AI Legal Research for Small Law Firms: How to Use It Without Citing Fake Cases

Updated 2026-10-19 · Some links may be affiliate links. Disclosure

Fake citations are no longer a one-off story. A public database kept by researcher Damien Charlotin had logged 2,202 court decisions worldwide dealing with AI-hallucinated material as of October 19, 2026, and 872 of them involved lawyers, not just self-represented litigants (AI Hallucination Cases database). The best-known case set the pattern: in Mata v. Avianca, a federal judge in New York fined two lawyers and their firm $5,000, jointly, and ordered them to send letters to their client and to every judge falsely named as the author of a fake opinion, after they filed made-up cases that ChatGPT produced and then stood by them (S.D.N.Y. sanctions order, June 22, 2023).

Small firms aren’t the only ones caught out, and a firm policy alone doesn’t prevent it. In Johnson v. Dunn (N.D. Ala., July 2025), three lawyers at the firm Butler Snow were publicly reprimanded, removed from the case, and referred to licensing authorities over ChatGPT-invented citations. The firm’s policy already required written approval to use ChatGPT for legal research, and the lawyer who added the citations admitted he hadn’t checked them in Westlaw or PACER (sanctions order).

Legal research AI tools are better than general chatbots, but they still make mistakes. A peer-reviewed Stanford study found the 2024 versions of the AI research tools from LexisNexis and Thomson Reuters each hallucinated between 17% and 33% of the time on the researchers’ test questions. All the products tested have changed since, but the lesson holds: verification is part of the job, whatever tool you use (Magesh et al., Journal of Empirical Legal Studies, 2025).

This guide covers research and citation checking. For drafting and redlining contracts, see our contract drafting guide.

Not legal ethics advice. Check your state bar’s guidance and your court’s rules, which may be stricter.

Ranked picks for AI research with citation checking

Why this order: The most useful feature is a citator in the same product as the AI, so every case it suggests already carries a good-law signal you can click through. All three paid picks have one. Clio Work comes first for small firms because it pairs that citator with research on your own matter documents, works with or without Clio Manage, and can be cancelled at the end of any monthly or annual term (Clio). Westlaw with CoCounsel is second: it has the clearest published pricing for firms of 10 or fewer attorneys and a dedicated brief checker, but it’s sold on 1- to 3-year terms, and Thomson Reuters’ plans page lists annual price increases of 12% on 1-year law-firm plans (5% on 2-year and 4% on 3-year) (Thomson Reuters). Lexis+ with Protégé is close behind. Its earlier version was the most accurate tool in the Stanford study, but there’s no published price to compare. If you already pay for Westlaw or Lexis, try its AI first. Switching research platforms just to get AI is rarely worth the disruption. The two free options belong in every firm’s routine, whatever else you buy.

Where general chatbots fit: ChatGPT, Claude, Gemini and the like can help you outline an argument or brainstorm search terms, with no client identifiers. They should never be the source of an authority you cite. Every case, statute and quote has to come from a research database and be read by a lawyer.

Step 1: Know what’s expected before you research

1. ABA Formal Opinion 512 (July 2024). As a matter of competence, lawyers should review all generative AI output for accuracy. Before submitting anything to a court, they must review it, “including analysis and citations to authority,” and correct “misstatements of law and fact, a failure to include controlling legal authority, and misleading arguments.” The opinion adds that even an unintentional misstatement to a court can be a misrepresentation under Rule 8.4(c) (ABA Formal Opinion 512; ABA summary).
2. Rule 11. In federal court, signing a filing certifies that its legal contentions are warranted by existing law or a nonfrivolous argument for changing it (FRCP 11(b)(2)). Mata was sanctioned under Rule 11, and the rule holds a law firm jointly responsible for its lawyers’ violations absent exceptional circumstances.
3. Your court’s own AI rules. Some courts now require disclosure. In the Northern District of Texas, Local Civil Rule 7.2(f) says a brief prepared using generative AI must say so on the first page under the heading “Use of Generative Artificial Intelligence,” and filing a brief without that disclosure certifies that no part of it was prepared with generative AI (N.D. Tex. Special Order 2-97; Judge Starr’s requirements). Check the local rules and the assigned judge’s standing orders in every court you file in.

Step 2: Pick one tool and test it on a closed matter

1. Start with what you have. If you’re on Westlaw or Lexis, ask your rep what its AI adds and costs. If not, check whether your state bar gives you vLex Fastcase free, then trial Clio Work or Lexis+ with Protégé.
2. Read the data terms before entering client information. Clio says Clio Work processes client files under zero-data-retention agreements and never uses them to train external AI models (Clio). Thomson Reuters says it never uses customer data to train underlying models (Thomson Reuters), and LexisNexis says it never uses customer data to train AI models (LexisNexis). ABA Opinion 512 calls for informed client consent before putting client information into self-learning tools and says boilerplate engagement-letter language isn’t enough; our contract drafting guide walks through that checklist.
3. Run a research question you already know the answer to from a closed matter. Note which key cases the tool found, which it missed, and whether any citation or quote was wrong. That tells you how much checking it needs better than any demo.

