AI for Legal Research

Finding and analysing precedent

The traditional case law research workflow is iterative: start with a legal question, identify search terms, run queries on Westlaw or LexisNexis, review results, follow citations, read headnotes, refine your search, and repeat until you have a comprehensive picture of the relevant authority.

AI accelerates several steps in this process:

Query formulation. Instead of spending time crafting Boolean search strings, you can describe your legal question in natural language: "Find federal circuit court decisions from the last ten years addressing whether a software licence constitutes a sale for purposes of the first-sale doctrine under copyright law." AI can generate targeted search queries for Westlaw or Lexis, suggest relevant key numbers, and identify secondary sources that may provide a starting point.

Case summarisation. When your search returns 50 potentially relevant cases, reading each one takes hours. AI can summarise the holding, key facts, and reasoning of each case, allowing you to quickly triage: which cases are directly on point, which are merely tangentially related, and which can be set aside.

Distinguishing cases. One of the most valuable analytical tasks in legal research is distinguishing unfavourable precedent. AI can compare the facts and holding of an adverse case against your client's situation and identify factual distinctions, procedural differences, and reasoning that may limit the adverse case's applicability.

Tracing doctrinal development. AI can map how a legal doctrine has evolved across a line of cases — identifying the foundational decision, subsequent expansions or limitations, circuit splits, and the current state of the law. This is particularly valuable for constitutional questions, statutory interpretation issues, and areas of law where the doctrine is actively developing.

The critical rule: every case AI identifies must be verified. You verify the citation exists, confirm the holding is accurately stated, check that the case has not been overruled or distinguished, and run a Shepard's or KeyCite report. No exceptions.

Navigating statutes, regulations, and agency guidance

Statutory interpretation and regulatory research present a different challenge from case law: the relevant text is known, but its meaning, scope, and application may be contested. AI assists differently in this context.

Statutory analysis. When you need to interpret a statute, AI can pull together the statutory text, legislative history, relevant committee reports, and judicial interpretations in a single analysis. You can ask: "Analyse Section 230 of the Communications Decency Act, including the statutory text, the key judicial interpretations from Zeran through NetChoice, and the current circuit split on the scope of publisher immunity." The AI synthesises a body of authority that would take hours to compile manually.

Regulatory research. Federal and state regulatory frameworks are vast, frequently updated, and spread across multiple sources: the Code of Federal Regulations, Federal Register notices, agency guidance documents, no-action letters, enforcement actions, and informal staff interpretations. AI can navigate this landscape faster than manual research. You can ask: "What are the current SEC disclosure requirements for cybersecurity incidents, including the 2023 final rule, staff guidance, and any enforcement actions to date?"

Multi-jurisdictional analysis. For matters that span multiple states or countries, AI can conduct parallel analysis across jurisdictions. A question like "Compare the enforceability standards for non-compete agreements in California, New York, Texas, and Illinois, including recent legislative changes" produces a structured comparison that would take a full day of manual research.

Regulatory change monitoring. AI can review newly published regulations, proposed rules, and agency guidance and identify which ones are relevant to your practice or your client's industry. This is not predictive — it is pattern matching against your defined criteria — but it is far more efficient than manually scanning the Federal Register.

Mata v. Avianca: the case that every lawyer must know

In June 2023, attorney Steven Schwartz of the firm Levidow, Levidow & Oberman submitted a brief in the Southern District of New York in the case of Mata v. Avianca, Inc. The brief cited six cases as precedent. None of them existed.

Schwartz had used ChatGPT to conduct his legal research. The AI generated citations that looked perfectly authentic: correct reporter formatting, plausible party names, realistic procedural histories. When Schwartz asked the AI to confirm the cases were real, it assured him they were. He did not verify the citations on Westlaw or LexisNexis.

Opposing counsel could not find the cited cases. The court ordered Schwartz to produce copies of the decisions. He went back to ChatGPT, which generated fabricated opinion text for each case. When the court investigated, it found that every case was fictional.

Judge P. Kevin Castel imposed sanctions on Schwartz and his colleague, Peter LoDuca. The court's opinion stated: "Technological advances are commonplace and there is nothing inherently improper about using a reliable artificial intelligence tool for assistance. But existing rules impose obligation to ensure the accuracy of their filings."

This case is not an outlier. It is the predictable consequence of treating AI as a research database rather than a text generation tool. AI does not retrieve cases from a database — it generates text that matches the pattern of legal citations. Those patterns can map to real cases or to fictional ones, and the AI does not know the difference.

