AI for law firms is safest and most useful on administration: intake and triage of new enquiries, chasing clients for documents, capturing time, and summarising or sorting material under a lawyer’s supervision. It should not give legal advice, and the lawyer remains responsible for everything that goes out, including anything the AI drafted.
This article is for principals and practice managers of small and mid-sized firms in Australia. It is general information, not legal advice, and professional rules differ by state and territory.
The admin that eats a lawyer’s week
Ask a solicitor where unbilled time goes and the answers repeat: returning calls from people who turn out not to be a fit, collecting ID and documents from clients who forget, reconstructing the week’s time entries on Friday afternoon, answering “where is my matter up to” and summarising a bundle someone else has already read. None of it needs a legal opinion. All of it takes time from fee-earners.
AI for law firms: intake and first contact
New enquiries arrive by phone, form, email and after hours, and many are not yet clients. An AI intake assistant can collect the same details a receptionist would: who the person is, the type of matter, the other party’s name for a conflict check, how urgent it is and how they would like to be contacted. It can book a call into the right lawyer’s diary and write the details to your practice or CRM system.
Two rules shape the design.
- It collects, it does not advise. It must not tell a caller whether they have a case, what to do about a deadline or what their rights are. If asked, it says a lawyer will discuss that and offers a time.
- It does not create a retainer by accident. Wording should say clearly that submitting details does not make the firm the person’s lawyer until an engagement is confirmed. Settle the wording with your own professional indemnity insurer and regulator guidance.
The phone side of intake is covered in AI receptionists for law firms.
Chasing documents and keeping clients informed
Many matters stall waiting on the client: ID, a signed form, a contract, bank statements. A system can send a checklist at the start, remind on a schedule, receive uploads into the matter and tell the lawyer what is outstanding. It can also answer “what is the next step” from a short, lawyer-approved status that you maintain, rather than from the AI’s guess. That removes a large share of status calls without anyone giving an opinion.
Time capture
AI can draft time entries from calendar events, emails and documents opened, leaving the lawyer to confirm, edit and bill. It is a reconstruction aid, not a billing authority. Each entry stays the lawyer’s responsibility, because costs disclosure and fairness obligations sit with the practitioner, so review every line.
Summaries and first drafts: where the risk lives
Summarising a long email chain or document bundle, drafting a chronology, or producing a first-pass letter can save real hours. These are also the uses courts have focused on. The NSW Supreme Court’s Practice Note SC Gen 23 says generative AI may be used for tasks such as chronologies, indexes and witness lists, and for summarising or reviewing documents and transcripts, while warning of its limits (Practice Note SC Gen 23, opened November 2026). The same note:
- warns of “hallucinations”, including false citations and fabricated legislative, case or other references;
- warns that data behind a tool may not be current or relevant to New South Wales or Australia;
- says information subject to suppression or non-publication orders, material produced under the Harman undertaking or on subpoena, and similar material must not be entered into a Gen AI program unless the practitioner is satisfied it stays within a controlled environment under confidentiality restrictions and is not used to train models;
- bars the use of Gen AI in generating the content of affidavits, witness statements and similar evidence, though it allows work merely preparatory to drafting.
That practice note governs the Supreme Court of New South Wales, and the District Court has adopted it. Other courts and other states have their own directions, so check the ones that apply to your matters, and your law society’s current guidance.
Confidentiality and privilege
A solicitor’s duty of confidentiality and legal professional privilege are the reasons to be strict about where client material goes. Before using any AI tool on matter material, settle:
- Whether the vendor keeps inputs, for how long, and whether they train models on them.
- Where the data is stored, and who the sub-processors are.
- Whether you have a written business agreement rather than a consumer terms-of-use click-through.
- Whether the tool can be limited to one matter or one client’s data.
- What your engagement terms and privacy policy tell clients.
Free public chat tools are the wrong place for client documents. The privacy side is in AI and the Privacy Act and is ChatGPT confidential. Put your rules in writing in an AI policy that every staff member signs.
Who checks what
| Task | AI | Lawyer |
|---|---|---|
| Intake details and conflict data | Collects | Reviews, runs the conflict check, accepts or declines |
| Document chase | Sends and reminds | Handles anything the client questions |
| Time entries | Suggests from activity | Confirms and bills |
| Summaries and chronologies | Drafts | Checks against the source |
| Legal research and citations | May suggest leads | Verifies every authority in a primary source |
| Advice, strategy, court documents | None, or preparatory only | All of it |
Cutting down invented answers
Every citation, quote and statute reference an AI supplies must be checked in the original. Use the habits in our guide on reducing AI hallucinations: restrict the AI to documents you give it, ask it to quote, let it say it does not know, and never accept a reference you have not opened.
A worked example: after-hours intake
Suppose a person contacts a firm at 7pm about a dispute with a neighbour over a boundary fence. An intake assistant takes their name, contact details, the address, the other party’s name for the conflict check and a short description, and tells them that a lawyer will review it and contact them by the next business day. It does not say whether they have a claim. The next morning the matter appears in the system with a conflict-check flag, and the practice manager decides whether to book a call. The enquiry has been captured and nothing has been advised.
The firm’s risk lies in the wording and escalation settings, so test them with awkward callers: someone describing a court deadline tomorrow, someone in distress, someone asking for advice outright.
Supervision, billing and telling clients
A lawyer who delegates to a human assistant remains responsible for the work. The same principle is the sensible starting point with AI, and your regulator and professional indemnity insurer may have specific expectations. Practical steps:
- Record which tools are used for which tasks, and who reviews the output.
- Consider whether your costs agreements and client communications should say that you use AI, and how time saved is reflected in fees.
- Do not bill a client for time you did not spend because a tool did the work faster, without checking how your jurisdiction’s costs rules apply.
- Train every fee-earner and support staff member on what may and may not be entered into any tool.
Your law society or legal services regulator publishes current guidance. Read it for your state, because it changes as the technology does.
Starting small
Pick one admin job, usually document chasing or intake after hours. Write the script and the lines it must never cross, run it on a small set of matter types, and read every conversation for a month. CentriWeb builds AI-assisted admin for professional services firms and starts with the job that costs the firm most. If your practice management software already has intake forms and reminders, switch those on before buying anything.