AI for Law Firms: Where to Start

AI for law firms means using AI tools to draft, summarize and search legal work faster, with a lawyer checking every result. Start with low-risk tasks that use no client information, a one-page policy and one tool tested for a month.
This guide covers what AI can do for a firm, the rules that apply, why checking matters, and a first-month plan. If you want help setting it up, see my AI consulting page.
Who is writing this, and what are the limits?
Nothing here is legal advice. I have been out of the legal profession since 2019, so I do not speak on the legal industry. What I do know is software and automation, and I co-founded PracticePanther, legal practice management software, and was its CEO. Read this as a technology guide, and ask your bar association or an ethics lawyer about your own duties.
What can AI do for a law firm today?
AI can already help with research, drafting, document review, contract analysis, summaries and admin work. The ABA’s ethics opinion on generative AI lists legal research, contract review, due diligence, document review, regulatory compliance and drafting letters, contracts and briefs as typical uses.
The useful question is which task to try first. The table below is my own ranking, based on how much damage a mistake can do and how much client information the task touches. It is a judgment, not a study.
| Task | Uses client information? | Risk if the output is wrong | Good first step? |
|---|---|---|---|
| Drafting the firm’s own blog posts, emails to prospects, job ads | No | Low | Yes |
| Summarizing public material, such as a new statute or a published opinion | No | Medium: check against the source | Yes |
| Meeting notes, scheduling, reminders | Often | Low to medium | After the policy is in place |
| First drafts of letters and routine documents | Yes | Medium: a lawyer reviews every line | After consent and tool review |
| Contract review and summaries | Yes | High | Test on a small sample first |
| Legal research and citations | Yes | High: wrong or invented cases | Only with checks on every cite |
| Anything filed with a court | Yes | Very high | Never without full human review |
Notice what sits at the top: work that does not touch a client file. That is where I would begin.
How many law firms use AI, and what holds the rest back?
Adoption is growing but uneven, and accuracy is the main worry. In the American Bar Association’s 2024 Legal Technology Survey Report, 30% of responding lawyers said they use AI. Use was 46% at firms of 100 or more lawyers and 18% among solo lawyers. Three-quarters of those surveyed said concern about AI hallucinations, meaning confident but wrong answers, was why they had hesitated.
Clients are moving too. Clio’s 2026 Legal Trends Report summary says 24% of US adults have used AI to answer a legal question and 46% would consult AI before contacting a lawyer. Clio sells legal software, so treat that as one vendor’s survey, but the direction matters. Prospective clients may arrive having already asked a chatbot.
Friction loses clients, and it did long before AI. At an earlier company we sent contracts that clients had to print, sign, scan, fax or email back, and we lost a lot of clients that way. I still think the client experience wins the client, and AI is one more way to make a firm easier to deal with.
What rules apply when a lawyer uses AI?
The main guide is ABA Formal Opinion 512, issued July 29, 2024. It applies the existing Model Rules of Professional Conduct to generative AI. It does not create a new rulebook, and each state sets its own rules, so use it as a map and then check your state.
Here is what it says, in plain English:
- Competence. A lawyer has to understand the tool well enough to use it, and must review its output. The opinion says AI can be a springboard for a draft, but a lawyer cannot rely on it alone for work that needs professional judgment.
- Confidentiality. Many tools learn from what you type. The opinion says informed consent from the client is generally needed before client information goes into such a tool, and boilerplate in an engagement letter is not enough. Lawyers should also read the tool’s terms of use and privacy policy.
- Communication. Sometimes you must tell a client you are using AI, for example when it influences a significant decision or affects the fee.
- Candor to the court. Rules on meritorious claims, honesty to a tribunal and misrepresentation can all apply to AI-written filings. The opinion says AI output must be carefully reviewed so that nothing false reaches the court, and it names citations to nonexistent opinions as a problem that has already happened.
- Supervision. Managing lawyers must set clear policies on AI use, and supervising lawyers must make sure lawyers and staff follow them and are trained.
- Fees. Hourly bills must reflect actual time, a flat fee can become unreasonable if the tool does most of the work, and a lawyer cannot bill a client to learn a tool the firm will use regularly.
That list is a summary, not a substitute for the opinion. Your state bar may have its own guidance, and the opinion tells lawyers to check each court’s local rules.
Why does every AI answer need a check?
