Which AI Is Best for Lawyers in Australia? Choose by Task

There is no single best AI for an Australian law practice. The choice depends on the task, with drafting and client updates suiting a general assistant on a business plan and matter-linked work suiting the AI inside your practice software. Whichever you choose, check your regulator's guidance before client material goes in.

The question "which AI is best" hides three separate decisions: what task, what data it touches, and which tools you already pay for. A practice that answers those in order rarely needs more than one general assistant and the AI built into its practice management system. The sections below set out a decision rule, then test it on real tasks.

Follow me on Instagram@sagnikteaches

A decision rule in three questions

Ask these in order for each task you want to hand over. Stop at the first one that settles it.

Connect on LinkedInSagnik Bhattacharya

1. Does the task need the matter file? If it must read your client's documents, time entries or correspondence, prefer AI that lives inside your practice software, because the data stays in a system you have already assessed.
2. Does the task need a legal authority to be right? If an incorrect statement of law would matter, use a research tool that links to sources you can open, and open them. A general assistant is for questions and structure, not authority.
3. Is it mostly language? If the task is rewording, summarising a document you supply, or drafting a plain-English letter from your instructions, a general assistant on a business plan is the efficient choice.

A practice with no practice-management AI answers question 1 differently: either it finds a tool that offers it, or it keeps matter-based work manual and uses a general assistant on anonymised material. General models versus a dedicated legal assistant goes further on question 2.

Subscribe on YouTube@sagnikteaches

Task one: drafting letters, emails and routine documents

Drafting is the strongest case for a general assistant. ChatGPT Business and Claude Team are the business tiers of two of the main assistants. ChatGPT Business was listed at $25 a user a month on monthly billing or $20 on annual billing, with a two-seat minimum, and Claude Team also requires at least two seats; check current figures and your local billing currency on each vendor's page, because both change often. Business plans from both do not train on your content by default, but verify that in the current terms.

A workable pattern is to keep a Project (ChatGPT Projects or Claude Projects) for the practice holding your style notes and precedent letters, and to start each draft from facts you type in, not from client files. An illustrative prompt for a residential conveyancing update:

Draft a client update email in our practice style (see project notes). Facts: contract of sale received from the vendor's solicitor today; special conditions include a sunset clause and an easement; we will review and advise by Friday; client should not sign anything yet.
Do not give advice on the sunset clause or easement. Say we will call before Friday if anything needs a decision.

What you would fix: any sentence that sounds like advice on the special conditions, and any reassurance ("nothing unusual here") the facts do not support. A second example: a plain-English explanation of a court process for a client. Ask for a reading age around a general audience and then check every legal statement yourself, because a smooth paragraph can hide a wrong step. Rough saving: a routine update email drops from about ten minutes to about four including review.

Task two: legal research and authorities

For research, the rule is the opposite: choose for source links and coverage, not for fluency. A general assistant can suggest headings and arguments, but it can invent case names or misstate a holding. Courts now write that risk into their rules: the practice note described further down requires the author of any AI-assisted written submission to confirm that every authority and legislative reference exists, is accurate and is relevant, and says that check cannot be done by an AI tool alone. Anything you cite is opened and read in the primary source.

When you trial a research tool, use questions with known answers. Ask three things from your own practice area, one with a recent change in the law, one with a trap (a question that sounds like a common issue but turns on a statutory exception), and one for which you know the leading authority. Score each answer on: does it cite, do the citations open, does the proposition match the source, and does it say when it is unsure. A two-week trial scorecard gives you a template, and best AI tools for attorneys compares tools built for research, drafting and intake, though much of that comparison is based on overseas products, so check coverage of local legislation and the courts you appear in.

A realistic mistake for a practice working across state lines: a junior asked a general assistant how long a buyer had to cool off after signing a residential contract, and pasted the answer into a client email. The period was real, but it belonged to a different state, and nothing in the answer said so. The supervising lawyer caught it because her rule was that any statutory period is checked in the current legislation for the right state. General assistants tend to give the most common answer they have seen, which is often not the local one, so ask "which state's law is this?" every time.

Task three: client updates, intake and file notes

Client updates sit between the first two. They are language-heavy but use matter data (what has happened, what is next). This is where practice-software AI is attractive. Clio's help pages say its AI watches matter activity (completed tasks, uploaded documents, stage changes) and drafts a client update when one looks due. Its AI page says some features come with every plan while others, including AI drafting and billing, need select plans or add-ons; Clio's main pricing page, as archived on 20 October 2026, starts at $49 per user per month and puts automatic client-update drafting in the Core plan and above. Check the pricing page for your own market before comparing.

Smokeball, a practice management product, offers Archie AI. Its Australian pricing page lists four plans: Bill from $59, Boost from $129 and Grow from $249 per user per month before tax, with Prosper+ on custom pricing, and Archie appears as an add-on on Boost, Grow and Prosper+ without a published add-on price. Smokeball says Archie runs under zero-data-retention agreements and that firm data is never shared with third parties for training. Whether Archie suits you depends on how you work, so request a demo using your own matter types, ask for the Archie price in writing, and ask where data is processed and retained.

