For small and solo practices, AI for law firms pays off most on admin work: sorting client email, updating matter status, answering new inquiries and pulling key facts out of documents. A lawyer still reviews anything that involves legal judgment, and client data only goes into tools that don't keep it. One solo M&A lawyer we work with got back more than 10 hours a week this way.
This guide covers where AI fits in a small firm, what ABA Formal Opinion 512 asks of you, how zero data retention works, and when to buy a tool instead of building one.
How Law Firms Are Using AI in 2026
Most coverage of AI in law firms is about legal research and drafting tools, which also get most of the vendor marketing. In a solo practice or a firm of five, the lost hours usually sit elsewhere: in the inbox, the matter list, intake and documents you read for a handful of facts.
| Use case | Typical tool | Fit for a small firm |
|---|---|---|
| Legal research and drafting | Legal AI platforms such as CoCounsel or Lexis+ AI | Good if research fills a large share of your week |
| Document review and contract analysis | Contract review tools, AI inside document management | Good for transactional work with heavy document volume |
| Client intake | Intake forms, chat and automated follow-up | Good when inquiries arrive after hours or while you're in meetings |
| Billing and time capture | AI features inside your billing or practice-management software | Worth turning on if your software already includes it |
| Email and matter admin | Custom workflows connected to your email and practice-management system | Often the largest time sink, and the one off-the-shelf tools cover least |
Research tools get most of the attention when lawyers look into AI for lawyers. Your own calendar is a better guide. Start with where last week's non-billable hours went.
Where AI Saves Small Law Firms the Most Time
The best first projects with AI for small law firms are tasks that repeat daily and follow the same steps each time. This kind of law firm automation runs alongside the software you already pay for and handles the copying and sorting between systems. Think of it as an AI legal assistant for admin work that leaves the legal advice to you.
Email triage and matter updates
Client emails carry information that belongs somewhere else. A signed document changes a matter's status, and a client question becomes a task. In most small firms, someone reads the email, finds the matter and updates it by hand.
AI email management can do that matching and updating. An agent reads each message, works out which matter it belongs to, and updates the status and tasks in your practice-management system. The case study below shows this workflow in practice.
Client intake and lead follow-up
A prospective client who fills out your contact form at 9 p.m. may be contacting other firms the same evening. AI can reply right away, ask the screening questions you would ask, and book a consultation on your calendar. You still decide whether to take the matter, and you still run the conflict check.
Our guide to speed to lead covers why response time matters for new inquiries.
Pulling data from documents
Engagement letters, purchase agreements and closing documents hold details you end up retyping elsewhere, such as party names, dates and deal amounts. AI document extraction reads those fields and fills them into your system for a person to check.
Our guide to AI document processing covers how this works.
What AI should not do on its own
AI should not make legal judgments, and it should not cite case law that nobody has checked. In Mata v. Avianca, Inc. (S.D.N.Y. 2023), lawyers filed a brief citing cases that ChatGPT had invented. Judge P. Kevin Castel imposed a $5,000 penalty on the two lawyers and their firm (LawSites).
The judge wrote that there is nothing "inherently improper" about a lawyer using AI. He found bad faith because the lawyers kept standing by the fake cases after the court questioned them and made false and misleading statements to the court. For a small firm, that means AI can sort, draft and summarize, while a lawyer reviews anything that goes to a client, a court or opposing counsel.
Case Study: AI Email Automation That Saves a Solo M&A Lawyer 10+ Hours a Week
Nocturnal Legal is a solo M&A and business law firm in Scottsdale, Arizona. The firm runs its client work in Notion, with a project for each client.
The problem
Every client email meant finding the right Notion project and updating its status and tasks by hand. The firm wanted email automation it could trust to be accurate, and it needed confidential client information to stay private.
What we built
We built an AI email processing agent that makes the update for every incoming email:
- A client email arrives.
- The agent reads it and identifies the client project it belongs to.
- It updates that project's status in Notion.
- It updates the related tasks on the same project.
The attorney keeps working in Notion as before, and the project statuses there are already current.
How client data stays private
The agent runs on OpenAI models with zero data retention (ZDR). OpenAI processes each email to produce the update and never stores the email content.
The result
- 10+ hours saved per week. That is more than a full working day, every week.
- Project and task statuses in Notion update automatically from incoming email, without manual entry.
Read the full Nocturnal Legal case study.
If your firm loses hours to the same kind of manual updates, we can look at your workflow with you and tell you which steps AI can take over. Book a Free Strategy Call.
Client Confidentiality: What to Check Before AI Touches Client Data
Your duties under Rule 1.6 apply to every tool you use, AI included.
What ABA Formal Opinion 512 says
The ABA issued Formal Opinion 512 on July 29, 2024. It applies the existing Model Rules to generative AI. The main points (Frankfurt Kurnit summary):
- Competence (Rule 1.1). Understand the tool well enough to use it properly. Relying on AI output without an "appropriate degree of independent verification or review" can violate your duty of competence. How much checking is enough depends on the tool and the task.
- Confidentiality (Rule 1.6). Get a client's informed consent before you enter their confidential information into a self-learning AI tool. Boilerplate in an engagement letter does not count. You have to explain the risk.
- Communication (Rule 1.4). Consider when a client needs to know how AI is being used on their matter.
- Fees (Rule 1.5). You can't charge a client for the time it takes you to learn a tool, unless the client asked you to use that specific tool.
- Supervision (Rules 5.1 and 5.3). Firm leaders are responsible for how lawyers and staff use AI. When you work with a vendor, investigate the tool's reliability, security measures and policies.
