Can You Use An AI Voice Clone In Your Marketing?

You can use an AI voice clone in your marketing when the person whose voice it is has agreed in writing to that use. Copy a voice without consent, especially a well-known one, and you are relying on a part of Australian law that has barely been tested.

The technology is the easy part now. A few minutes of clean audio is enough for tools like ElevenLabs to produce a convincing copy, which is why the decision has moved from the production team to whoever signs off the brand.

What Australian law says

Australia has no standalone right in a person's voice. MinterEllison's review of the area, published in August 2025, describes a patchwork:

• Copyright can stop you reusing an existing recording, but it offers little against a new soundalike.
• The Australian Consumer Law bans misleading conduct and false representations under sections 18 and 29. An ad that makes listeners believe a person is speaking or endorsing the product is exposed.
• Passing off needs a reputation in the voice, a misrepresentation and damage, so it mainly helps singers, actors, podcasters and other people known for how they sound.

The closest Australian case is CBS Records v Telmak from 1987, about a soundalike album. It settled before a final judgment. Get legal advice before running a clone of anyone with a public profile, because this article is a marketing view and no court had settled the question when that review was written.

Tool and platform rules

The tools have their own terms. ElevenLabs' use policy bans cloning the voice of a living person without their "explicit and informed consent", so an unapproved clone also breaks the terms you agreed to when you signed up.

Platforms care too. YouTube requires creators to disclose realistic altered content, and synthetically generated voices are one of its examples. Set that option at upload, and treat TikTok's AI labelling the same way.

Where a clone earns its place

The strongest uses keep a real person in control of a real message:

• Fixing one line in a founder's recorded voiceover when a product name or a date changes, without booking another session.
• Narrating a long run of product explainers or training modules that the person has read and approved.
• A properly licensed synthetic voice for explainers, where no real person is being imitated at all.

Where a real person is involved, they have seen the script and would happily say those words in a meeting.

Where to leave it out

Never clone a customer, a client or a reviewer. A testimonial in a voice the person did not record is an invented testimonial, whatever the words say.

Stay away from staff who have left the business, unless their consent clearly covers that. Keep it out of emotional or personal founder content as well. A viewer who trusts a founder's story and later learns the voice was generated will doubt everything else that founder has said.

The consent document

Write one page and have the person sign it before the model is built. It covers:

• Whose voice it is, and which recordings can train the model.
• Which tool holds the voice, and who has the login.
• What the voice can be used for, on which channels, and for how long.
• Who approves each script.
• Any payment, plus what happens when the person leaves the business.
• How and when the model gets deleted.

A voice actor will price this as a licence, with a term and a list of uses, and that is the right way to treat an employee's voice too. Limit the login as well. Anyone with access can make your managing director say anything, including an instruction to approve a payment.

Whether to tell the audience

MinterEllison's review treats disclosure as an open question, so set your own rule. If the voice is presented as a person speaking directly to the buyer, say so in the description. If it is narration over a product demonstration that the person approved, the platform label is enough.

A founder's voice takes years to earn trust. Treat a clone of it the way you would treat the company seal.