The test for either: take one closed matter, ask the tool to produce a client update and a file note from the record, and compare with what you actually sent. If the tool invents a step that did not happen, that is a red flag; if it leaves out something important, it is less severe but tells you what to add to the prompt.

General assistants versus built-in practice-software AI

For a practice that has not chosen software yet, a side-by-side of Clio, Smokeball and Lawmatics sets out the platforms, and the point worth carrying over is that practice software is a ten-year decision, so do not pick it for its AI alone.

What to verify against your regulator before you start

Local rules matter more than any product feature. On 6 December 2024 the regulators of the three Uniform Law jurisdictions issued a joint Statement on the use of artificial intelligence in Australian legal practice. It says lawyers cannot safely enter confidential, sensitive or privileged client information into public chatbots, must review a commercial tool's contract terms before using it with client information, cannot rely on AI output in place of their own judgement, must personally verify anything AI helped prepare, and must bill only for work actually done. It also suggests limiting AI to lower-risk tasks that are easy to verify. Courts have issued their own practice notes too. One state Supreme Court's note, in force since 3 February 2025, bars generative AI from generating the content of affidavits, witness statements and character references, requires those documents to state that it was not used, and needs the court's leave before AI is used on an expert report. Other courts take different approaches and the position keeps changing, so read the current note for each court you appear in, and do not rely on a summary like this one.

Practical steps that come out of that:

• Read your own state or territory regulator's current page on AI, and note the date you read it.
• Check each court's practice note before using AI on anything filed.
• Do not put confidential or privileged material into a public or free tool; use a business plan whose terms you have read.
• Be ready to explain your AI use to a client and check whether your costs agreement should mention it.
• Keep the checks in writing: who reviewed which output and when.

Privilege is worth a separate thought. Sharing privileged material with a third-party service raises the question whether confidentiality is preserved, which is why the vendor's data terms (no training, limited retention, access controls) are part of the decision. A practical test is whether you could explain the arrangement to a client without embarrassment. The checklist in AI tools for lawyers in Australia: a safe starting checklist covers this at more length.

Worked choice: a four-lawyer practice with three jobs

Say a four-lawyer practice does wills and estates, with a practice management system and no AI so far. Its three recurring jobs are: client update emails, summarising incoming medical or financial records, and drafting standard letters to banks and registries. Walking the rule:

• Client updates: need matter data (stage, dates), so first check the practice software's AI. If it drafts updates from tasks, trial it on three closed matters. If not, use a general assistant with facts typed in.
• Summarising records: contains sensitive personal data, so only a business-plan tool or practice-software AI whose terms the partners have read; strip identifiers when testing. Check each summary against the source pages.
• Standard letters: language-only, low risk, use the general assistant with a letter library as a Project.

Cost for the general part, illustrative: four seats at $25 a month is $100 a month (a minimum of two seats applies). Time saved: if each lawyer saves 30 minutes on a working day, that is around two hours a day across the practice, which will not be exact, so time five tasks both ways first. The decision to add a dedicated research tool comes later and depends on how much research the practice does.

A research prompt and the answer you should expect

Here is what a careful use of a general assistant for research looks like. The aim is a map of the issues, with every authority left for you to find.

I am a solicitor advising on a dispute over a residential building contract where the builder abandoned the job after a 30 percent payment. List the legal issues I should consider (breach, termination, remedies, limitation, statutory protections for home owners, evidence).
For each issue give one sentence on why it matters and a question I should answer from primary sources.
Do not cite cases or section numbers. If you are not sure, say so.

An illustrative answer comes back as eight issue headings with a question under each, such as "was the termination valid under the contract's own notice provisions?" and "what is the limitation period for each head of claim?". That is useful, because it is a checklist you can take into a proper database. What you would fix: the assistant may include an issue that does not apply in your state, or leave out a statutory scheme that does, so treat the list as a prompt for your own knowledge. The instruction "do not cite cases or section numbers" is deliberate, since it removes the one thing the tool is most likely to invent.

Questions to put to any vendor, in writing

Whether it is a general assistant or practice-software AI, ask the same questions and keep the answers on file.

1. Is our content used to train or improve any model, for us or for anyone else?
2. Where is our content processed and stored, and can that be restricted to a region?
3. How long are prompts and outputs retained, and can an administrator delete them?
4. Who at the vendor can read our content, and under what conditions?
5. Can we export our data, including prompt history, if we leave?
6. What happens to our content if a subprocessor changes?
7. Can AI features be switched off per user or per matter type?

Clio's help centre, for instance, warns that depending on where a firm is located, its AI features may process queries on servers outside the home jurisdiction, with the resulting data then stored in the firm's own region. That is the kind of answer you need to have read before relying on a feature for client data. Where a vendor's answer is vague, ask for the contract clause, not a summary on a sales call. Published plan prices often leave out AI add-ons, so also ask for a written quote that names the users, plan, add-ons and contract term.

Four mistakes practices make when choosing

Buying the research tool first. A practice with mostly transactional work, such as conveyancing and wills, may do little research. Spending several hundred dollars a month on a research subscription for it is poor value compared with a general assistant and a precedent library. Count how many research hours you really bill in a month before deciding.

