What to say when a client asks if AI drafted their advice
A commercial client reads your advice on Wednesday night and emails at 6am: 'Genuine question, did you write this or did you put it through ChatGPT?' There is no accusation in it. They are asking because they use the same tools in their own business and they know what those tools get wrong.
The question has become normal for three reasons. Australian courts, including the Supreme Court of NSW and the Supreme Court of Victoria, have published guidance on generative AI in litigation. The press has run story after story about fabricated citations turning up in filed material. And clients now have a rough sense of how fast a machine can produce something that looks like legal writing.
The worst response is a defensive one. Waffle reads as concealment, and a client who suspects concealment starts checking your citations themselves. Here is what the question is really asking, and what a straight answer sounds like.
The question behind the question
Almost no client cares which software sits on your desktop. They care about four things, and each one is hiding inside the question they actually asked.
Answer the surface question and you have said nothing useful. Answer the four underneath it and the conversation ends in about ninety seconds.
Answer it in one sentence, then give the detail
The structure that works is simple: direct answer, then scope, then responsibility, then verification, then stop. The stopping matters. Solicitors who keep explaining past the point of the client's satisfaction sound like they are talking themselves out of something.
Said out loud it runs about like this: 'Yes, I use research software. It helped me find the authorities and gave me a starting structure. Every case in that advice I have opened and read, the analysis is mine, and my name is on it, which means I carry it.'
Not all AI use sits the same way with a client
A client's tolerance is not a single setting. It tracks how far the machine's output travelled before a solicitor looked at it. Summarising a brief they already sent you is invisible and uncontroversial. Sending them a paragraph containing a case nobody has opened is the thing that ends retainers.
Be honest with yourself about which rung you are standing on before you get asked, because the answer sounds completely different depending on the rung.
Confidentiality is the part that can actually go wrong
Fabricated citations get the headlines, but they are recoverable: you check the source and the problem disappears before it leaves your office. Confidentiality is different, because the mistake completes itself the instant you hit paste and there is no undo.
Your duty to keep a client's affairs confidential does not soften because the recipient is software. If a client asks where their material goes, 'I am not sure' is a worse answer than 'I used AI'. Know what your tools retain, whether the content trains a model, and which jurisdiction the servers sit in, before a client makes you find out live.
Put it in writing before anyone asks
The firms handling this well are not the ones with a clever line. They are the ones where the answer already exists in a document, so every solicitor gives the same answer and nobody improvises.
It is a small amount of work: a clause, a list of approved tools, and a habit of noting verification on the file. Do it while it is a housekeeping task rather than a response to a complaint.
- Write down which tools are approved and which are not
- Write down what never gets pasted into an unapproved tool
- Tell the paralegals and junior solicitors, not just the partners
- One short clause on technology use, in plain English
- State that a solicitor reviews and remains responsible for all advice
- State that confidential material only goes to systems the firm has assessed
- Note on the file that authorities were checked against primary sources
- Check the current guidance of every court you appear in, because it is still moving
- Review one live file the way opposing counsel would
When the honest answer is yes, partly
Some clients will follow up with the real question: if the software did part of it, why is the bill the same? Answer that one head on too. If you charge by time, the time recorded should be the time spent, and efficiency belongs to the client. If you charge a fixed fee, they are buying an outcome and your professional risk, and the tooling behind it is your business, the same way your precedent bank always was.
The other thing worth saying: a general chatbot and a tool built for Australian practice are not the same conversation with a client. Something like Legal Brain, which answers from Australian legislation and case law and puts the source in front of you to open, is far easier to describe honestly, because the verification step is right there rather than something you promise you did later.
The client asking whether a robot wrote their advice is not hostile. They are checking that there is still a solicitor on the other end who has read the cases, formed a view and will stand behind it. That is an easy question to answer well, provided the answer is true.
Frequently asked questions
Do I have to tell my client if I used AI to draft their advice?
There is no blanket rule in Australia requiring you to disclose every piece of software you use, in the same way you do not disclose your precedent bank or your research database. What does not change is your duty to be competent, honest, confidential and accurate about costs. If a client asks directly, answer honestly, and check the practice notes of any court you are before, because several impose specific requirements in litigation.
What do Australian courts require about generative AI in filed documents?
Several Australian courts, including the Supreme Court of NSW and the Supreme Court of Victoria, have issued practice notes or guidance dealing with generative AI, with particular restrictions around affidavits, witness statements and expert material. The requirements differ between courts and have been revised more than once. Check the current version for the specific court and division before you file, rather than relying on what applied last year.
Can I put client documents into ChatGPT or a similar tool?
Not without knowing where the material goes and whether that is consistent with your confidentiality obligations. Consumer accounts and enterprise or business accounts have materially different retention and training settings. The practical test: could you tell the client today, in one sentence, which server holds their contract and who can access it? If not, do not paste it in.
How should I respond if a client says they are not paying for work a machine did?
Separate the bill from the tool. On a time basis, you record the time you actually spent, so any efficiency has already been passed to them. On a fixed fee, they are buying the advice and your responsibility for it, which is unchanged. Then point at the part that only a solicitor can do: applying the law to their specific facts and standing behind the result.
Two quick questions
No score is stored. Pick an answer to see why it is right.
-
1A client emails asking whether AI drafted their advice. What is the strongest opening response?
The direct answer removes any suspicion of concealment, and naming who is responsible addresses the actual worry, which is whether a solicitor's judgment still sits behind the document. Deflecting, minimising or interrogating the client all read as evasion.
-
2Which use of AI is hardest to defend if something goes wrong on the matter?
The other three all end with a solicitor checking the output against the source or the client's own material. An uncited-but-unread authority is unverified output reaching the client and potentially the court, which is the failure mode the practice notes are aimed at.
Research Australian law without handing over client data
Legal Brain searches Australian legislation and case law, shows you the source behind every answer, and anonymises client-identifying detail before anything reaches a model.