In June 2023, a federal judge in Manhattan fined two New York lawyers and their firm $5,000 after they filed a brief citing six cases that didn't exist. ChatGPT had produced them. One of the lawyers had even asked the chatbot whether the cases were real. It said yes.
That case, Mata v. Avianca, still gets passed around as proof that AI doesn't belong in a law office. I read it differently. The tool wasn't the only failure. The missing review step was, and that's a problem a firm can design around.
Which AI tools help law firms? The safest wins are administrative: intake summaries, scheduling, document cleanup, first drafts of routine letters and internal meeting notes. Each has clear inputs and a named reviewer. Legal advice, citations, evidence analysis, strategy and decisions about confidential data still need a licensed lawyer every time.
Start smaller than the vendor demo
A local law firm doesn't need an enterprise AI transformation program to benefit from automation. It needs one repetitive task with four traits: clear inputs, a predictable output, a human reviewer, and low risk if the first draft is imperfect.
Look at how a small practice already presents itself. Someone visiting the website of a Grand Strand DUI lawyer may encounter practice-area information, educational content, consultation pathways and office contact details. Johnny Gardner Law currently maintains DUI-focused pages, legal articles, consultation options and public information for clients across Horry County, South Carolina.
Every one of those public pages implies a quiet internal workflow behind it: answering the same intake questions, confirming consultations, updating an FAQ when a statute changes. AI could potentially help a firm organize or draft internal material supporting those workflows. It shouldn't be anywhere near the decisions those workflows lead to.
10 law firm tasks you can hand to AI this quarter
- Intake summaries. Turn a web-form submission into a short summary for the attorney. The person who books the consult reviews it.
- Appointment confirmations and reminders, written from templates the firm has already approved.
- Internal meeting notes. Summarize staff meetings. Leave client calls out unless the client agreed and the vendor's retention terms are clear.
- Routine letter drafts. Status updates, document-request reminders and cover notes, edited by the sender.
- Formatting and cleanup. Make scanned PDFs searchable, fix heading styles, build an exhibit table.
- Plain-English FAQ drafts built from the firm's own published pages, approved by a lawyer before they go live.
- Document sorting. Group a file set by date, sender and type before a person reads it.
- Time-entry narratives drafted from calendar entries and corrected by the attorney who did the work.
- Process checklists. Turn a partner's spoken routine into a written step list the team can follow.
- Marketing drafts, like blog outlines and newsletter copy, checked against your state bar's advertising rules.
Notice what these share. If the AI gets one wrong, a paralegal or the sending attorney catches it before anything leaves the building. That's the test I'd apply to any task on a firm's list: could the mistake reach a client, an opposing party or a judge without a person seeing it first? If yes, it isn't an automation candidate yet.
5 jobs that stay with the lawyer
AI should not be allowed to independently decide any of these:
- What legal advice a client receives. A model can describe DUI law in general terms. It can't weigh this client's record, this officer's report and how this county's courts tend to handle it.
- Whether a citation is genuine. Pull every case from Westlaw, Lexis or the court's own records before it goes in a filing. Mata v. Avianca is why.
- What evidence means. Summaries flatten detail. A breath-test log or a bodycam timestamp needs a trained reader.
- What legal strategy should be used. Strategy rests on judgment, risk tolerance and the client's goals. No prompt captures that.
- Whether confidential information can safely be uploaded. That's a firm policy question, answered before anyone pastes a client file into a chatbot.
The professional remains responsible for the work. AI changes how fast the first draft arrives, not who signs it.
The American Bar Association said much the same in Formal Opinion 512, its first ethics guidance on generative AI, issued in July 2024. It points lawyers to duties they already have: competence, confidentiality, honest communication with clients and reasonable fees. It's worth reading before you sign any vendor contract.
Before you upload a single client file
If nobody at the firm can answer those four questions for a tool, the tool shouldn't see client data. Use it on internal templates and public material until someone can.
Where I'd start on Monday
Pick the intake summary. Run it for four weeks with a paralegal reviewing every output, and write down how many minutes each one actually saves compared with doing it by hand. If the number is real and the error log is boring, add task two. If it isn't, you've lost a month of tinkering, not a client.