Australian professional firms are quietly folding AI into everyday work: drafting file notes, summarising long contracts, sorting inbound queries, preparing first-pass advice. Most of that happens behind the scenes, and clients never see it. The awkward question every accountant, lawyer, broker and consultant now faces is simple to ask and hard to answer well. Do you tell the client you used AI, and if so, when and how?
There is no single Australian rule that says "disclose all AI use". Instead there is a patchwork of professional obligations, the Privacy Act, client contracts and plain commercial trust. This piece sets out a practical way to decide, and gives you wording you can adopt this quarter. We use Claude for this kind of work every day, so the guidance here comes from real client engagements, not theory.
When disclosure actually matters
Disclosure is not a binary switch you flip for every email. The useful test is whether a reasonable client would feel misled if they later found out. If the answer is yes, say something up front. A few situations clearly cross that line for Australian firms:
The AI output goes directly to the client as a deliverable, such as a drafted letter, a report, or advice they will rely on and pay for.
Client data leaves your controlled environment and is processed by a tool, which engages your Privacy Act obligations around how personal information is handled and disclosed.
You are in a regulated setting where APRA, ASIC or a professional body expects a clear record of who, or what, produced the work.
The engagement letter or tender promised human expertise as the core of what the client is buying.
By contrast, using Claude to tidy your own internal notes, brainstorm an approach, or check your grammar rarely needs a formal notice. The client is buying your judgment, and the tool is closer to a calculator than a co-author. The risk is not that you used software. The risk is silence in the cases where the client would reasonably expect to know.
How to word it without over-promising or scaring people
The most common mistake is treating disclosure as a legal warning. Clients do not want a wall of disclaimers. They want to know their work is accurate, their data is safe, and a qualified human stands behind the result. Good disclosure is short, calm and specific.
Three sentences usually do the job. Name the tool category, say what a human still owns, and point to how you protect their information. For example: "We use AI assistants, including Claude, to help prepare and review parts of your work. A qualified member of our team checks and signs off everything before it reaches you. Your information is handled under our privacy policy and is not used to train public models." That last clause matters in Australia, where clients increasingly ask where their data goes.
Keep these principles in mind when you draft your standard wording:
Be truthful about the human sign-off. Never claim review you do not actually perform.
Describe the tool by role, not brand hype. "AI assistant that drafts and checks" is clearer than a marketing slogan.
Address data handling directly, because that is the client's real worry, not the technology itself.
Match the channel to the stakes: a line in your engagement letter for routine work, a direct conversation for high-value or sensitive matters.
Whether you always need to say something
Blanket disclosure on every interaction trains clients to ignore it, the same way nobody reads a cookie banner. A tiered approach works better. Put a standing statement in your engagement letter and privacy policy so the baseline is covered. Then disclose actively, in the moment, only when the work crosses the thresholds above.
Consider the commercial maths. A mid-sized Sydney advisory firm we work with was spending roughly $120K a year of senior time on first-draft documents. Moving that first pass to Claude, with human review retained, freed about a third of it. They chose to disclose the change to clients because the alternative, a client discovering it through a stray reference, would have cost far more than $45,000 in lost trust and rework. Transparency was cheaper than a repair job.
There is also a reputational upside. Australian clients are already assuming firms use AI. A firm that explains its approach clearly looks more in control than one that stays quiet and hopes the question never comes up.
A practical policy you can adopt this quarter
You do not need a committee to get this right. A one-page internal policy plus two lines of client-facing wording covers most firms. Here is a starting structure:
A standing disclosure clause in your engagement letter and website privacy page, covering routine AI-assisted work.
A short list of trigger situations where a team member must disclose actively before delivery.
A fixed rule that a named human reviews and approves any AI-assisted output before it reaches a client.
A data-handling note confirming client information is not fed into public training and stays within approved tools.
Set it once, brief your team, and revisit it every six months as tools and expectations change. The goal is a firm where any staff member can answer "did you use AI on this?" without hesitation, because the policy already told the client the honest answer.
If you would like help drafting disclosure wording and an AI-use policy that fits your obligations and your clients, book a brainstorm with us. We will map where disclosure genuinely matters for your firm and where it just adds noise.
