Legal AI Prompts That Actually Work
Interactive tools, copy-paste templates, and research-backed frameworks to get reliable results from AI—without the hallucinations.
"You shouldn't use these tools with no human oversight, but a person using this has a jetpack now."
— Daniel Martin Katz, Chicago-Kent College of Law
Prompt Templates for PI Workflows
Production-ready prompts for common personal injury tasks. Each includes citation requirements, uncertainty handling, and defense-awareness.
Before & After: Generic vs. Optimized Prompts
The same task can produce wildly different results depending on how you prompt.
What are the requirements for patent infringement?
Issues:
- ✕No jurisdiction specified
- ✕No time frame for precedent
- ✕No output format defined
- ✕No citation requirements
- ✕Open-ended question invites generic response
You are a patent litigation analyst. TASK: Provide a structured analysis of the elements required to establish direct patent infringement under 35 U.S.C. § 271(a). SCOPE: Federal Circuit decisions from 2022–2025 ANALYSIS REQUIREMENTS: 1. Include the standard of proof for each element 2. Distinguish between literal infringement and the doctrine of equivalents 3. Cite three recent leading cases with full citations OUTPUT FORMAT: Structured memo with numbered elements CONSTRAINT: If any element lacks clear precedent, note: "REQUIRES FURTHER RESEARCH"
Improvements:
- ✓Defines specific role (patent litigation analyst)
- ✓Specifies exact statute (35 U.S.C. § 271(a))
- ✓Limits timeframe (2022-2025)
- ✓Requires specific output format
- ✓Mandates citations with clear standard
- ✓Includes uncertainty handling
Result: Produces defensible, citable analysis instead of generic overview
Adapted from US Legal Support prompt engineering research
Security & Compliance Checklist
ABA Formal Opinion 512 (2024) requires lawyers to have a "reasonable understanding" of AI capabilities and limitations. These practices help you stay compliant.
Never Include PII or Privileged Data
Even with "zero-retention" promises, never enter client names, SSNs, account numbers, or attorney-client privileged communications.
Do
- Use pseudonyms: "Client A" or "[PLAINTIFF]"
- Redact identifying information before upload
- Create sanitized versions of documents for AI analysis
Don't
- Include full names, addresses, or contact info
- Upload documents with SSNs or financial account numbers
- Paste privileged attorney communications
Verify Every Citation
AI models hallucinate citations at alarming rates. Every case, statute, and regulation must be independently verified.
Do
- Cross-reference every citation in Westlaw/Lexis
- Check that quotes match actual source text
- Verify cases haven't been overruled or distinguished
Don't
- File any AI-generated citation without verification
- Assume case names and citations are accurate
- Trust AI summaries of holdings without reading originals
Maintain Audit Trails
Document your AI usage for ethics compliance and malpractice defense. ABA Opinion 512 requires "reasonable understanding" of AI tools.
Do
- Log which prompts were used for each task
- Track which outputs required correction
- Document your verification process
Don't
- Use AI without any record of interaction
- Claim AI output as entirely your own work
- Ignore jurisdiction-specific disclosure requirements
Human Review Is Non-Negotiable
Per Florida Bar Advisory 24-1, supervising AI is equivalent to supervising a paralegal. Final judgment must be yours.
Do
- Review all AI output for accuracy and tone
- Assess whether conclusions fit your case strategy
- Make independent judgments on legal questions
Don't
- Submit AI drafts without substantive review
- Let AI make strategic decisions for your case
- Delegate legal judgment to any tool
Recent AI-Related Sanctions
These attorneys learned the hard way—verification isn't optional:
Issue: Six fabricated case citations from ChatGPT
Result: Attorneys sanctioned $5,000 jointly, required to notify affected judges
Issue: 32 fabricated citations including fake case "Brasher v. Stewart"
Result: 6 hours mandatory AI ethics training
Issue: Nearly 30 defective citations, including nonexistent cases
Result: Lead counsel admitted using AI without verification
Key Regulatory Guidance
ABA Formal Opinion 512
July 2024 · First ABA opinion on generative AI
Requires lawyers to have "reasonable understanding" of AI capabilities and limitations. Covers competence (1.1), confidentiality (1.6), and supervision (5.1-5.3).
