In February 2026 Adobe Express published a survey of 850 consumers and 205 marketers on AI-generated audio in branded content. Fifty-eight per cent of consumers said they’d trust a brand less if it used an AI voice in its advertising. Seventy-seven per cent said they trust human voices most. Then the context. Acceptance swung hard by it, with 51% comfortable hearing an AI voice in a video game and 45% in customer service, against 9% in news and 8% in podcasts.
The survey was run in the US, and I haven’t seen a UK equivalent with a sample that size, so treat the exact percentages as a direction of travel. The shape of it holds. The useful question isn’t whether your business should use an AI voice. It’s which jobs you’d hand to one, and the answer is narrower than most cost cases assume.
Table of Contents
Where synthetic voice earns its keep
Start with the honest half of the argument, because pretending synthetic voice has no place is a good way to lose the reader. The marketers in the same survey pointed to faster turnaround (49%) and cost savings (41%), and those are real. Scratch tracks for timing an edit before the proper recording. Temporary reads so a client can approve a script. Internal updates that nobody outside the building will ever hear. Prototypes. Long, low-stakes narration where the alternative is no audio at all.
The strongest case is content that changes constantly. An e-learning module that gets amended every time a policy or a price changes is painful to maintain with human recording, because altering one sentence means booking a session, matching the original performance and re-cutting the file. Synthetic voice handles that well. If the same module also opens with a message from your chief executive about why the training is compulsory, that part is a different decision.
Where it costs you
Advertising, brand films, founder stories, customer testimonials, anything whose whole job is to persuade somebody who doesn’t know you yet. The marketers gave the more useful answer here. Asked about their biggest challenge with AI audio, 48% named the lack of real emotion, ahead of legal risk and brand reputation. That’s the people buying the tools saying the tools don’t do the hard part.
The context figures should also change how you read your own content plan. Eight per cent of consumers found AI voices acceptable in podcasts. If you’ve spent two years building a branded podcast on the strength of the host’s voice, that number is a warning, and swapping in a synthetic read to save a day of studio time is a poor trade against the audience you’ve built.
Disclosure stopped being optional in August
Article 50 of the EU AI Act (Regulation (EU) 2024/1689) became applicable on 2 August 2026. Providers of systems that generate synthetic audio, images, video or text have to mark that output in a machine-readable way, with a deferral to 2 December 2026 for generative systems already on the market. Deployers have their own duty: anyone using AI to produce a deepfake, meaning content resembling real people, places or events that could pass as authentic, has to disclose that it’s artificial. Breaches sit in the AI Act’s general penalties framework, which reaches €15 million or 3% of worldwide annual turnover, whichever is higher. The obligations follow the audience, so a UK business whose content reaches EU users is caught.
The European Commission’s guidance is specific about what disclosure means in practice. It has to be clear and perceivable to an ordinary person on first exposure, which rules out relying on the invisible marking your supplier’s tool embeds in the file. A line in the small print at the end of a 30-second radio spot is unlikely to do it either.
Consumer expectation is running the same way anyway. Nearly three in four people in the Adobe survey said brands should disclose when AI-generated voice or music is used. That’s a large enough majority that disclosure is going to become normal whether or not the regulation reaches your market, so the planning question is how you say it without making the ad about the disclosure.
Whose voice is it
Cloning a real person’s voice raises a set of questions that a subscription doesn’t answer. You need written consent, a licence with a defined term, and a decision about what happens when that term ends. A cloned founder or staff voice can easily outlive the person’s employment, and the awkward conversation happens later, when they’ve left and the induction video is still speaking in their voice.
Library voices carry a quieter problem. A synthetic voice from a shared catalogue usually isn’t exclusive, which means a competitor can licence the same one. If you’re building brand recognition on how your company sounds, check the terms before you commit to the voice across a campaign, a phone system and a year of video.
Making the decision once, properly
This is a per-asset call, not a company-wide policy you set and forget. Somebody has to own the rules: which categories of content may use a synthetic voice, which may not, who signs off on rights and consent, and who writes the disclosure line when it’s needed. In most businesses that ownership is nobody’s, which is how a temp read intended for internal approval ends up in a paid campaign.
It’s easier to enforce when one supplier covers both sides of the choice. A voice over agency that casts and records human reads and also produces synthetic ones can hold the consent forms, the licence terms and the disclosure wording alongside the delivered files, so a year later, when somebody asks who approved a particular voice and on what basis, there’s an answer. Split the work between a marketplace freelancer, a subscription tool and whoever happened to be editing that week, and the paperwork lives in three places, which usually means none.
What to ask before you switch
Ask where the voice model came from and whether the people whose recordings trained it agreed to that use. Ask what rights you hold over the output, for how long, and across which media, because web and broadcast are rarely covered by the same licence. Ask whether the voice is exclusive to you. Ask whether the supplier marks the audio in the way Article 50 expects, and who is responsible for the on-screen or spoken disclosure when the content runs in the EU. If a supplier can’t answer those in writing, you’ve learned something useful for the price of an email.
One last thing about that 58% figure. Those consumers weren’t played clips and asked to spot the fakes, which is the test people usually imagine when they argue that synthetic voices have got good enough to pass. They were asked how they’d feel about a brand that used one. As disclosure spreads, that’s the question your customers will actually be answering.
