Will AI Replace Voice Actors? A View From the Person Hiring
Last updated: September 16, 2026
For some work, it already has. For the work where a voice has to carry meaning, it has not, and I do not expect it to soon. AI voices have taken most of the jobs where the brief was "read this clearly". They have not taken the jobs where the brief was "make people feel this". The real change for voice actors is less about replacement and more about consent. A voice can now be copied from a few minutes of audio, and UK law has not caught up with that.
I have booked voice artists for films for start-ups and for billion-dollar brands, so I sit on the side of the table that decides whose voice goes on a project. I understand why AI is attractive. Most people reaching for generated voices, and generated video, are trying to make content on a smaller budget, and that pressure is real. But I am a filmmaker whose work relies on filming real people, so I am not a neutral observer either. We are all part of the same machine now, and this piece is written from inside it.
A recent job made this concrete for me. I booked a voice artist through one of the more reputable voiceover sites. When the client heard the read, their first reaction was not whether they liked it. It was to ask, "Is this an AI voiceover?"
As far as I was aware, it was a real person. But the question landed twice. First, that a client's starting assumption is now that a polished voice might be synthetic, which is a strange place for any working artist to be. Second, and more uncomfortably, I realised I could not prove it either way. I had trusted the platform and the profile. If a seller had quietly used a synthetic voice, would I have known? I am honestly not sure I would.
You are probably here for one of four reasons. You voice for a living and want a straight read on what is changing. You commission video and want to know when an AI voice is a sensible saving and when it is a false one. Your employer has asked you, or you are about to ask a colleague, to voice something for the company's channels. Or you have just heard a voice that sounded suspiciously like someone real, and you want to know what the rules are. All four are answered below, followed by where I think this is all heading.
Which voiceover work is AI already taking?
The work going first is the work where nobody was listening closely. Where the voice was functional rather than performed. That has been true of every wave of automation in this industry, and it is true again.
| Type of voice work | Where it is heading | Why |
|---|---|---|
| Phone menus, on-hold, system prompts | Largely AI already | Pure information, heard once, updated often. Nobody is buying a performance. |
| Internal training and e-learning | Mostly AI, for now | Scripts change constantly and budgets are thin. The quality gap shows on long modules, where a flat read loses people. |
| Explainers and social cut-downs | Split | Fine for volume content. The brands that care how they sound are coming back to human voices. |
| Dubbing and localisation | Most exposed of the skilled work | Timing and translation can now be automated together. Human work is moving towards reviewing and directing. |
| Placeholder and scratch voice | AI, and that is fine | It was never paid work for most artists. It speeds up the edit before the real session. |
| Brand films, adverts, documentary narration | Human | The voice is the brand. Direction, interpretation and distinctiveness are the product. |
| Character, animation and games | Human, with licensed AI for pickups | Performance defines the character. AI is being used for small line changes, with the actor's consent. |
The dubbing row deserves a number. A global study for CISAC, the international body for authors' societies, estimated that translators and adaptors for dubbing and subtitling have 56% of their revenue at risk by 2028, and audiovisual creators as a whole 21%. Those are projections, not counts, but they point at the same place my own work does. The exposure is highest where a voice is matching something that already exists, and lowest where it is creating something that does not.
Where does a human voice still win?
A voiceover is not someone reading words aloud. It is someone deciding what the words mean. Where to lean in, where to pull back, what to leave unsaid, how much warmth a line can take before it tips into selling. On a directed session, a good artist gives me a read I did not know I wanted, and then gives me three more when I ask. That exchange is where most of the value sits, and it is the part AI tools still handle worst. You can prompt a synthetic voice towards an emotion. You cannot have a conversation with it about why the line matters.
The second thing is distinctiveness. AI voices are competent and they are converging. Spend a week listening to them and they start to sound like the same person in slightly different moods. For a brand, that is a real cost. How a company sounds is part of how it is remembered, which is the whole argument of auditory branding. A voice that sounds like everyone else's does the opposite job.
The third is trust. When an audience realises a voice was generated and nobody told them, the problem is rarely the voice. It is the feeling of having been handled. That is a disclosure problem more than a quality problem, and it is solvable, but only if someone decides to solve it.
None of this means AI voices have no place. They are useful as scratch tracks and rough timing guides, and in openly synthetic pieces where nobody is pretending otherwise. The line I hold is simple. If the voice is doing the persuading or the storytelling, it should be a person, or a licensed and disclosed version of one. It is the same judgement that sits behind choosing between human and synthetic in filmmaking more broadly, and voice is simply where most audiences hear the difference first.
What if a voiceover sold as human is actually AI?
