In 2023, attorneys Roberto Mata and Steven Schwartz submitted a brief in federal court containing citations to six cases that didn't exist. They had been generated by ChatGPT and submitted without verification. Judge P. Kevin Castel imposed sanctions and described the situation as "an unprecedented circumstance." The lesson wasn't that AI research is worthless — it's that unverified AI research is professionally dangerous. Here's how to use AI for legal research correctly.
The Two Categories of Legal Research AI
Understanding the distinction between these two categories determines every subsequent decision about legal research tools.
Category 1: Generative AI (ChatGPT, Claude, Gemini) These tools generate legal analysis, summarize legal concepts, and produce text that sounds like legal research — but they are not connected to verified legal databases. They can hallucinate case citations, fabricate holdings, misstate statute text, and produce plausible-sounding analysis that is factually wrong. Their value in legal research is as a starting point — framing issues, identifying research directions, understanding unfamiliar areas — not as a source of citable authority.
Category 2: Verified Legal Research AI (CoCounsel, Westlaw AI, Lexis+ AI) These tools are built on top of authoritative legal databases — Westlaw, Lexis Nexis, Thomson Reuters. When they cite a case, it exists. When they summarize a holding, it reflects the actual decision. When they identify relevant statutes, those statutes are current. Their value is in delivering verified legal research faster than traditional research methods, not in generating content that looks like research.
The professional responsibility rule is simple: Category 1 AI can inform your research; it cannot replace verified legal authority.
What Verified Legal Research AI Does Differently
The architecture explains the difference. A generative AI model like ChatGPT was trained on text data that included legal documents — court decisions, law review articles, legal blogs — but that training data has a cutoff date, contains errors, and is weighted by frequency rather than authority. When ChatGPT generates a case citation, it's predicting what a case citation in that context looks like based on patterns — not retrieving an actual case from a database.
CoCounsel, Westlaw AI, and Lexis+ AI work differently: they query verified legal databases and use AI to summarize, synthesize, and present the results. The underlying authority is a real case in a real database — the AI is the retrieval and synthesis layer, not the source.
The practical result: if CoCounsel tells you that Palsgraf v. Long Island Railroad stands for the proposition that liability for negligence requires proximate causation, that's a verified statement about a real case. If ChatGPT tells you that Smith v. Jones, 987 F.3d 456 (9th Cir. 2021) established the same proposition, you need to verify that case exists before relying on it.
CoCounsel (Thomson Reuters) — Verified AI Legal Research
CoCounsel is Thomson Reuters' AI legal research tool, built on the Westlaw database — the largest commercial legal database in the US. It delivers verified case law, statute summaries, and regulatory analysis with citation links to the underlying authority.
What CoCounsel does:
- Answers legal research questions with verified citations from Westlaw
- Summarizes case holdings and extracts key legal points from lengthy decisions
- Identifies relevant statutes and regulatory provisions
- Assists with deposition preparation — identifying key facts and questions from case documents
- Reviews contracts and documents for specific provisions or issues
- Drafts initial research memos from research results
The hallucination-resistant architecture: CoCounsel is built to minimize hallucination by grounding responses in verified database content rather than generated text. This doesn't make it perfect — no AI system is — but it makes it meaningfully more reliable than generative AI for legal research purposes.
Pricing: ~$150/month for solo practitioners (Thomson Reuters has adjusted pricing; verify current plans at thomsonreuters.com). Enterprise plans for larger firms are significantly more expensive.
Best for: Any attorney who does regular legal research across practice areas and needs verified citations rather than research starting points.
Westlaw AI and Lexis+ AI — The Established Platforms with AI Layers
If your firm already subscribes to Westlaw or LexisNexis, the AI features within those platforms provide verified research assistance without adding a new vendor.
Westlaw's AI features:
- Natural language search that understands legal concepts, not just keywords
- Case law recommendations based on documents you're working on
- Quick summary of case holdings and headnotes
- Key Cite integration — immediate verification that a case is still good law
Lexis+ AI features:
- Conversational legal research interface
- Document comparison and analysis
- Statute and regulation tracking
- Integration with Lexis' secondary sources and practice guides
The consideration for small firms: Westlaw and LexisNexis subscriptions are expensive for small firms that don't do high-volume research. If your research needs are moderate, CoCounsel's per-task pricing model may be more cost-effective than a full Westlaw subscription.
