Short answer: AI for lawyers is most useful as a fast first-draft and summarizing assistant: research starting points, document drafts, record summaries, meeting notes, intake triage and marketing drafts. It is not a substitute for legal judgment. ABA Formal Opinion 512 (2024) applies your existing duties of competence, confidentiality, communication, supervision and reasonable fees, so verify every output and vet every vendor.
A solo practitioner reading about AI for lawyers usually hears two stories at once. One says the tools will do half the work of a practice. The other says a lawyer somewhere just got sanctioned for filing fake cases. Both are true in a narrow sense, and neither helps a three-lawyer firm decide what to try on Monday. This guide is the practical middle: which jobs the tools handle well, which jobs they must never handle alone, what the ethics rules require, and how to adopt them in stages without betting the practice on a demo.
Where AI for lawyers actually saves time
Think of the useful tools in seven categories. General assistants such as ChatGPT, Claude, Gemini and Microsoft Copilot can do several of these, while legal-specific products wrap similar models in a research database, a document workflow or a practice management system.
Legal research assistants. These answer a research question in prose and point to authorities. The good ones work from a defined database of cases and statutes and link every proposition to a source you can open. Their real value is orientation: getting from a blank page to the right doctrine and the right search terms in minutes. They are a starting point for research, never the end of it.
Drafting and document review. First drafts of demand letters, engagement letters, discovery requests, routine motions and client emails come out quickly and usually need structural rather than cosmetic edits. Review tools compare a contract against your preferred positions, flag missing clauses and summarize differences between versions. A lawyer still decides what matters.
Summarizing records. This is where many small firms feel the difference first. Medical records, deposition transcripts, police reports, financial statements and long email chains can be condensed into timelines and issue lists. The summary tells you where to look; it does not replace looking.
Intake and after-hours chat. A configured assistant can collect contact details, practice area and location, screen out matters the firm does not take, and offer consultation times. It must not answer legal questions. The intake form behind it matters just as much; our guide to the law firm intake form covers fields, disclaimers and routing.
Marketing drafts. Outlines, first drafts of practice pages, social posts and newsletter copy. Useful as a starting point, weak as a finished product, because generic drafts read like everyone else’s and can state the law wrongly for your state.
Meeting notes and transcription. Recording and summarizing internal meetings, client calls (with consent where required) and dictation. Valuable for follow-up lists, risky if the recording or transcript sits with a vendor whose terms you never read.
Billing and time narratives. Turning a day of calendar entries and emails into draft time entries. Helpful for capturing time you would otherwise forget, but every entry must reflect work actually done, at the time actually spent.
Task by task: the machine’s share and the lawyer’s share
The pattern across every category is the same. The tool produces volume quickly; the lawyer supplies judgment, verification and accountability. Here is how that split looks for the work a small firm does every week.
| Task | Good use of AI | Keep human |
|---|---|---|
| Legal research | Orienting on an unfamiliar issue, suggesting doctrines and search terms, summarizing an opinion you have already pulled | Reading every authority cited, confirming it exists, says what is claimed and is still good law |
| Drafting pleadings and letters | A first draft from your facts and your template, alternative phrasings, tightening long passages | Strategy, every factual assertion, every citation, the decision to file or send |
| Contract review | Flagging missing or unusual clauses against your checklist, comparing versions | Deciding which risks matter to this client and what to negotiate |
| Record and deposition summaries | Timelines, issue lists, finding where a topic appears in a long transcript | Checking key facts against the source pages before relying on them |
| Client intake | Collecting contact details and matter type, screening fit, booking consultations | Conflict checks, case evaluation, any statement about the merits or likely outcome |
| Marketing content | Outlines, first drafts, headline options, repurposing an article into shorter posts | Legal accuracy, state-specific detail, anything stating results or comparing the firm |
| Billing narratives | Drafting entries from calendars and documents | Approving each entry and the time charged |
What stays with the lawyer, always
Three things never transfer to software. The first is legal judgment: deciding what the client actually needs, which argument to make, which risk to accept. A model predicts plausible text; it does not weigh a client’s interests or understand what a particular judge will tolerate.
The second is final review. Anything that leaves the firm with a lawyer’s name on it, whether a filing, a letter, a contract or a web page, has been read in full by a lawyer who is prepared to defend every sentence. “The tool drafted it” is not a defense anyone should plan to use.
The third is anything client-facing that describes outcomes. A chatbot telling a visitor they have a strong claim, a generated page describing the firm’s record, or an email estimating what a case is worth all carry professional risk. ABA Model Rule 7.1 bars false or misleading communications about a lawyer’s services, and a machine-written statement on the firm’s site is still the firm’s statement. Advertising rules vary by state; our attorney advertising rules by state hub summarizes the differences.
Formal Opinion 512, applied to a small practice
In July 2024 the ABA issued Formal Opinion 512, its main guidance on lawyers’ use of generative AI. It creates no new rule. It reads the existing Model Rules against a new tool, and for a small firm the practical consequences sort into five duties.
Competence under Rule 1.1 means understanding, in reasonable terms, what a tool does well and how it fails before relying on it. You do not need to know how a model is built. You do need to know it can produce confident errors, and to check accordingly.
Confidentiality under Rule 1.6 is the duty most often broken by accident. Pasting a client’s facts into a consumer chatbot whose terms allow the provider to keep and learn from inputs is a disclosure risk. The opinion takes the view that informed client consent is needed before confidential information goes into a tool that may use it in ways that could expose it, and that generic language buried in an engagement letter is not enough. Business-tier accounts with contractual protections change the analysis, which is why vendor terms matter.
