
- Quoted language matches the text of the judgment
- What each case decided is accurately described
- No fabricated textbooks, practitioner works or articles
Logical Consistency
- Arguments don't contradict themselves
- Conclusions follow from stated premises
- No circular reasoning or logical fallacies
- Analogies are actually analogous
3. Edge Case Testing
Boundary Conditions
- Tested with unusually high/low values
- Tested with missing or incomplete inputs
- Tested with conflicting instructions
- Tested with ambiguous fact patterns
Jurisdictional Variations
- Applies the law of England and Wales unless the matter says otherwise
- Does not import the law of Scotland, Northern Ireland or another country
- Handles governing law and jurisdiction clauses correctly
- Says so when a question falls outside England and Wales
4. Professional Standards
Tone and Formality
- Appropriate professional register maintained
- No colloquialisms or inappropriate casualness
- Hedging language appropriate to certainty level
- Client-facing language is reassuring but accurate
Professional Conduct
Paragraph numbers refer to the SRA Code of Conduct for Solicitors, RELs, RFLs and RSLs.
- The service to the client is competent (paragraph 3.2)
- Work carried out through others has been effectively supervised (paragraph 3.5)
- The client's affairs are kept confidential in prompts and outputs (paragraph 6.3)
- Conflicts of interest considered
- Personal data handled in line with UK GDPR and the Data Protection Act 2018
- Appropriate disclaimers included where necessary
5. Format and Usability
Document Structure
- Proper legal document formatting
- Logical organisation and flow
- Appropriate use of headings and numbering
- Tables, schedules, and appendices correctly formatted
Practical Usability
- Next steps are clear
- Actionable recommendations provided
- Timeline and deadlines identified
- Responsible parties identified where relevant
Implementation Guide
For Individual Lawyers
- Create a physical checklist - Print it and keep it visible
- Mandatory check before sending - Never skip, even for "quick" outputs
- Log failures - Track what types of errors you find most often
- Refine prompts based on patterns - If you're always fixing the same issues, improve the prompt
For Teams
- Two-person rule - One person generates, another person validates
- Rotating validation - Different team members validate each week to prevent blind spots
- Monthly calibration - Review flagged items as a team to align on standards
- Metrics tracking - What percentage of AI outputs pass on first check?
For Firms
- Quality gates in workflow - AI outputs can't progress without checklist sign-off
- Training requirement - All lawyers using AI must demonstrate checklist proficiency
- Audit trail - Keep records of who validated what and when
- Continuous improvement - Update checklist based on discovered failures
Common Pitfalls
"It looks right, so it probably is"
AI outputs look professional and authoritative whether or not they are right. This is exactly why systematic validation matters—you can't rely on vibes.
Solution: Verify citations and legal principles even when they sound plausible.
"We're in a rush, we'll check it later"
The entire point of AI tools is speed. If you're too rushed to validate, you're too rushed to use AI safely.
Solution: Build validation time into project timelines. AI should speed up drafting, not skip quality control.
"Only junior lawyers need checklists"
Cognitive biases affect everyone, whatever their seniority. Confidence that you would spot an error is not a check.
Solution: Everyone uses the checklist, regardless of seniority.
Advanced Variations
Tiered Validation
Not all AI outputs need the same level of scrutiny:
Tier 1 - Full Checklist (Client-facing work, high stakes)
- Complete all checklist items
- Two-person validation
- Documentation of validation process
Tier 2 - Abbreviated Checklist (Internal memos, low stakes)
- Focus on accuracy and hallucination detection
- Single-person validation
- Spot-check citations
Tier 3 - Spot Check (Brainstorming, ideation)
- Quick review for obvious errors
- No formal documentation
The key is explicitly deciding which tier applies before generating the content.
Domain-Specific Extensions
Extend the base checklist for specific practice areas:
Litigation
- Civil Procedure Rules and Practice Directions correctly stated
- Procedural deadlines calculated correctly
- Limitation periods checked
Corporate
- Company types and structures correct
- Board and shareholder approvals and Companies House filings addressed
- Regulatory compliance requirements identified
IP
- Patent and trade mark numbers verified
- Filing deadlines calculated correctly
- Prior art references accurate
Measuring Success
Track these metrics to evaluate whether your testing process is working:
- First-pass quality rate: Percentage of AI outputs that pass validation without corrections
- Error type distribution: What kinds of errors are most common?
- Time to validate: How long does validation take on average?
- Downstream corrections: How often do partners/clients identify issues that validation missed?
If first-pass quality is low, your prompts need improvement. If validation time is high, you may need better tooling. If downstream corrections are common, your checklist needs expansion.
How This Fits with the Rest of the Methodology
This testing checklist is the foundation. Layer it with:
- Tracer Bullets - Test prompts with simple examples before complex ones
- Five Validation Layers - Add automated checks before human review
- Context Architecture - Manage token limits to prevent degraded outputs
Worked Example
This scenario is illustrative, not a record of a real matter.
Scenario: A trainee solicitor uses AI to draft a research note for a client.
Without checklist: The trainee reviews the output, it looks good, and it goes to the client. The client later discovers that the AI cited a case that doesn't exist. Embarrassing correction required.
With checklist: The trainee follows the "Citation Verification" step, discovers the fabricated case, and corrects the note before it is sent. The client receives an accurate note.
The checklist adds a step before the note is sent. Without it, the error reaches the client.
Getting Started Today
- Print this checklist - Make it physical and visible
- Use it on your next AI output - Don't wait for perfect conditions
- Note what you find - Track the first 5 validations to identify patterns
- Refine your prompts - Fix recurring issues at the source
Legal AI tools are powerful, but they're not magic. Systematic validation is what transforms AI from "interesting experiment" to "professional tool."
Questions about implementing this at scale? The consulting page describes what is offered.