The verification workflow: non-negotiable steps

Every AI-assisted research product must go through a verification workflow before it is used in any filing, memo, or client communication. This is not optional. It is a professional obligation.

Step 1: Verify every citation exists. For every case citation in AI output, confirm that the case exists by looking it up on Westlaw, LexisNexis, or another verified legal database. If you cannot find the case, it does not exist. Do not ask the AI to confirm — it will tell you the case is real even when it is not.

Step 2: Confirm the holding is accurately stated. Even when AI cites a real case, it may misstate the holding, attribute reasoning to the wrong opinion (majority vs. dissent vs. concurrence), or overstate what the case actually decided. Read the relevant portion of each cited case.

Step 3: Check the current status. Run a Shepard's report (on LexisNexis) or KeyCite report (on Westlaw) for every cited case. Confirm it has not been overruled, reversed, or limited by subsequent authority. AI has no real-time knowledge of subsequent case developments.

Step 4: Verify the factual accuracy. AI may accurately cite a case but describe the facts incorrectly, or state that the case involved a contract dispute when it actually involved a tort claim. Confirm that the factual description matches the actual case.

Step 5: Confirm jurisdictional relevance. AI may cite persuasive authority from another jurisdiction when binding authority exists. Verify that the cited cases come from the correct jurisdiction and have the appropriate precedential weight for your matter.

This five-step workflow adds time to the research process. It is still dramatically faster than conducting the entire research process manually. The time savings come from AI's speed in identifying potentially relevant authority and drafting the initial analysis — not from skipping verification.

Complement, not replacement: how AI fits with your existing research tools

A common misconception is that AI will replace Westlaw and LexisNexis. It will not — at least not in the foreseeable future. Here is why, and how the tools work together.

Westlaw and LexisNexis are databases. They contain verified, authoritative text of judicial opinions, statutes, regulations, and secondary sources. When you find a case on Westlaw, that case exists. The text is the actual opinion. The citation is accurate. The editorial enhancements (headnotes, key numbers, KeyCite) are maintained by human editors.

AI is a reasoning and synthesis engine. It excels at understanding your question, identifying relevant areas of law, analysing patterns across multiple cases, and drafting structured memoranda. But it does not retrieve information from a verified database — it generates text based on patterns in its training data.

The optimal workflow uses both:

1. Frame the question with AI. Describe your legal issue. Have AI suggest relevant areas of law, key cases to investigate, and search terms for Westlaw/Lexis.
2. Search and retrieve on Westlaw/Lexis. Use the AI-generated search terms and case leads to conduct verified searches. Retrieve actual case text.
3. Analyse with AI. Feed the retrieved, verified case text back to AI for analysis: summarise holdings, identify distinguishing facts, map doctrinal development.
4. Draft with AI. Have AI draft the research memorandum using only the verified authority you retrieved from Westlaw/Lexis.
5. Verify the draft. Confirm that the final memo accurately represents the cases you verified, and that no new citations were introduced during drafting.

This workflow is dramatically faster than either tool alone while maintaining the accuracy standard your practice requires.

Drafting research memoranda: from analysis to work product

The research memo is the standard output of legal research: a structured document that presents the legal question, the relevant authority, the analysis, and a conclusion. AI can draft this document from your verified research, saving significant time on the writing phase.

Here is a prompt template for memo drafting:

Draft a legal research memorandum with the following structure:

QUESTION PRESENTED:
[State your legal question here]

SHORT ANSWER:
[Provide a brief answer — 2-3 sentences]

STATEMENT OF FACTS:
[Include relevant facts from the client matter]

DISCUSSION:
Analyse the following authorities in relation to the question presented.
Organise the analysis by issue, not by case. For each authority, state
the holding, identify the relevant reasoning, and explain how it applies
to our facts.

Authorities (verified):
[Paste the case summaries and key excerpts you verified on Westlaw/Lexis]

CONCLUSION:
Provide a clear conclusion with practical recommendations.

Requirements:
- Cite only the authorities provided above. Do not add new citations.
- Use ALWD or Bluebook citation format.
- Flag any areas where the law is unsettled or where additional research
  may be needed.
- Note any adverse authority and explain how it can be distinguished.

The critical instruction is "cite only the authorities provided above." This prevents the AI from introducing hallucinated citations during the drafting phase. You have already verified your authority — the AI's job is to synthesise and write, not to find new cases.