AI in Wills and Estates Work: Drafting, Capacity Notes and the Limits

By Michael Nadalin, Founder, Market Lead · 7 October 2026 · 9 min read

An 84 year old client arrives with her daughter to change her will. The daughter does most of the talking. The new will cuts out a son who received an equal share under the last one. You have forty minutes, a fixed fee, and a matter that may be picked apart in a Supreme Court probate list a decade from now, long after anyone remembers the conference.

That file is the test for any AI tool in a wills and estates practice. The drafting is the easy part. The value of the solicitor sits in what happened in the room: who was present, what the client said in her own words, what you asked, and what you wrote down at the time. No language model was there.

This post sets out where AI earns its place on an estates file, where it does not, and how to keep your capacity notes and your citations defensible. It is written for practitioners who would rather hear what fails than read another promise.

Where AI helps on an estates file, and where it does not

Estates work has a lot of text that is repetitive but must be exact: clauses, letters of advice, asset schedules, executor explanations, chronologies in contested matters. AI is good at first drafts of that material, and at summarising long documents such as trust deeds, superannuation fund rules and earlier wills so you can see what changed between versions.

It is poor at anything that depends on facts it was not given or judgment it cannot exercise. It does not know that the house is held as joint tenants and will never reach the estate. It does not know that the client's largest asset is super sitting outside the will. It cannot tell whether the client understood your questions or was repeating what her daughter said in the car.

A sensible ranking of tasks, from the ones you should keep entirely to yourself to the ones where a tool does the heavy lifting, looks like this.

Drafting wills with AI: useful for clauses, risky for structure

The right way to use AI in will drafting is as an assistant working from your instructions and your firm's precedents, not as a will generator. Give it the clause you want, the facts that matter and the precedent language, and ask for a draft. Ask it to explain a gift over, a substitution clause or a testamentary trust power in plain English for the client letter. That is real time saved.

The wrong way is asking a general chatbot for a complete will. The output tends to look competent and may borrow from other jurisdictions: wrong terminology, attestation wording that does not match local practice, references to concepts that do not exist in your State or Territory. Formal validity is governed by State and Territory legislation, such as the Succession Act 2006 (NSW) and the Wills Act 1997 (Vic), and the requirements are not identical across the country. A document that looks like a will is not the same as one that meets the formal requirements, and relying on a court's dispensing power to rescue it is an expensive plan.

The other structural risk is silence. A model drafts what it is asked to draft. It will not ask whether the family trust has an appointor succession clause, whether there is a binding death benefit nomination, or whether a company loan account needs dealing with. Those questions are the job.

Capacity notes: AI can structure them, it cannot observe for you

Testamentary capacity in Australia is still assessed against the test from Banks v Goodfellow: whether the will-maker understands the nature and effect of making a will, has a general understanding of the extent of the property, comprehends the claims of those who might expect to benefit, and is not affected by a disorder of mind or delusion that influences the dispositions. Your file note is the evidence of how you tested each limb.

AI has two legitimate uses here. Before the conference, it can turn the test into a question plan for the particular client: open questions, nothing leading, with space for the client's own words. After the conference, it can take your dictated or handwritten notes and arrange them under each limb so gaps are obvious while you can still remember what happened.

What it must not do is supply content. If a tool turns 'client a bit vague on assets' into a polished paragraph saying she accurately described her property, the note is now false, and it is your note. Record the client's actual answers, record who was in the room and when they left, and record why you were satisfied. If you record or transcribe conferences with an AI tool, get informed consent first and check the surveillance and listening devices legislation in your jurisdiction. Where you have real doubt, the answer is a medical assessment with a properly framed letter of instruction, not better prose.

Contested estates and probate: research, chronologies and affidavits

Family provision claims and probate disputes are document heavy. AI is useful for building a chronology from years of correspondence and medical records, for comparing successive wills, and for a first pass at research on eligibility and the factors a court may consider under the relevant State or Territory legislation.

Two cautions apply. First, research output is a lead, not an authority. Read every case and every section before it goes into advice or submissions, and confirm it is still good law. Australian courts and regulators have dealt with practitioners who filed material containing authorities that did not exist, and the consequences have included costs orders and referrals. Second, evidence is different from argument. The Supreme Court of New South Wales has issued a practice note on generative AI, SC Gen 23, that restricts its use in preparing affidavits and witness statements, and other courts have published their own guidelines. Check the current position in the court you are filing in before any AI tool goes near a deponent's evidence.

A practical adoption plan for a wills and estates practice

Most of the risk in adopting AI comes from doing it informally: one solicitor using a personal account, client details pasted into a consumer tool, and no record of what was checked. A short written policy fixes most of that. It should cover which tools are approved, what information may be entered, what must be verified and by whom, and how supervision of junior staff works.

Start with low risk tasks and move up only when the checking habits are in place. Costs matter too: if a task takes a fraction of the time it used to, think about how that sits with your costs disclosure and with fixed fee pricing.

The limits, stated plainly

AI does not take instructions, assess capacity, detect influence, witness execution or carry professional responsibility. It will not notice that the client looked to her daughter before every answer. It will not tell you that the asset the client cares most about is not hers to give by will. If a will fails or a claim succeeds because one of those things was missed, the explanation that a tool produced the draft will not help anyone.

What it does well is remove hours of drafting, summarising and organising, which leaves more time for the conference and the note, the two things that decide whether a will holds up. For research, the difference between tools matters: a general chatbot answers from memory, while a tool built on Australian primary sources, such as Legal Brain, points you to the judgment or provision so you can read it yourself. Either way the rule is the same. If you have not read the source, do not cite it, and if you did not observe it, do not put it in the file note.

Frequently asked questions

Can AI write a valid will in Australia?

AI can produce text that reads like a will, but validity depends on meeting the formal requirements of the relevant State or Territory legislation, including proper execution and witnessing, and on the will-maker having capacity and acting freely. General AI tools often mix in wording from other jurisdictions and miss assets that sit outside the estate, such as superannuation and jointly held property. A solicitor should take the instructions, settle the document and supervise execution.

Can I use AI to write a testamentary capacity file note?

You can use AI to prepare a question plan based on the Banks v Goodfellow test and to organise your own notes under each limb afterwards. It must not add observations or conclusions you did not make yourself. Keep your raw notes on the file, check the structured version against them, and record the client's actual words.

Is it a breach of confidentiality to put client information into an AI tool?

It can be. Solicitors owe duties of confidentiality under the conduct rules, and entering client details into a tool that retains data or uses it for training may breach them. Before using any tool, confirm where data is stored, who can access it and whether it is used for training, and consider de-identifying information or obtaining client consent.

Can AI be used to prepare affidavits in probate or family provision proceedings?

Be very careful. The Supreme Court of New South Wales has a practice note, SC Gen 23, that restricts the use of generative AI in preparing affidavits and witness statements, and other Australian courts have issued their own guidelines. Check the current requirements of the court you are filing in, and make sure any affidavit reflects the deponent's own knowledge in their own words.