AI for law firms in Australia: what actually works

Short answer: AI earns its keep in five places inside an Australian law firm: drafting correspondence and letters of advice for lawyer review, triaging discovery documents, building clauses from your own precedent bank, preparing matter intake and conflict checks, and capturing time entries with billing narratives. Everything outside those five either puts the practising certificate at risk or creates a privilege problem.

Most of what is published about AI for law firms is written by vendors selling legal AI products. That matters, because the gap between what the sales demo shows and what the conduct rules allow is where a principal's risk sits. This guide covers the workflows that actually return time, the ones that will get a practitioner into trouble, and the compliance reality that most vendor articles skip.

If you are a principal, practice manager or managing partner at an Australian firm of 3 to 50 lawyers, this is written for you. If you are still working out where to start with AI in your business, start there.

First-draft correspondence and letters of advice

This is the highest-return, lowest-risk use of AI in a law firm, and it is the one to try first.

A general-purpose AI assistant on a business or enterprise tier (Claude, ChatGPT or Microsoft Copilot for Microsoft 365) can produce a serviceable first draft of routine correspondence in under a minute: client update letters, letters of advice on settled areas, briefing letters to counsel, responses to straightforward queries, and the internal file notes that always get pushed to Friday afternoon.

The return is not in the quality of the first draft. It is in the time recovered from the blank page. A routine letter of advice typically takes 20 to 40 minutes to draft from scratch. Reviewing and editing an AI draft of the same letter takes 5 to 15 minutes. Over a week with 15 or 20 such letters, that compounds.

Smokeball's Archie AI works embedded in Word and Outlook and pulls context from the matter file, so the draft already references the right parties, dates and key facts.¹ LEAP's LawY integration answers legal questions with cited AI answers verified by a panel of 300+ Australian lawyers with at least five years' post-qualification experience, and it is included in every LEAP subscription at no extra cost.²

What it costs to set up: a business-tier AI subscription runs roughly $25 to $50 per user per month. If your practice management system already includes AI features (Smokeball Archie, LEAP LawY, or Clio's AI), you may already have the tool and not be using it. The real setup cost is writing the prompt templates that produce output in your firm's tone with the right disclaimers, sign-off conventions and matter references. A senior associate or practice manager needs to spend half a day building and testing those templates.

One hard rule: no letter of advice leaves the firm without a lawyer reviewing every sentence. AI drafts confidently and incorrectly. The practitioner whose name is on the letter is responsible under rule 4.1.3 of the Australian Solicitors Conduct Rules 2015, which requires a solicitor to deliver legal services competently, diligently and as promptly as reasonably possible.³ "The AI drafted it" is not a defence.

Discovery and document review triage

The second highest-return use, and the one where the time savings are most visible in a litigious practice.

Legal AI research platforms can review large document sets, flag documents likely to be relevant, identify potentially privileged material, and produce first-pass summaries that a reviewer can assess in a fraction of the time a manual review would take. The lawyer still decides what is relevant and what is privileged. The AI reduces the pile they have to read.

Thomson Reuters CoCounsel grounds its document analysis in Westlaw and Practical Law content, holds ISO/IEC 42001:2023 certification for AI management systems, and stores all Australian customer data locally in Australia.⁴ Lexis+ AI (renamed Lexis+ with Protégé in March 2026) integrates AI-assisted research with LexisNexis Australia's legal content library and includes citation validation through CaseBase.⁵ Harvey targets large firms and runs enterprise contracts with minimum seat requirements, which puts it outside the reach of most Australian practices under 50 lawyers.⁶

What it costs to set up: the practical starting point for a smaller firm is the AI research features already in your Lexis or Westlaw subscription, which have expanded significantly in the past 12 months. Budget a day for a senior associate to test the AI review against a known document set (one where the relevant documents have already been identified manually) before using it on live discovery.

Precedent and clause drafting from the firm's own precedents

This is where the practice management platforms earn their keep over a standalone AI tool.

When a lawyer asks a general-purpose chatbot to draft a restraint clause or a sunset provision, the AI draws on its training data: a mix of everything it was trained on, with no certainty about jurisdiction, currency or quality. When the same lawyer asks an AI tool embedded in their practice management system to draft a clause, the tool can pull from the firm's own precedent bank: the clauses that have been reviewed, used and tested in that firm's matters.

Smokeball Archie drafts from matter context and can reference the firm's own document library and 20,000+ court forms.¹ Clio's AI features work with documents already in the Clio workspace.⁷ The output still needs a lawyer's eye, but the starting point is the firm's own work product, not the internet.