Communication under Rule 1.4 may require telling clients when AI use affects their matter, for example when a client asks, when the engagement terms address it, or when use bears on decisions the client should make. Supervision under Rules 5.1 and 5.3 means partners set firm policy and supervising lawyers make reasonable efforts to ensure staff, contractors and vendors use the tools consistently with the firm’s obligations. And reasonable fees under Rule 1.5 means hourly billing reflects time actually spent; the opinion also cautions against charging clients for the time a lawyer spends learning a general-purpose tool.
Many state bars have issued their own AI guidance, and some go further or differ in emphasis, so read yours before writing a firm policy. This is not legal advice; check your state bar’s rules.
Fabricated citations and the verification habit
The most visible AI failure in law is the invented authority. Courts in several jurisdictions have sanctioned lawyers for filing briefs that cited cases which did not exist, or quoted language no court wrote, after a general assistant produced them and nobody checked. Some judges now require a certification about AI use in filings, so check the standing orders and local rules of the courts where you appear.
The cause is not mysterious. A general assistant generates text that looks like a citation because citations are a pattern it has seen many times. Tools built on a verified legal database reduce the risk but do not remove it; they can still misdescribe a holding or cite a case for a proposition it does not support.
The fix is a habit, written into firm policy: no citation goes into any document until a person has pulled the source, read the relevant passage and confirmed it is still good law. Quotations are checked word for word. The same rule applies to statutes, regulations and deadlines quoted in client letters and on the firm’s website. Build the check into the workflow so it does not depend on someone remembering at 11 p.m. before a filing deadline.
Questions to put to any AI vendor before signing
Vendor diligence is where confidentiality is won or lost. Get written answers, from the contract or the vendor’s published terms rather than a sales call, to a short set of questions.
Does the vendor train its models, or let a model provider train, on your inputs and outputs, and is that off by default or something you must switch off? How long are prompts, uploaded documents and transcripts retained, and can you delete them on demand? Where is the data stored and processed, and which subcontractors touch it? Who at the vendor can view your content, and under what circumstances? Will the vendor agree to confidentiality terms suited to a law firm and notify you promptly of a security incident? What happens to your data if you cancel or the vendor is acquired? Does the product keep an audit log of who used it on which matter?
For a free consumer account, the honest answer to several of those questions is often unfavorable, which is why firms that use general assistants for client work usually move to a business or enterprise tier with clearer terms. Read the current terms yourself; they change.
A staged adoption plan for a small firm
Consider a hypothetical four-person family law firm: two lawyers, a paralegal and an assistant. Trying everything at once produces a mess of accounts and no policy. A staged approach works better.
Stage one, the first month, uses no client data at all. The lawyers use a general assistant for internal tasks: drafting a firm newsletter outline, rewriting a template letter, summarizing a public opinion. The point is to learn how the tools behave, including how they fail. One partner writes a one-page AI policy covering approved tools, banned uses and the verification rule.
Stage two, the next two months, brings in a business-tier tool after the vendor questions are answered, and limits it to tasks with a clear review step: record summaries checked against source pages, first drafts of routine letters, meeting notes. The paralegal is trained on the policy and the partners spot-check outputs.
Stage three adds client-facing and research uses: an after-hours intake assistant restricted to collecting details and booking times, and a research tool built on a legal database. Engagement letters are updated to describe AI use where the firm’s state guidance suggests it. Stage four is a quarterly review: which uses saved real time, which produced errors, and whether the policy still matches state bar guidance.
How AI is changing the way clients find a lawyer
The other side of AI for lawyers is the client’s side. People who once typed “divorce lawyer near me” now also ask an assistant to explain their situation and suggest who to call. Assistants often answer recommendation questions by pointing at directories, and when they do name firms, they tend to name ones that are consistently described across many independent sources. Our post on whether ChatGPT recommends your law firm includes a ten-minute test you can run today.
Google’s AI Overviews add a second shift: informational legal questions increasingly get a summary at the top of the results, while hiring searches still lean on the map and reviews. Our guide to AI Overviews for law firms covers what a firm can and cannot control there. The broader discipline of making a firm easy for these systems to identify and cite is usually called generative engine optimization, and it rests on the same foundations as good search work: accurate, specific pages and consistent firm details everywhere.
Frequently asked questions
Is it ethical for lawyers to use AI?
Yes, within existing duties. ABA Formal Opinion 512 (2024) says competence, confidentiality, communication, supervision and reasonable fees all apply to generative AI. State bars add their own guidance, so check yours.
Can I put client information into ChatGPT?
Not casually. Check whether the account’s terms allow the provider to retain or train on inputs, consider whether client consent is needed, and prefer a business-tier tool with contractual confidentiality protections for any client work.
Do I have to tell clients I use AI?
Sometimes. Under Rule 1.4 disclosure may be needed when a client asks, when AI use affects decisions the client should make, or when confidential information goes into a tool. Many firms address it in the engagement letter.
Can I bill clients for time saved by AI?
Hourly billing must reflect time actually spent, so time the tool saved cannot be billed as if it were worked. Flat fees and how tool costs are passed on raise separate Rule 1.5 questions to settle in advance.
What is the safest first use of AI in a small firm?
Internal tasks that involve no client data, such as rewriting templates or outlining a newsletter. They teach how the tools fail before anything confidential or client-facing is at stake.
The short version
Use AI for volume: first drafts, summaries, notes, intake triage and marketing outlines. Keep judgment, verification and anything stating results with a lawyer. Read Formal Opinion 512 and your state bar’s guidance, get written answers on data retention and training before you sign with a vendor, check every citation by hand, and adopt the tools in stages with a written policy.
Wondering whether your website is ready for clients who arrive through AI tools, and whether its intake and disclaimers hold up? Run our free law firm website scan to see what needs fixing first.