ChatGPT for Legal Research: The Safe Workflow
ChatGPT is genuinely useful in legal research — but only for specific, clearly defined purposes that don't involve relying on its output as legal authority.
Legitimate uses of ChatGPT in legal research:
Issue spotting: When you're entering an unfamiliar area of law, ChatGPT can quickly surface the major issues, relevant legal frameworks, and leading cases in that area — giving you a research roadmap before you open Westlaw.
I need to research [legal issue] in [jurisdiction].
I'm not familiar with this area of law.
Give me:
1. The primary legal framework (what body of law governs)
2. The key issues typically at stake
3. The leading cases or statutes I should research
4. Common defenses or counterarguments
5. Jurisdiction-specific considerations if any
Note: I will verify all of this in Westlaw before relying on it.
This is to help me understand the research landscape,
not to provide citable authority.
Plain language explanation: When you've found the law but need to understand it (or explain it to a client), ChatGPT can translate legal language into plain English.
Explain this legal standard in plain language
for a non-lawyer client:
[paste the statutory text or case holding]
Context: [brief description of the client's situation]
Tone: clear and accessible, not technical
Do not add legal advice or case-specific conclusions.
Research memo structuring: Once you've done verified research, ChatGPT can help structure and draft the research memo.
I have completed legal research on [issue] and found
the following verified cases and statutes:
[paste your research findings — no need for exact
citations here as this will be filled in]
Draft a research memo structure that:
1. States the legal question
2. Summarizes the applicable legal standard
3. Analyzes how the standard applies to our facts
4. Reaches a conclusion
I will populate this structure with the actual
verified citations and complete analysis.
This is a first draft requiring attorney review.
What not to do:
- Don't ask ChatGPT for case citations and use them without verification
- Don't rely on ChatGPT's summary of a case without reading the actual decision
- Don't submit any legal document citing cases you haven't personally verified
The Verification Protocol
Every AI research result — from any source — should go through a verification protocol before being relied upon in a client matter.
For verified legal research AI (CoCounsel, Westlaw AI):
- Click through to the underlying case or statute — verify it exists and says what the summary says
- Check KeyCite or Shepard's — confirm the case is still good law, hasn't been overruled or distinguished in ways that affect your argument
- Read the relevant portion of the actual decision — AI summaries can miss nuance that affects your analysis
For generative AI (ChatGPT):
- Before using any AI-generated citation, search for it in Westlaw or Lexis — does it exist?
- If it exists, does it say what ChatGPT said it says?
- Never rely on a ChatGPT citation without independent verification. Never.
The time investment for verification: For verified research AI, verification adds 10–15 minutes per research issue. For generative AI research, the verification burden is higher — plan for 20–30 minutes per citation to verify existence and accuracy. If verification is taking this long, the time savings from AI research are largely eliminated. Stick to verified legal research AI for any citation-dependent work.
Building Your Research Workflow
A practical research workflow for small firm attorneys using AI:
Step 1 — Issue framing (ChatGPT, 10 minutes): Use ChatGPT to understand the legal landscape, identify the key issues, and develop a research plan. Confirm jurisdiction, applicable body of law, and key terms of art.
Step 2 — Verified research (CoCounsel or Westlaw AI, 30–60 minutes): Run your research in verified tools. Get case citations, statute text, and regulatory provisions. Let the AI summarize holdings to identify the most relevant authorities.
Step 3 — Verification (Westlaw/Lexis, 15–30 minutes): Click through to underlying authorities. Verify currency (KeyCite/Shepard's). Read the portions of decisions that directly support your analysis.
Step 4 — Analysis and memo drafting (ChatGPT or attorney, 30–60 minutes): Once your verified research is complete, ChatGPT can assist with drafting the analysis memo. But the legal conclusions — how the law applies to your client's facts — are attorney work, not AI work.
Total research time with this workflow: 1.5–3 hours for a typical issue. Total research time with purely manual research: 4–8 hours for the same issue. Time saved: 50–70%.
