Client Disclosure for AI-Assisted Design

There’s no rule that says you must tell a client you used AI. There is a rule that says you can’t mislead them. Knowing the difference is the skill.

Copyright and Licensing opens on a freelancer at 11pm, deck due Monday, asking whether she can commercially use an AI-generated image. There’s a second question waiting right behind it, same Friday night: do I have to tell the client I used AI at all?

Here’s the honest answer. There is no universal requirement to disclose AI tool use to a client. There is a universal requirement, under Australian Consumer Law, not to mislead them. So the test isn’t “did I use AI?” It’s “would this client reasonably expect to know how the work was made, and does leaving it out change a decision they’re making?”

Designers who set that expectation early, at the brief, never have to navigate the awkward version of this conversation later. Designers who don’t, eventually do. This is the practical sibling to Copyright and Licensing: that asks what the client can own, this asks what they need to know. This is professional-practice guidance for designers, not legal advice.

Designers tend to meet AI uncertainty in one of two unhelpful ways: avoid the tools, or use them and hope nobody asks. The professional-practice frame is neither.

Treat an AI-generated asset the way you already treat stock photography. Know the licence. Document the source. Tell the client when it changes their risk. You’ve done this for decades with stock and font foundries; AI adds new variables to a familiar habit. The Graphic Artists Guild, whose Handbook of Pricing & Ethical Guidelines (17th edition, 2025) now covers generative AI, takes the same line in its published ethical-use guidance: tell clients when AI is part of the creative process. Routine, not confession.

One caveat, same as the copyright article: for genuinely high-stakes work (a bet-the-brand campaign, public-figure likenesses, rights-managed franchise material), get a lawyer. For everyday client work, the habit below is enough.

Run every disclosure decision through one diagnostic: would the client reasonably expect to know how this was made, and does the omission matter to their decision?

• Yes to both, disclose.
• Yes to one, judgment call; lean toward disclosing.
• No to both, no disclosure obligation.

The clearest case sits at the top. A client commissions an original illustration, pays for a designer’s hand on the artwork, and receives an AI generation styled to look hand-drawn. That fails both halves: they expected something specific, and the substitution changed what they bought. That’s misrepresentation. Now the other end: you use AI to spin up a dozen rough concept directions, pick one, and develop the finished work yourself. The client expected a finished design from a designer; the route you took to get there isn’t material to the deliverable. That’s a workflow choice, not a disclosure event.

Misrepresentation test (the two-part check, reasonable expectation plus materiality, that decides whether disclosure is required). Material omission (a fact the client would treat as important to their decision; under the ACL, leaving one out can itself be misleading).

Both failure modes are real. “Disclose nothing unless asked” walks straight into the commissioned-illustration trap. “Disclose everything, always” sounds safe but turns the disclosure into noise: when every project carries the same boilerplate, clients stop reading it, and the one that mattered slips past. The test is judgment, not a blanket rule.

Disclosure isn’t a delivery-time event. It’s a brief-stage habit with two fallbacks.

Brief stage (preferred). Before scope is locked: “My workflow uses AI tools at these stages. I’ll include a full disclosure with the deliverables; if that raises anything, now’s the time.” The client knows upfront and can opt in, opt out, or renegotiate. No surprise at handoff.

Revision stage (acceptable). When the brief evolves mid-project and AI becomes part of the scope: “In this round I used AI for this specific purpose; the deliverables will carry the detail. Flagging now in case it affects how you want to scope the next phase.” Less clean, still proactive, client keeps agency over what’s downstream.

Delivery stage (last resort). At handoff: “Here’s the work, and here’s the AI-asset disclosure attached. If anything in it changes how you want to handle usage or attribution, tell me.” This works least well: the client is already committing to the deliverable, so the disclosure reads like a footnote, and the only moves left are accept or object.

The cost of disclosing at the brief is one extra paragraph of discussion. The cost of disclosing at delivery is, occasionally, a renegotiation, and rarely a relationship.

Production-ready. Copy it into your project record and fill in the placeholders.