What it costs to set up: the AI capability is typically included in your practice management subscription or available as an add-on. LEAP includes all AI features (LawY, Matter AI, AI Prompts, and Generator) in every subscription at no extra cost.² The real cost is organising the precedent bank. If your precedents live in 15 different folders on a shared drive with inconsistent naming, the AI has nothing clean to draw from. Budget two to three days of a senior practitioner's time to organise, tag and update the firm's core precedent set before expecting the AI to do useful work with it.

Matter intake and conflict-check preparation

Matter intake is largely a data-extraction and matching problem, which is the kind of work AI handles well.

When a new matter instruction arrives by email, an AI tool can extract the party names, addresses, matter type and key dates, populate the intake form in your practice management system, and cross-reference the party names against the firm's existing records to flag potential conflicts. The conflict check itself remains a lawyer's obligation under the conduct rules, but the preparation (the data matching) is the part that currently takes a paralegal 15 to 30 minutes per matter and that AI can reduce to seconds of review time.

If your intake process starts with phone calls, an AI receptionist can handle the initial enquiry and book the consultation before the matter reaches a lawyer. Smokeball and LEAP both include workflow automation for intake, and Smokeball's Intake feature captures client information from web forms and pre-populates matter files.¹

What it costs to set up: if you already use a practice management system with intake features, the setup is workflow configuration, not new software. Budget a day for an operations person to map the current intake process, configure the automation, and test it on a week's worth of new matters alongside the manual process. The return is clearest in practices that open more than 20 matters a month.

Time capture and billing narratives

The least glamorous and most immediately profitable use of AI in a law firm is recovering the billable time that currently leaks between the work and the timesheet.

Smokeball's AutoTime feature captures time automatically by tracking which matter documents, emails and applications a lawyer is working in, then generates time entries with billing narratives for review.¹ The lawyer reviews and approves the entries rather than reconstructing the day from memory at 6pm. LEAP's AutoTime works the same way and is included in every subscription.²

General-purpose AI can also clean up billing narratives: turning "Reviewed docs and called barrister" into a narrative that meets the firm's billing standards and is defensible in a costs assessment. This is a small task on a single entry and a large return across a practice billing 500 or more entries a month.

What it costs to set up: if your practice management system includes automatic time capture, turn it on and review the output for a week. If it does not, a general-purpose AI on a business tier can still batch-review and improve billing narratives at the end of each day. Budget an hour to set up the narrative templates and another hour to review the first week's output.

Where it fails and what to never automate

Every vendor article about AI for lawyers ends with a section about "the future of legal AI." This is not that section. This is the list of things that do not work today and that a prudent principal should never automate, regardless of what a product demo implies.

Anything filed in court without a lawyer reading every word. In June 2023, a New York federal court sanctioned attorneys Steven Schwartz and Peter LoDuca USD 5,000 after they filed a brief citing six judicial decisions that did not exist, all generated by ChatGPT.⁸ In Australia, a Victorian solicitor became the first practitioner sanctioned for AI misuse after submitting fabricated case citations generated by an AI legal tool in the Federal Circuit and Family Court in August 2024. He was stripped of the right to practise as a principal and restricted to supervised practice for two years.⁹ Since then, multiple Australian courts have encountered AI-generated false citations, including in the Supreme Court of Victoria, the Supreme Court of NSW and the Federal Court.¹⁰ The pattern is always the same: the tool produced something plausible, and nobody checked.

Advice that goes out under a practising certificate. AI can draft the analysis. It cannot apply professional judgement to a client's specific circumstances. It does not know about the side deed your client signed last year, the undertaking given to the other side in 2019, or the fact that the opponent's solicitor has a history of enforcing costs orders. The practitioner who signs the advice is personally responsible under the conduct rules.³

Pasting privileged or confidential material into a consumer AI tool. A joint statement issued on 6 December 2024 by the Law Society of NSW, the Legal Practice Board of WA and the Victorian Legal Services Board and Commissioner is direct: "Lawyers cannot safely enter confidential, sensitive or privileged client information into public AI chatbots/copilots (like ChatGPT), or any other public tools."¹¹ Free-tier tools typically retain input data for training. Uploading a privileged document into such a tool may waive legal professional privilege, because privilege depends on confidentiality, and once confidentiality is lost, privilege is lost with it. Business and enterprise tiers generally do not retain data, but the data processing agreement must be reviewed against your obligations under the Privacy Act and rule 9.1 of the conduct rules.

Trust accounting. Trust money is governed by Part 4.2 of the Legal Profession Uniform Law in NSW, Victoria and WA, and by equivalent provisions in Queensland. The obligations are statutory and heavily regulated: separation, reconciliation, annual audit, designated trust account signatories. Practice management systems like Smokeball, LEAP and Clio handle trust accounting within their own audited modules, and that is the right approach. Do not use a generative AI tool to produce trust account transactions. A plausible but wrong ledger entry is not a rounding error. It is a compliance breach.