AI tools produce plausible text, not verified facts. The clearest example is Mata v. Avianca in the Southern District of New York. In a June 22, 2023 sanctions order, Judge P. Kevin Castel found that lawyers had submitted fake judicial opinions and citations created by ChatGPT, then stood by them after the court questioned them. He ordered a $5,000 penalty, jointly on the lawyers and their firm, along with letters to the client and to the judges whose names were falsely attached to the fake opinions. The judge also wrote that there was “nothing inherently improper about using a reliable artificial intelligence tool for assistance.” The problem was the missing check.
Legal-specific tools are better, but not clean. A Stanford study, first posted May 30, 2024, tested leading legal research products from LexisNexis and Thomson Reuters. It found each gave wrong or made-up information between 17% and 33% of the time, which is less than a general chatbot but far from zero.
My rule is simple. AI has access to a huge amount of data, and some of that data is wrong. A blog post that Claude or ChatGPT cites is not automatically accurate. For legal research I would teach the AI to cite and source everything, to work from your firm’s own research account, and to verify that every case is real. Then a person opens each case and reads it. Skipping that last step is how sanctions happen.
Where should a firm start?
Start small, in this order. It takes about a month.
- Week 1: pick the owner and write one page. Name one partner or manager who owns AI. Write a one-page policy: approved tools, what may never be pasted in, who reviews output, and who to ask. Supervision under the ABA opinion starts with a clear policy. For a fuller template, see AI governance.
- Week 2: list what staff already use. Ask everyone, with no blame. People often use AI tools on their own, and you cannot set rules for tools you do not know about.
- Week 3: choose one tool and one task. Pick a task from the top of the table above. Read the tool’s terms and privacy policy before anyone signs in.
- Week 4: test, compare and decide. Run the task by hand and with AI, compare time and quality, and have a lawyer review every output. Keep it only if it clearly helps.
- After that: train and widen. Train everyone on the tool and its limits, then move to the next task. The AI training guide covers how to teach a team, and my AI workshops teach it live.
Small automations pay off first, and reminders are the classic example. You can schedule reminders automatically and never miss a deadline again, and the keyword is automatic. The same logic applies to routine tasks that someone repeats every day. If you would rather have a tool built around your firm’s own rules, I build custom AI apps that you own.
People also make AI agents sound fancier than they are. You can create a paralegal agent by typing “act as my paralegal” into Claude or ChatGPT. Learning to build agents like that is easier than most people think. In a law firm, though, every confidentiality rule above still applies to whatever you feed it. See AI agents for business for how agents work, and my agents video shows how one sentence is enough to start an agent.
How do you choose an AI tool for a law firm?
Choose on data handling first and features second. These questions come from the confidentiality and vendor-review points in the ABA opinion:
- Where does our data go? Who can see what you type, and is it used to train the tool?
- Can we read the terms? If nobody at the firm has read the terms of use and privacy policy, do not use the tool for client work.
- Does it show its sources? You need to be able to click through to each case or clause.
- What controls does an administrator get? Look for user accounts, logs and the ability to turn features off.
- Is the vendor credible? The opinion points to reference checks, security policies and confidentiality agreements when a firm relies on an outside provider.
- Does it fit our work? Test it on real but non-confidential samples before you pay for a year.
Ignore broad claims that a product is “hallucination-free.” The Stanford study above is a reason to test for yourself.
What should a law firm AI policy say?
A short policy is enough to start. It should name the approved tools, ban pasting in anything confidential until the firm has done its review, require a lawyer to check every output, say who may approve a new tool, and explain how and when to tell a client. Add training and a date to review it. You can adapt the checklist in AI governance and have your own lawyer read it. For the wider risk picture, see AI risks for business.
What should you do next?
This week, do three things. Name an owner, ask every person on the team which AI tools they already use, and pick one low-risk task to test. Keep client information out of it until your policy and consent process are ready.
I wrote a book called The Secrets to Marketing & Automating Your Law Practice, which was a number one best seller in law office marketing and law office technology. The tools have changed since then, but the principle has not: automate the repeatable work and keep the people for the judgment. If you want a senior person to help you set up the policy, choose the tool and train the team, start on the consulting page.
Sources
- American Bar Association, Formal Opinion 512: Generative Artificial Intelligence Tools, July 29, 2024.
- ABA Journal, ABA tech report finds that AI adoption is growing, but some are hesitant, March 7, 2025, reporting the 2024 Legal Technology Survey Report.
- Clio, Legal Trends Report, 2026 edition summary, read November 14, 2026.
- U.S. District Court, S.D.N.Y., Mata v. Avianca, Inc., No. 22-cv-1461 (PKC), Opinion and Order on Sanctions, June 22, 2023.
- Magesh, Surani, Dahl, Suzgun, Manning and Ho, Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools, Stanford University, submitted May 30, 2024.


