AI Legal Assistant: What It Can Take Off a Law Firm's Desk

An AI legal assistant can draft routine letters, summarize records, pull dates into a calendar, send intake forms, and chase clients for missing files. It cannot give legal advice, say what a document means for the case, or sign anything. In a small firm, the biggest win is the follow-up work, not the legal work.

If you're not a lawyer and need help with a legal problem, this page isn't for you: an AI chatbot is not a lawyer and can't represent you in anything. A legal aid office or your state bar's lawyer referral service is the right place to start.

Key Takeaways

• An AI legal assistant drafts, summarizes, and chases clients for documents well. It doesn't give advice, judge what a document means, or sign anything.
• "Legal assistant" covers two separate O*NET occupations — Legal Secretaries and Administrative Assistants, and Paralegals and Legal Assistants — and small firms blend both.¹²
• The task every AI product page skips is also the one that eats the most unbillable time: getting records out of clients who never log in.

A tool built around the one task on this page that's unbillable, not legal: see pricing if the follow-up work below sounds familiar.

What is an AI legal assistant?

An AI legal assistant is software that does part of a legal assistant's job: drafting, summarizing, scheduling, intake, and following up with clients. It can be a general chatbot such as ChatGPT or Claude, a feature built into practice-management software, or a tool built for one task. None of them replace the person; each takes a slice of the work.

"Legal assistant" is actually two jobs filed separately. O*NET tracks Paralegals and Legal Assistants (23-2011.00) on one record and Legal Secretaries and Administrative Assistants (43-6012.00) on another.¹² In a small firm, the person everyone calls "the legal assistant" usually does tasks from both lists — answering the phone and drafting, but also researching and summarizing records. That blend is the honest frame for the rest of this page.

Three kinds of tool do this work. General assistants like ChatGPT and Claude draft and summarize well, with no legal specialty built in. AI built into the practice-management software a firm already pays for handles tasks like intake forms, time capture, or document assembly. Single-task tools — intake forms, document collection, transcription — do one job and nothing else. None of them replaces judgment; each takes a specific task off the desk.

The legal assistant's week, task by task

Take the legal assistant's real task list and AI covers some of it well: first drafts, summaries, pulling dates into a calendar, and chasing clients for records. It covers intake partly. It doesn't cover judgment — what a document means, what to tell the client, what to file.

Here's the week, in the order a matter usually moves, in O*NET's own wording for each task:

Tool names and prices aren't on this table on purpose: see Best AI tools for lawyers for the attorney-facing list, dated.

Intake and opening the matter

At intake, AI is good at the parts that are really data entry: a form that pre-fills from the prospect's earlier answers, an online booking link, structured questions that route to the right practice area. What it can't do is say whether the firm takes the matter.

Two things stay with a person no matter how good the form is. The conflict check is a judgment call against the firm's own client list, not a form field. And opening the attorney-client relationship — the engagement itself — is the attorney's call, every time.

Start from a list rather than a blank form: the free law firm client intake checklist lays out what to ask for at intake, by practice area, so the form isn't invented from memory on day one.

Getting documents from clients, and following up

This is the task AI handles best, and the one most product pages skip. The work here isn't legal: send an itemized request, follow up until it arrives, and notice when the bank statement is for the wrong month or the ID photo is unreadable. A person still has to look at each document and judge whether it's acceptable.

The reality of the task doesn't match a tidy list. Clients reply by text with a photo of page 1 of 3. They go quiet for a week, then send the wrong year's return. Run your own arithmetic on what that costs: 15 open matters, 3 late documents each, 2 follow-ups per document at about 5 minutes to find the thread, write it, and note it — 15 × 3 × 2 × 5 = 450 minutes, or about 7.5 hours a month, on follow-up a schedule can take over instead of a person retyping the same message. Your numbers will differ; the shape of the problem usually doesn't.

I built DokuTrak for this one task. You send one itemized request per matter. The client gets one secure upload link, with no account to create. Reminders go out on a schedule and read as a file status rather than a nag, something like "Your file: 4 of 6 documents received." An AI first pass flags a file that's the wrong document, the wrong period, unreadable, or expired. You or your assistant still accept every file; the software never does.

Where it honestly loses: if your clients already use the portal inside your practice-management system and actually log in, the request belongs there. DokuTrak is for the clients who never log in and keep emailing photos — which, for most small firms, is most clients. See how the AI first pass flags bad files for what that check does and doesn't do.