Assuming built-in means safe. A feature inside your practice software is not automatically covered by the data terms you signed. Check whether it uses a separate model provider and whether that provider's terms are passed through.

Letting each lawyer pick their own tool. Three lawyers with three free accounts means client material in three places and no way to audit any of it. Approve one or two tools and ban the rest for client work.

Skipping the time test. A common result when a practice does time it: drafting is faster, but for complex letters the checking takes as long as writing from scratch. The sensible response is to keep the tool for routine letters only. Without timing five tasks, a practice can pay for a seat nobody uses well.

How the rule plays out in conveyancing, litigation and family law

A high-volume conveyancing or wills practice leans on templates and client updates, so the best return is usually practice-software AI plus a general assistant for plain-English explanations. A litigation boutique spends more on research and document review, so a research tool with source links and a careful trial matter more, and the court practice notes loom larger. A family law practice handles highly sensitive personal data, so privacy terms weigh more than features; privacy and cost for family law goes into that. A solo practitioner cannot afford several subscriptions and gains most from one business-plan assistant, one note tool and disciplined checking.

The same decision rule produces different answers because the tasks differ. That is the reason a "best AI" list is the wrong starting point and a task list is the right one.

Filling in the decision for your own practice

Copy this into a document and complete it for each recurring task. It takes about fifteen minutes for ten tasks and gives you a defensible record of why you chose what you chose.

1. Task and how often it happens each month.
2. Does it need matter data? If yes, practice-software AI first.
3. Does a wrong answer on the law matter? If yes, a research tool and primary-source checks.
4. Reviewer and what they check (dates, authorities, tone, anything that sounds like advice).
5. Tool approved, plan, seat cost, and the date you last read its data terms.

A completed row might read: "Client updates, 60 a month, needs matter data, practice-software AI trial on three closed matters, reviewed by the matter lawyer for promises and dates, terms read this week." Rows like that make the choice visible to partners and easy to revisit when a tool changes its plans, which happens often.

A short trial before any commitment

Run any tool for two weeks on non-sensitive or anonymised material, and score it. Use five real tasks: a client email, a summary, a research question you know the answer to, a time narrative, a letter from a precedent. Record time with and without the tool, the number of corrections, and whether anything was wrong in a way you would not have noticed. Agree beforehand what result would make you buy: for example, at least 20 percent faster on three of the five with no uncaught error.

If a vendor will not let you trial with your own task types, or will not answer data questions in writing (where it is processed, how long it is retained, whether it is used for training), treat that as a reason to wait. Record the trial scores next to the task rows you filled in above, so the partners can see why a tool was bought or turned down when it comes up for renewal.

If you would rather map this with someone, that is what my AI implementation consultation covers: finding the jobs where AI pays off and automating them or adding AI to the tools you already use.

Questions before paying for legal AI

Should I start with a general assistant or practice-software AI?

If you already use a practice management system, look at its built-in AI first, because it sits inside your matters and billing and usually has the right data terms. Add a general assistant for drafting and thinking tasks that do not need matter data.

Is there an Australian-specific legal research AI I should use?

Several vendors sell research tools with local content, and the market changes fast. Judge them on whether answers link to primary sources you can open, on coverage of the courts you appear in, and on a short trial with questions you already know the answers to.

Can I put client documents into ChatGPT or Claude?

Only on a plan whose current terms protect your content and with your confidentiality duties satisfied, and never on a free personal plan for identifiable client matter. Read the data terms, switch off training where offered, and follow your regulator's guidance.

How do I know if my regulator has said anything?

Check the website of your state or territory law society or legal services regulator, and the practice notes of each court you appear in. Several have issued guidance on generative AI, and courts in particular can require a declaration or restrict its use.

Further reads

• AI Voice Agent for Lawyers: Cost and Safe Handling of Intake Calls — Cost and safe handling of intake calls.
• Free ChatGPT for Lawyers: Where the Free Plan Stops Being Safe — Where the free tier stops being safe for client work.
• Best AI for Immigration Lawyers: Intake to Filing, Stage by Stage — Intake to filing, stage by stage.
• AI Billing Software for Lawyers: Which Features Earn Their Cost — Which billing features earn their cost.
• AI Setup for a Two-Lawyer Estate Planning Firm: The First Month — A two-lawyer practice's first month with AI.
• Lexis+ With Protégé Pricing: Online Plans for Firms of One to Three — What a research subscription costs a small firm.
• Best AI for Solo Lawyers: A Stack That Fits One Person — A lean tool stack for a one-lawyer practice.
• AI Contract Review Software for Lawyers: Spellbook vs Gavel Exec — Two contract review tools compared, if that is your work.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Clio's AI page, help centre and pricing page (archived 20 October 2026), Smokeball's Australian pricing and Archie pages, OpenAI and Claude pricing pages, the joint regulators' Statement on the use of artificial intelligence in Australian legal practice (6 December 2024) and Supreme Court Practice Note SC Gen 23. Check each vendor, your regulator and each court for current terms.