- Candor (Rules 3.1 and 3.3). Check every citation before you file.
ABA opinions guide state bars but don't bind them, and several state bars have published their own AI guidance. Read your state's rules alongside Opinion 512.
Consumer AI vs. business plans vs. zero data retention
"We don't train on your data" and "we don't store your data" are separate promises. A provider can keep what you send for weeks without ever training a model on it. OpenAI's plans show the difference:
| Level | Used to train models? | Stored by the provider? |
|---|---|---|
| Personal ChatGPT plans (Free, Plus, Pro) | Yes by default, unless you turn it off in Data Controls (OpenAI) | Yes, in your chat history |
| OpenAI API (standard) | No by default | Abuse monitoring logs kept for up to 30 days by default (OpenAI) |
| OpenAI API with zero data retention | No | Customer content is excluded from abuse monitoring logs. ZDR needs OpenAI's approval (OpenAI) |
Zero data retention AI means the provider processes your request and keeps no copy of the content. Opinion 512's consent rule targets the risk of client information leaking out of a tool later, and ZDR leaves nothing on the provider's side to leak. Whether a given use still needs client consent is a judgment call under your state's rules.
Questions to ask any AI vendor
Before client data goes into any AI product, get written answers to these:
- How long do you keep our inputs and outputs, including logs?
- Do you use our data to train or improve models?
- Do you offer zero data retention, and does it cover every AI model your product calls?
- Where is our data processed and stored?
- Who at your company, or at your subprocessors, can access it?
- Can we review a SOC 2 report?
- Will you delete our data on request and confirm the deletion in writing?
Write a one-page AI use policy
Opinion 512 makes firm leaders responsible for how the firm uses AI. For a small firm, a one-page written policy covers it:
- Which tools are approved, and which plan or setting each one must use.
- What client information may go into which tool.
- Which outputs a lawyer must review before they leave the firm. For anything legal, that means all of them.
- How you tell clients about AI use and get consent when it's needed.
- Who approves a new tool before anyone uses it.
Choosing AI for Law Firms: Legal AI Tools vs. a Custom Workflow
You can buy a product that already does the job, or have a workflow built around the systems your firm runs on. Buying makes sense when the task looks the same in every firm, like research and drafting. Building makes sense when the job depends on how your firm works: your stack has no AI for the task (Nocturnal Legal runs on Notion), the workflow spans several systems, or you need control over which models see client data.
| Your situation | Better fit |
|---|---|
| You want help with legal research and first drafts | Buy a legal AI platform such as CoCounsel or Lexis+ AI |
| Your practice-management software, such as Clio, has an AI feature for the task | Use the built-in feature |
| The work happens inside Outlook and Word | Try Microsoft Copilot first |
| The task moves information between email, your matter system and documents | Build a custom workflow |
| Your firm runs on a general tool such as Notion | Build a custom workflow |
| You need ZDR on every model call and the vendor can't confirm it | Build a custom workflow |
So the best AI for law firms depends on the job you hand it. AI tools for lawyers cover research well. Admin work that crosses systems usually needs custom AI automation built around your stack.
How to Start Using AI in a Small Law Firm
Start with one workflow and measure it.
- List your repetitive admin tasks. For two weeks, note each one and roughly how long it takes.
- Pick one workflow. Choose the one that costs the most hours and follows the same steps every time.
- Set your data rules first. Decide what client information the workflow will touch, and confirm the tools meet your retention and training rules before anything is built.
- Pilot it with human review. Run it on real matters and check every output until you trust it.
- Measure after 30 days. Compare the hours the task takes now against your notes from step 1.
FAQ
Can lawyers use ChatGPT with client information?
Be careful on a personal plan. OpenAI can train on personal ChatGPT conversations unless you turn that off, and Opinion 512 requires informed client consent before client information goes into a self-learning tool. Business plans and the API don't train on your data by default, and API setups with zero data retention don't keep it. Confirm the vendor's terms and your state bar's guidance either way.
What does ABA Formal Opinion 512 require?
Opinion 512 (July 29, 2024) says lawyers using generative AI must understand their tools, verify AI output, protect confidentiality, get informed consent before entering client information into self-learning tools, bill fairly and supervise staff use of AI.
How much does AI cost for a small law firm?
It depends on what you're doing. Off-the-shelf legal AI tools are usually sold as per-user subscriptions. A custom workflow has a one-time build cost plus ongoing AI model usage, which is billed per use. The build cost depends on how many systems the workflow connects, how much email or document volume it handles, how much human review it needs and how strict your data rules are. The useful comparison is against the hours it saves. If a workflow gives back 10 hours a week, compare its cost with what 10 hours of your time is worth.
Will AI replace lawyers?
The ethics rules assume a lawyer stays responsible for the work. Opinion 512 requires lawyers to review AI output, and Mata v. Avianca shows what happens when they don't. What AI can take over today is admin. At Nocturnal Legal, that was more than 10 hours a week.
What is the best AI for law firms?
It depends on the job. For legal research and drafting, a legal AI platform such as CoCounsel or Lexis+ AI is the usual choice. For admin work that crosses your email, matter system and documents, a custom workflow often fits better. The decision table above maps common situations to each option.
Find the Hours AI Can Give Back to Your Firm
Most small firms already know which tasks eat their week. On a strategy call, we go through your workflow with you and point out the steps AI could take over. If confidentiality is your first question, bring it. We'll walk you through how the Nocturnal Legal agent keeps client email off the AI provider's servers.
Not a law firm? Our guide to AI automation for small business covers the same approach for other industries.