The compliance reality for Australian law firms

Most articles about AI for lawyers treat compliance as an afterthought. For an Australian practice, it is the section that matters most, because the regulatory environment is specific, current and enforceable.

The conduct rules

The Australian Solicitors Conduct Rules 2015 apply in NSW, Victoria and WA (which joined the Uniform Law scheme on 1 July 2022). Queensland's Australian Solicitors Conduct Rules 2023, commenced 27 September 2024, contain equivalent provisions. Three rules bear directly on AI use:³

• Competence (rule 4.1.3): a solicitor must deliver legal services competently, diligently and as promptly as reasonably possible. This means the practitioner must understand the AI tool well enough to evaluate its output. "I did not understand how the tool worked" was the explanation given by the first Australian lawyer sanctioned for AI-generated citations.⁹ It was not accepted.
• Confidentiality (rule 9.1): a solicitor must not disclose any information which is confidential to a client. Entering client data into a third-party AI tool is a disclosure. The tool's data processing agreement must be reviewed, and the client's informed consent obtained.
• Supervision (rule 37.1): a solicitor with designated responsibility for a matter must exercise reasonable supervision over all employees engaged in providing legal services for that matter. AI does not supervise itself. Every AI-assisted workflow must have a named lawyer responsible for reviewing the output.

Court practice notes

Two practice notes directly govern AI use in court proceedings:

NSW Supreme Court, Practice Note SC Gen 23: Use of Generative AI (commenced 3 February 2025).¹² AI must not be used to write, reword or edit affidavits, witness statements or character references; these must be in the deponent's own words. A statement must confirm that no part of the final content was generated by AI. Expert reports cannot use AI without the court's permission. All legal citations must be independently verified, and verification cannot be done by using an AI tool.

Federal Court of Australia, Generative Artificial Intelligence Practice Note GPN-AI (16 April 2026).¹³ Presenting false or inaccurate information to the court is "unacceptable." Privileged, confidential or private information must not be entered into an AI tool in a way that is inconsistent with the obligations applying to that information. The note requires disclosure of which AI tool was used, how it was used, and where in the document it was employed. It distinguishes between public AI tools and closed enterprise systems, with heightened caution for public platforms.

The Federal Circuit and Family Court of Australia also issued a Practice Direction on AI use in May 2026. Check the practice note that applies in your jurisdiction before filing.

Law society guidance

All three major state law societies have published guidance:

• Queensland Law Society, Guidance Statement No. 37: Artificial Intelligence in Legal Practice (31 May 2024), the first Australian law society to issue formal AI guidance. Covers competence, confidentiality, transparency, supervision, accountability and legal costs. References conduct rules 4, 5, 9, 17, 19 and 37.¹⁴
• Law Society of NSW, A Solicitor's Guide to Responsible Use of Artificial Intelligence (January 2026). Published alongside an AI hub, online learning modules and a glossary of AI terms for lawyers.¹⁵
• Law Institute of Victoria, Ethical and Responsible Use of Artificial Intelligence (August 2025). Covers duties of competence, diligence, honesty, confidentiality, privilege and supervision.¹⁶

In February 2026, the QLS released the AI Selection and Use Checklist for Legal Practitioners, developed in collaboration with interstate law societies: the first multi-state resource on AI evaluation and implementation in Australian legal workplaces.¹⁴

Legal professional privilege and waiver

This is the risk that gets the least attention and carries the most consequence.

Legal professional privilege attaches to confidential communications made for the dominant purpose of giving or obtaining legal advice, or in connection with litigation. Privilege depends on confidentiality. If the confidentiality is lost, the privilege is lost with it.

Uploading a privileged document into a public AI tool may waive privilege. The joint statement from the three regulators is unambiguous: lawyers cannot safely enter privileged information into public AI chatbots.¹¹ Enterprise AI tools with contractual data isolation and zero-data-retention agreements are a different matter, but the data processing agreement must be reviewed before any privileged material enters the system. If your firm uses AI for document review or correspondence drafting, the question to answer first is: does this tool's data handling preserve the confidentiality that privilege requires?

The Privacy Act

When a practitioner enters a client's personal information into an AI tool, that is a disclosure under Australian Privacy Principle 6 of the Privacy Act 1988 (Cth).¹⁷ If the tool's servers are overseas, it may also trigger cross-border disclosure obligations under APP 8. From 10 December 2026, the Privacy and Other Legislation Amendment Act 2024 will require organisations to disclose in their privacy policies how personal information is used in substantially automated decisions.¹⁸

If your practice does not have an AI acceptable-use policy, that is the first step. If you have one but it does not address the specific obligations of a legal practice (privilege, the conduct rules, trust data), it needs updating. An AI governance framework is the operating system that keeps the policy current.