Drafting, research and notes: useful, and where it goes wrong

Drafting and summarizing are a good fit for a general assistant: a first draft of a routine letter, a plain-English summary of a long file, a rough outline of a motion. Research is where the same tool gets dangerous, because a fabricated citation reads exactly like a real one until someone checks it.

Courts have been sanctioning attorneys for filed AI-generated citations since 2023, starting with Mata v. Avianca in the Southern District of New York, in a pattern that's grown rather than faded — I covered the cases, the fines, and what a verified legal research tool does differently in ChatGPT for lawyers; I won't repeat it here. The short version: verification isn't optional, and it's part of competence, not an extra step. ABA Formal Opinion 512 treats a working understanding of a tool's limits as part of the competence a lawyer owes a client, with more scrutiny for tasks like research than for something like brainstorming.³

Meeting and deposition notes sit in the same useful-but-supervised category. Transcription and a first-pass summary save real time. What goes in the file — which parts matter, which are privileged, which get flagged for the next hearing — is still a judgment only a person makes.

Scheduling, deadlines and billing

Pulling a deadline out of a filed document and putting it on a calendar is mechanical, and AI does it well: read the date, write the entry. Time capture from activity logs is the same kind of task — it records what happened; it doesn't judge what's billable.

Both still belong to a person at the edges. Missing a deadline is the kind of mistake that ends a case, so whoever owns the calendar checks the entry against the court's own rules, not just the tool's read of the document. And on billing, ABA Formal Opinion 512 is direct: when a lawyer bills hourly, only actual time worked may be billed — hours a tool saved can't be billed as if they'd been worked.³

Is there a free AI legal assistant?

Free general chatbots can draft and summarize at no cost, but many consumer tiers learn from what you type, so they are the wrong place to put client information. ABA Formal Opinion 512 calls for the client's informed consent before information about the representation goes into a self-learning AI tool — and says a boilerplate waiver doesn't meet that bar.³

A paid business plan, AI bundled into your practice-management software, or a single-task tool can each be a fit. What matters is whether the tool learns from your inputs, and what its terms say about confidentiality — a price tag doesn't settle the consent question. Read both before the first upload.

Is it legal to use AI as a lawyer?

Yes, under the same ethics rules that apply to any other tool. ABA Formal Opinion 512, issued July 29, 2024, addresses competence, confidentiality, communication, supervision, and fees for generative AI specifically.³

Three points carry the most weight in practice. Before client information goes into a self-learning tool, the client's informed consent is required — and a generic clause in the engagement letter isn't enough; the consent has to name why the tool is used and the specific risk.³ Managerial lawyers are expected to set a clear policy on what AI the firm allows, and supervisory lawyers have to make a reasonable effort to see that lawyers and nonlawyers at the firm follow it.³ And on fees, a lawyer billing hourly may only bill for actual time worked.³

One more thing worth saying plainly: state bars publish their own guidance on top of the ABA's, and yours governs over anything on this page. This is commentary on a published ethics opinion, not legal advice.

What an AI legal assistant should never do alone

Five things stay with a person, full stop:

• Give legal advice to a client.
• Judge whether a document is sufficient for the matter.
• File or send anything without a lawyer's review.
• Treat a client's document as final without a human look.
• Rule on privilege.

None of this is a limitation to work around. It's the same rule that governs every other tool a firm has ever used: the software flags, drafts, and reminds. A person reads, judges, and signs. That line doesn't move because the tool got better at sounding confident.

Start your 14-day free trial. No card required. Solo is $79/month, Team $199, Agency $449.

Footnotes

1. O*NET OnLine, "Legal Secretaries and Administrative Assistants (43-6012.00)," retrieved 2026-10-01. https://www.onetonline.org/link/summary/43-6012.00 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰

2. O*NET OnLine, "Paralegals and Legal Assistants (23-2011.00)," retrieved 2026-10-01. https://www.onetonline.org/link/summary/23-2011.00 ↩ ↩² ↩³

3. NCBE, The Bar Examiner, Wendy J. Muchman, "Generative Artificial Intelligence Tools: ABA Formal Opinion 512 Provides New Guidance on the Use of AI," Fall 2024 (Vol. 93, No. 3), summarizing ABA Formal Opinion 512 (July 29, 2024). Paraphrased, not quoted from the ABA opinion. Retrieved 2026-10-01. https://thebarexaminer.ncbex.org/article/fall-2024/generative-artificial-intelligence-tools/ ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