> **AI-Asset Disclosure, Project [Name]**
>
> For the [project name] deliverable, the following assets used AI tools:
>
> 1. Tool used: [tool name + version]
> 2. Subscription tier: [Free / Pro / Enterprise]
> 3. Commercial-use clearance: [Yes, under tool terms / Conditional, see notes / N/A]
> 4. Indemnification: [Yes, covered by tool / No, designer carries risk; see Copyright and Licensing]
> 5. Asset usage: [hero / supporting graphic / reference / development input]
> 6. Reverse-image-search performed (recommended for commercial work): [Yes, no issues found / No, see notes]
> 7. Jurisdiction of use: [AU only / AU + EU / AU + US / Global]
> 8. Notes: [client-specific risk flags]
>
> This disclosure forms part of the project record. The designer warrants reasonable
> diligence; the client retains final responsibility for usage decisions.

Each row earns its place. Tool and tier decide what’s commercially usable and what the client can own. Indemnification decides who carries the risk if a third party objects; the tool-by-tool detail (including which tools indemnify paying users) lives in Copyright and Licensing, not here. The reverse-image-search line records your diligence against inadvertent-replication claims. Jurisdiction flags whether obligations like EU AI Act Article 50 transparency apply (again, see Copyright and Licensing).

Brief-stage, a logo project. You’re pitching a small business on a logo and you’ll use AI to generate early visual directions before drawing the mark yourself. At the brief: “I’ll use AI tools to explore initial directions quickly, then design the final mark by hand. The delivery pack will include an AI-asset disclosure.” Run the test: the client would reasonably expect to know, and at brief stage it costs nothing to say so. Disclose early, done.

Delivery-stage, a brand identity. A larger identity package where AI generated some initial moodboard references, all developed and finalised by you. At handoff the disclosure names the tools, marks those assets as “development input,” and confirms the final artwork is your own. The test still passes, the client would want to know, but because the AI never reached the deliverable, the conversation is brief and the work stands on its own.

The statutory anchor is section 18 of the Australian Consumer Law (Schedule 2 of the Competition and Consumer Act 2010): you can’t engage in misleading or deceptive conduct in trade or commerce. And silence can mislead, too: not because there’s a blanket duty to disclose everything, but where the client would reasonably expect to be told and isn’t. Australia’s regulators have been clear this reaches AI like anything else: a 2025 Treasury review of AI and the Australian Consumer Law found the existing misleading-conduct rules apply regardless of the technology involved, and the ACCC has named “AI-washing” (overstating what AI did or can do) an enforcement priority.

The professional codes point the same way. AIGA’s standards predate generative AI, so read them as orienting principles, not rules written for this. Two carry over cleanly. §6.2: a designer “shall communicate the truth in all situations and at all times; his or her work shall not make false claims nor knowingly misinform.” And §1.3, the one most freelancers miss: a designer shall treat “all knowledge of a client’s intentions, production methods and business organization as confidential.” Pasting a client’s brief, files, or unreleased product into an AI tool is itself a confidentiality act, know your tool’s data-retention terms before you do it.

Two privacy notes, as flags rather than a panic. Uploading client materials ties straight back to §1.3 above. And from 10 December 2026 the Privacy Act will require covered organisations to disclose significant automated decision-making in their privacy policies; whether that reaches a small studio depends on your circumstances, and the long-standing small-business exemption is itself under reform, so check the OAIC rather than assume you’re outside it.

When an Air Canada chatbot gave a customer wrong information, the airline argued the bot was “a separate legal entity” responsible for its own words. A tribunal disagreed and held the company liable (Moffatt v Air Canada, 2024). The lesson travels: the professional who put the tool in front of the client owns what it produces. You can’t subcontract responsibility to a model. Disclosure is how you keep that responsibility legible, to the client and to yourself.

Knowing what to declare, when, and how is professional judgment, the same muscle as crediting a photographer or scoping a contract. It isn’t an AI problem bolted onto design practice; it’s design practice meeting a new variable. The designers who handle it well aren’t the most cautious or the most casual. They’re the ones who made it routine.

TGDS Verdict. Disclosure done early is a one-paragraph habit; disclosure done late is a renegotiation. Build the habit.

Design@Work is the applied route for working freelancers turning this into standard practice. The Graphic Design School has run on professional-practice fundamentals for 18 years (850+ graduates, RTO #91706, zero ASQA complaints), and that grounding is exactly what tells you which conversations to have before the work ships.

Outbound: When AI Gets It Wrong (what to watch for in the output you’re disclosing), Brief Writing with AI (where the brief-stage disclosure habit actually lives), Copyright and Licensing (the canonical home for tool indemnity and the four-factor IP matrix).

Tell the client what they’d reasonably want to know. Do it at the brief. The awkward conversation only happens to the people who wait.