What your PI insurer will ask

Professional indemnity insurance is mandatory for Australian legal practitioners. Insurers are starting to ask specific questions about AI use in renewal applications.

The practical questions to expect: which AI tools does your practice use, what data goes into them, who reviews AI output before it reaches a client or a court, do you have a written AI policy, and how do you verify AI-generated legal research. If you cannot answer those five questions today, your renewal conversation will be harder than it needs to be.

The conduct rules, the law society guidance and the court practice notes all point in the same direction: AI is a tool, the practitioner is responsible, and the insurance follows the practitioner. A firm with a clear policy, documented review processes and training records is in a stronger position at renewal than one that cannot describe its AI governance.

How to run a two-week trial on one workflow

If you have read this far and want to test one AI workflow before committing to anything, here is a two-week trial that works for a firm of any size.

Week 1: pick one workflow and measure it.

• Choose one of the five categories above. Correspondence drafting is the safest starting point because the tools are mature, the compliance risk is lowest (no privileged material needs to enter the tool if you draft from a prompt rather than upload documents), and the time saving is immediately visible.
• Time the current process. How long does it take your senior associate to draft 10 routine client letters? Write that number down.
• Set up the tool. If your practice management system includes AI features, turn them on. If not, a business-tier subscription to Claude or ChatGPT costs roughly $25 to $50 per user per month. Write three prompt templates that match your firm's tone, include the standard disclaimers and sign-off, and specify the jurisdiction.
• Run it alongside the manual process for one week. The lawyer drafts the same letters both ways and compares time and quality.

Week 2: compare and decide.

• Time the AI-assisted process over the same volume. How long does the review-and-edit workflow take versus drafting from scratch?
• Check quality. How many of the AI drafts needed material corrections versus minor edits?
• Calculate the return. If the tool saves a senior associate four hours a week on correspondence, and the firm's average hourly rate is $400, that is $1,600 a week of recovered capacity for a tool that costs $50 a month.
• If the numbers work, write the AI policy, brief the team, and switch over. If they do not, try a different workflow. The point is to start with the boring, repetitive work rather than the ambitious idea.

That is the first thing we look for in a Bulletproof AI audit: the workflow that returns the most time for the least risk. For a law firm, that workflow is almost always correspondence, because the volume is high, the risk is manageable, and the saving is felt the same week.

Frequently asked questions

Can lawyers use AI in Australia?

Yes, with conditions. The Australian Solicitors Conduct Rules require competence, confidentiality and supervision in any workflow that uses AI. Court practice notes in NSW and the Federal Court require disclosure when AI has been used in preparing court documents. Law society guidance from NSW, Victoria and Queensland sets out the practical guardrails.³¹²¹³

What are the best legal AI tools in Australia?

There is no single best tool. For a small to mid-size Australian firm, the practical starting point is the AI already built into your practice management system (Smokeball, LEAP or Clio), a general-purpose AI assistant on a business tier for correspondence, and a legal research platform with AI (Lexis+ AI or CoCounsel) for discovery and research. Start with what you have before buying something new.

Do I need to disclose AI use in court documents in Australia?

In NSW, Practice Note SC Gen 23 requires that affidavits, witness statements and character references include a statement confirming no part of the final content was generated by AI.¹² The Federal Court's GPN-AI requires disclosure of which tool was used, how it was used, and where in the document it was employed.¹³ Check the practice note that applies to your jurisdiction before filing.

Can AI replace lawyers in Australia?

No. AI automates drafting, document review and administrative work. It does not replace professional judgement, the duty of competence under the conduct rules, or the legal accountability that sits with the practitioner whose name is on the document. The firms that use AI well will do more advisory work with the time they recover from routine production.

Does using ChatGPT with client data breach confidentiality?

It can. A joint statement by the Law Society of NSW, the Legal Practice Board of WA and the Victorian Legal Services Board states that lawyers cannot safely enter confidential, sensitive or privileged client information into public AI chatbots.¹¹ Free-tier tools may retain input data for training, making the entry a disclosure to a third party in breach of the conduct rules. Business and enterprise tiers with zero-data-retention agreements are different, but review the data processing agreement first.

What happens if a lawyer cites a fake AI-generated case?

In Australia, a Victorian solicitor who submitted fabricated AI-generated citations in 2024 was stripped of the right to practise as a principal and restricted to supervised practice for two years.⁹ In the US, lawyers in Mata v Avianca were sanctioned USD 5,000 for citing six fabricated cases generated by ChatGPT.⁸ Courts in both countries now require independent verification of all citations before filing.