AI for Legal Research: Tools, Prompts, and Pitfalls
⚖️ 🛠️ Try our free Client Email Drafter: generate results instantly, no signup needed.
Let’s get the scary part out of the way first: in 2023, attorneys in at least three separate cases were sanctioned for citing AI-fabricated cases. ChatGPT confidently generated case names, citations, and even summaries for cases that never existed. The attorneys trusted the output, filed it, and paid the price.
AI can accelerate legal research dramatically. It can also fabricate case citations that don’t exist. The line between these two outcomes is entirely about how you use it: and knowing that line is non-negotiable for any practicing attorney.
The ChatGPT Problem
ChatGPT will confidently cite cases that don’t exist. This isn’t a bug: it’s how language models work. They generate plausible-sounding text, not verified facts.
Multiple attorneys have been sanctioned for filing briefs with AI-fabricated citations. Don’t be next.
Rule: Never cite a case from ChatGPT without verifying it exists in Westlaw, LexisNexis, or another authoritative legal database.
Where ChatGPT IS Useful for Research
Despite the citation problem, ChatGPT helps with research in ways that don’t require case citations:
Understanding a new area of law:
Explain the key elements of a [type of claim] under [jurisdiction] law. What does a plaintiff need to prove? What are the common defenses? Write for a lawyer who’s new to this area, not a law student.
Identifying research angles:
I’m researching [legal issue] in [jurisdiction]. What are the key statutes, leading cases, and secondary sources I should look at? List them so I can verify each one independently.
Summarizing complex statutes:
Summarize [statute number] in plain language. What does it require, who does it apply to, what are the exceptions, and what are the penalties for non-compliance?
Brainstorming arguments:
I’m representing the [plaintiff/defendant] in a [type of case]. The key facts are: [brief summary]. What are the strongest arguments for my client? What are the likely counterarguments?
The pattern: use ChatGPT for frameworks and ideas, then verify everything in proper legal databases.
CoCounsel: AI Built for Legal Research
CoCounsel (Thomson Reuters) is designed to solve the hallucination problem. It:
- Searches actual Westlaw databases
- Cites real cases with proper citations
- Flags its confidence level on each result
- Summarizes holdings accurately
Workflow with CoCounsel:
- Ask a research question in natural language
- CoCounsel searches Westlaw and returns relevant authorities
- Review the cases it identifies
- Ask follow-up questions to narrow results
- Export citations and summaries to your memo
This replaces the hours spent running Boolean searches and reading through irrelevant results.
A Safe AI Research Workflow
- Start with AI for orientation. Use ChatGPT to understand the legal landscape and identify what you’re looking for.
- Research in authoritative databases. Use Westlaw, Lexis, or CoCounsel for actual case law and statutes.
- Use AI for synthesis. After finding your authorities, use AI to help organize and summarize them.
- Draft with AI assistance. Use AI to create a first draft of your research memo, incorporating the verified authorities.
- Verify everything. Check every citation, every holding summary, and every statutory reference before finalizing.
Prompt for Research Memos
Once you’ve done the actual research:
Draft a legal research memo. Issue: [state the question]. Brief answer: [your conclusion]. Authorities: [list the cases and statutes you’ve verified]. Format: IRAC (Issue, Rule, Application, Conclusion). Tone: objective, thorough. Include counterarguments.
This uses AI for writing efficiency while keeping the legal analysis grounded in verified sources.
Jurisdiction-Specific Research
Prompt:
What are the key differences between [legal concept] in [State A] and [State B]? Focus on: elements of the claim, statute of limitations, damages available, and any recent legislative changes. Note: I will verify all citations independently.
Adding “I will verify all citations independently” sometimes (not always) makes AI more cautious about fabricating sources.
The Bottom Line
AI makes legal research faster, not easier. The time savings come from:
- Faster orientation in unfamiliar areas
- Better research question formulation
- Quicker synthesis of findings
- Faster memo drafting
The judgment: which authorities matter, how they apply, what they mean for your client: is still yours. That’s what makes it legal research and not just a search query.
Related reading: 7 Best AI Tools for Lawyers · AI for Contract Review · AI Case Law Analysis: Find Relevant Precedent in M · AI Client Communication Templates for Every Case S
FAQ
Will ChatGPT fabricate legal citations?
Yes. ChatGPT (and similar language models) will confidently generate case names, citations, and even summaries for cases that never existed. This is not a bug: it’s how language models work. Multiple attorneys have been sanctioned for filing briefs with AI-fabricated citations. Never cite a case without verifying it in Westlaw, LexisNexis, or another authoritative legal database.
What’s the safest way to use AI for legal research?
Follow a five-step workflow: (1) use AI to orient yourself and understand the legal landscape, (2) research in authoritative databases like Westlaw or Lexis for actual case law, (3) use AI to synthesize and organize what you found, (4) draft your memo with AI assistance using verified authorities, (5) verify every citation and holding summary before finalizing.
How is CoCounsel different from ChatGPT for legal research?
CoCounsel searches actual Westlaw databases, meaning every citation it returns actually exists. It flags its confidence level and summarizes holdings accurately. ChatGPT generates plausible-sounding text with no connection to verified legal databases and cannot guarantee any citation is real. CoCounsel costs $100/user/month vs. $20/month for ChatGPT.
What is ChatGPT actually good for in legal research?
ChatGPT excels at understanding new areas of law, identifying research angles and key sources to verify, summarizing complex statutes in plain language, and brainstorming arguments and counterarguments. Use it for frameworks and ideas: then verify everything in proper legal databases. It’s a thinking tool, not a citation tool.
Does adding “verify all citations” to my prompt prevent hallucinations?
Adding phrases like “I will verify all citations independently” sometimes makes AI slightly more cautious, but it does not reliably prevent fabricated citations. The only solution is external verification: checking every citation against Westlaw, LexisNexis, or original sources before relying on it. No prompt engineering can guarantee citation accuracy.