That job left me with a question I think the industry has barely started asking. There are two problems hiding in it.
The first is unfair on artists. Real performers are now being suspected of using AI, simply because synthetic voices have become good enough that a clean, confident read can raise doubts. That is a reputational cost for people who have done nothing wrong.
The second is the one that crosses a line. If someone sells a voiceover as a human performance and delivers a synthetic or converted voice instead, the buyer has not got what they paid for. I am not a lawyer, but that looks like a misrepresentation of the service, and it is likely to break the platform's own terms. It also passes the risk down the chain. The commissioner, and then their client, may publish something as a real performance when it is not, with no disclosure where disclosure is needed. And if the synthetic voice was cloned from someone else, the buyer may be using a real person's voice without their permission and without knowing it.
How I would protect a commission now.
Book a live directed session where possible, even a remote one, so you hear the performance happen rather than receive a file.
Ask for a short custom test read with a specific change, such as a line read two ways, which is harder to fake convincingly on the spot.
Put it in writing. The recording is a live performance by the named artist, with no synthetic voice, voice cloning or voice conversion unless agreed.
Check what the platform's terms say about AI, and keep the session files and correspondence.
Tell your client what you booked and how, so they are not the ones left guessing.
If you are the artist, the same thing works in reverse. Offer live sessions, keep your raw takes, and state your position on AI in writing. Being able to show your work is becoming part of what you sell.
Can someone clone your voice without permission in the UK?
This is the part most articles skip, and it matters more than the quality debate.
UK law protects recordings of a performance. It does not protect the voice itself. So a synthetic voice trained to sound like you, without copying any specific recording, falls into a gap. Lawyers point to passing off, trade marks, defamation and data protection as partial routes, and all of them are patchy and mostly work for people who are already famous.
The government knows. Its March 2026 report on copyright and AI accepted that existing law may not deal with the harms of digital replicas, and promised a consultation on whether the UK needs a new personality or digital replica right, along with a taskforce on labelling AI content. At the time of writing, that consultation has not opened and no law has changed. Until it does, the practical position is blunt. In the UK, prevention beats remedy. Consent agreed in writing before a recording is made protects you far better than anything you can do after a clone appears.
Performers are pushing from the other side. In December 2025, Equity's film and TV members voted 99.6% in an indicative ballot that they were prepared to refuse digital scanning on set to secure AI protections, and the union has been negotiating with the producers' body Pact ever since. Their American counterpart SAG-AFTRA already has AI consent and payment terms written into its 2026 studio agreement, which is part of why UK performers feel exposed by comparison.
Two rules apply right now, whatever Parliament eventually does. If work reaches audiences in the EU, synthetic audio that could pass as real has to be disclosed under the AI Act, in force since August 2026. And in UK advertising, the codes apply however an ad was made, so an AI voice that misleads is a breach whether or not it carries a label. I set out how I handle both in my AI policy.
What does the licensed route look like?
There is a version of AI voice work that is fair to performers, and it is starting to look like a real market. In November 2025 ElevenLabs launched a marketplace for licensed voices, with Sir Michael Caine among the first living actors to license his. Games studios are using licensed versions of an actor's voice for small pickups, a changed item name or a late menu line, instead of booking a full session.
What makes that version acceptable is not the technology. It is four things agreed in writing before anything is made. What exactly is being used. What it can be used for and where. How long the permission lasts. How the performer is paid each time it is used. Take any of the four away and you are back in the gap described above. It is the same logic that runs through actors becoming royalty-earning IP.
If you commission voiceover, what should you ask?
Five questions, before you sign off a script, whether you are booking a person or a tool.
Is this voice real, synthetic, or a licensed copy of a real person? If it is a copy, who consented, to what, and for how long? Does our contract with the artist say anything about AI, and if not, why not? Where will the audience learn that a voice was generated, if it was? And if someone asks in a year how this was made, what do we have on file?
One more thing that catches people out. Permission to record someone is not permission to synthesise them. A standard voiceover agreement licenses the recording you made. It does not license a clone of that voice for next year's campaign. That needs its own consent, specific to the use. If you want wording for it, there is sample consent language in my AI policy.
What happens when an employee lends their voice to company content?
This is the version of the problem I see growing fastest, and it rarely involves a professional at all. A marketing team needs a voice for a social video. Someone in the office has a good one. They say yes, record it on a Tuesday afternoon, and the video goes out. Nobody writes anything down, because it feels like a favour between colleagues.