Florida Bar Advisory 24-1
January 2024 · First comprehensive state guidance
Treat AI like a nonlawyer assistant—same supervision standards apply. Requires informed client consent for third-party AI tools.
California State Bar Guidance
2024 · Ethics guidance on billing
Cannot charge hourly fees for time "saved" by AI. Billing must reflect actual value delivered, not hours that would have been spent manually.
Chief Justice Roberts Year-End Report
December 2023 · Federal judiciary perspective
"Any use of AI requires caution and humility." Signals federal judiciary's awareness of AI risks while acknowledging potential benefits.
What Legal AI Leaders Are Saying
"You shouldn't use these tools with no human oversight, but a person using this has a jetpack now."
Daniel Martin Katz
Professor of Law, Chicago-Kent College of Law
Prompting Frameworks That Reduce Hallucinations
These patterns are derived from legal AI research and designed to improve citation accuracy and reduce hallucinations.
IRAC Prompting
Force the model into formal legal reasoning structure. Structured legal prompts consistently outperform generic approaches on legal reasoning tasks.
ISSUE: Identify the specific legal question
RULE: Quote the statute or precedent exactly as provided
APPLICATION: Apply the rule explicitly to the facts
CONCLUSION: State the outcome based only on the application
Example Prompt Snippet
"Analyze using IRAC: If no relevant rule is found in context, output INSUFFICIENT_CONTEXT."
Common Pitfalls in Legal AI Prompting
Learn from the mistakes that have led to sanctions, malpractice claims, and wasted hours.
Mistake
Using generic AI without citation verification
Fabricated case citations in court filings
Use legal-specific AI with source linking and citation whitelists
Case: Mata v. Avianca (2023): NY attorneys fined for 6 fake cases
Mistake
Treating AI output as final work product
Logical fallacies and misapplied precedent slip through
Implement "Partner Review" step—run adversarial critique before finalizing
Mistake
Prompting without jurisdiction constraints
AI cites NY law for a California case
Always specify jurisdiction in prompts and filter retrieval results
Mistake
Using open-ended "summarize this" prompts
Vague outputs missing key legal elements
Force structured output with IRAC or element-by-element analysis
In-Depth Articles & Guides
How to Use LLMs in Legal Work Without Hallucinations
Learn four strategies to safely use Large Language Models in legal work while avoiding the headline-making mistakes that have led to sanctions.
Read articleWhy Generic AI Tools Put PI Firms at Risk
ChatGPT and generic AI tools lack HIPAA compliance, cite fake cases, and don't understand PI law. Learn what a secure, PI-ready AI platform looks like.
Read articleIntegrating AI Into Your PI Practice: A Step-by-Step Guide
A practical playbook covering pilot programs, prompt libraries, adoption training, and measuring ROI—based on what top firms are actually doing.
Read articleAn AI Assistant that Handles Your Casework Like Clockwork
Explore real-world insights on how firms use Precedent's legal AI assistant to generate timelines, assess injuries, and highlight case strengths and weaknesses.
Read articleHow AI Navigates the High-Stakes Nuance of Litigation
See how litigators use AI to draft interrogatories, prep depositions, audit discovery, and sharpen arguments—without sacrificing human judgment.
Read articleAI That Works for Anyone, No Prompting Expertise Needed
Built-in tools deliver timelines, chronologies, case strengths and weaknesses, and missing docs—without requiring prompt engineering skills.
Read articleLegal Prompt Writing 101
This instructional video showcases why AI prompts matter in your day to day legal work. Learn how to write strong prompts so you can get the most accurate and most reliable answers. Discover real-world examples from our customers and their prompting journey.
Watch the full video
PrecedentPrompt
Engineering
101
"AI should liberate lawyers to do more of the real work of lawyering, not less."
— David G. Yosifon, Santa Clara University School of Law
Want AI That Doesn't Require Prompting?
Precedent's purpose-built tools handle medical chronologies, damage calculations, and case analysis without requiring prompt engineering skills. See how it works.