Step 3: Research with AI the right way

1. Ask narrow questions with the jurisdiction, procedural posture and time frame, such as “Ninth Circuit standard for a preliminary injunction in a trademark case, decisions since 2020.”
2. Treat the answer as a map, not a memo. Open every case it relies on. The Stanford researchers documented these tools misreading holdings and citing real cases for points they don’t support, errors that are harder to catch than an invented case (Magesh et al.).
3. Ask for the other side. Ask what authority cuts against your position. ABA Opinion 512 lists failing to include controlling authority among the errors you must correct, and Model Rule 3.3 requires disclosing directly adverse controlling authority the other side hasn’t cited (ABA).
4. Keep a short research log of the tool, the questions and the date. It helps if a court asks how a filing was prepared.

Step 4: Check every citation before anything is filed

This is the step that would have prevented the sanctions above. Run it on every brief, motion and memo that cites authority, however the research was done.

1. It exists. Pull up every cited case in a research database. CourtListener’s free Citation Lookup Tool is a quick first pass, and its API can scan a whole draft and flag citations it can’t find (CourtListener). If a case can’t be found, treat it as fake until you’ve seen the opinion.
2. The quote is real. Match every quotation word for word against the opinion. In Mata, the filings contained fake quotes as well as fake cases (S.D.N.Y.).
3. It says what you say it says. Read the passage you rely on and confirm it supports your sentence, and that it’s the holding, not dicta or a dissent.
4. It’s still good law. Run KeyCite, Shepard’s or Cert on every authority and read any negative treatment.
5. Pin cites and parentheticals are right.
6. Use the built-in checkers, then check anyway. Westlaw Quick Check verifies the citations and quotations in an uploaded brief, Word files up to 20 MB or PDFs up to 30 MB, about 120 pages (Thomson Reuters). Lexis’s Shepard’s Verify checks citations in drafts and flags unsupported statements (LexisNexis). These speed up steps 1 to 5; they don’t replace your reading.
7. The signing lawyer owns it. A paralegal can do the first pass under supervision, but the lawyer who signs confirms the check was done. Add the AI disclosure wherever a court requires one.

Step 5: Put it in your firm’s AI policy

Johnson v. Dunn shows that a policy restricting tools isn’t enough on its own. Make the check step the rule:

• Approved research tools, listed by name; general chatbots are never the source of a cited authority.
• The Step 4 checklist is completed and initialed before any filing goes out.
• Who supervises staff and paralegals using AI (ABA Opinion 512 applies the supervision rules, Model Rules 5.1 and 5.3, to AI use).
• What to do if an error is found after filing: tell the supervising lawyer at once and correct it with the court.

Our accounting firm AI policy guide is a one-page template that adapts well.

Step 6: Bill fairly

Under ABA Opinion 512, lawyers billing hourly bill for the time actually spent, and generally can’t charge clients for time spent learning a general AI tool. You can treat the tool’s cost as overhead, or pass on a share of it only if you explain that to the client in advance (ABA summary).

Step 7: Check the other side’s briefs too

The same tools work on opposing filings. Quick Check can analyze an opponent’s brief for negative treatment of its authorities and relevant authority it left out (Thomson Reuters), and Clio says Clio Work flags weak citations in opposing arguments (Clio). A citation in the other side’s brief that doesn’t exist is worth knowing about before the hearing.

Step 8: Measure after 30 days

Track research hours per matter, the number of citation problems caught at the Step 4 check, and the subscription cost against hours saved. If the tool saves little time once you add the checking it needs, cancel it and fall back on your bar’s free database.

Quick checklist

• Local rules and judge’s standing orders checked for AI disclosure requirements
• One research tool with a built-in citator chosen and tested on a closed matter
• Data terms read; client consent handled where client information goes into the tool
• No cited authority taken from a general chatbot
• Every citation checked: exists, quote matches, supports the point, still good law, pin cite right
• Signing lawyer confirms the check; AI use disclosed where required
• Firm AI policy makes the check step mandatory
• Billing reflects time actually spent
• Results measured after 30 days

Tools in this guide

• Clio Work: Firms already on Clio Manage.
• CoCounsel Legal (Thomson Reuters): Small firms already on Westlaw, or litigators who want a citator and a brief checker in the same product as the AI.
• CourtListener Citation Lookup: A free first pass to confirm every case citation in a draft actually exists.
• Lexis+ with Protégé (LexisNexis): Small firms already on Lexis, or litigators who want Shepard's checks built into AI research and drafting.
• vLex Fastcase (Fastcase Library): Solos and small firms that want a free research database with a citator before paying for an AI research tool.

Related playbooks

• Best AI for Contract Drafting at Small Law Firms (and How to Use It Ethically)
• AI for Consultants and Small Agencies: Research, Proposals and Client Decks (Without the Deloitte Mistake)
• How to Write an AI Policy for Your Accounting Firm (Step by Step)

More AI tools and playbooks for Small & Solo Law Firms →