Then the recording is public. And public audio is raw material. A few minutes of clear speech is enough to build a convincing clone, and nobody needs to ask permission to try. This is not hypothetical. Fraudsters targeting WPP built a voice clone of its chief executive, Mark Read, using public YouTube footage and used it in a video meeting to try to get money and personal details from a senior colleague. It failed, but the method only needed what was already online.
I am not a lawyer, and anyone in this position should take proper advice. But these are the risks I would want both sides to understand before the microphone goes on.
For the business. A voice recording of an identifiable person is personal data. Consent is the obvious basis to rely on, and it is the weakest one in employment, because the UK data regulator names the employer and employee relationship as a clear example of a power imbalance. If staff feel they cannot say no, their yes may not count, and they can withdraw it later. A scripted read may also count as a performance in law, with rights of its own that do not automatically pass to the employer the way copyright in office work usually does. Then there is the practical exposure. The employee leaves, and their voice is still selling your product. The employee's voice gets cloned, and a fake of them calls your finance team. Or a vendor's tool keeps the raw recording and trains on it.
For the individual. Your voice is not like a photo you can take down. Once it is out, it can be copied, and in the UK there is currently no general legal right over your voice to stop an imitation. So the time to set terms is before you record, not after. Ask where the content will run, for how long, whether it can be edited or re-voiced, whether it can ever be used to create a synthetic version of you, and what happens when you leave. It is also fair to ask whether you will be paid if the piece becomes a paid advert rather than an organic post. If you are in a union, your rep will have seen this before.
What a sensible business puts in place.
A short written agreement for each staff voice, separate from the employment contract, so it is clearly optional.
A genuine choice, with no consequence for saying no, and a professional voice artist as the default alternative.
Stated scope. Which pieces, which channels, how long, and what happens to the content when the person leaves.
No cloning and no model training unless separately agreed in writing, with the default set to no.
Raw recordings stored securely and kept out of tools whose terms allow them to be retained or trained on.
A verification rule for any urgent request that arrives by voice or video, such as calling the person back on a known number, because every public recording of a senior leader makes impersonation easier.
None of this is heavy. It is a page of paper and a ten-minute conversation, and it is the same principle I apply on every shoot. Permission to record someone is not permission to synthesise them.
Where is AI voice heading, and what does it mean for traditional artists?
This is my view rather than a forecast, and it is shaped by what I see across video, sound and image work, not only voice.
The market is splitting into three. At the bottom, a large volume of synthetic voice that nobody will ever credit, doing the jobs that were always about information. At the top, human performance that is booked because it is human, with a named artist, a directed session and a voice nobody else has. In the middle, a growing licensed layer, where a real performer's voice is used under agreed terms for pickups, translations and updates, and they are paid each time. The middle is where most of the argument over the next few years will happen.
How the voiceover market is splitting
The top
Human performance
Booked because it is human. A named artist, a directed session, a voice nobody else has.
The middle
Licensed middle
A real performer's voice used under agreed terms for pickups and updates, and paid each time.
The bottom
Synthetic volume
Functional work nobody will ever credit. Phone menus, system prompts, throwaway reads.
nigelcamp.com
Human work becomes a signal. Recorded music did not end live music. It made the live version something people chose and paid more for. I think human performance is heading the same way. When synthetic is the default, a real voice, a real hand or a real lens says that someone thought the work was worth it. Brands will start saying so out loud, and some already are.
Provenance becomes infrastructure. The big platforms are building ways to show how content was made. OpenAI joined the C2PA content credentials steering committee and adopted Google's SynthID watermarking in May 2026, and Google has started surfacing that information in Search and Chrome. The UK government has committed to a taskforce on labelling AI content, with an interim report due this autumn. For artists, that shift matters. The ability to prove what you made, and what was made from you, is turning into part of the job.
The voice becomes an asset, not just a service. Performers who document their work, understand their contracts and decide on their own terms whether to license a synthetic version are in a much stronger position than those who find out after it has been copied. That applies to illustrators, photographers and musicians too. The value is moving towards authorship, taste and control.
The hardest problem is the first rung. The small, repetitive jobs AI is taking first were where most artists learned the craft and paid their rent while they did. If those disappear, the question is not whether top-end talent survives. It is where the next generation of it comes from. I do not think anyone has a good answer yet, and the industry, me included, should be spending more time on it.
If you are a voice artist, where is the work going?
Honestly, some of it is not coming back. The high-volume, low-stakes jobs that paid the bills between bigger bookings are the ones AI has taken first, and that hurts most for artists early in their careers.
What is holding up is work where the client is buying judgement. Brand voice, narration, character, anything directed live. Artists who can take direction quickly, offer options, and talk to a producer about why a read works are in a stronger position than they were, because the gap between them and a synthetic voice is audible. So is anyone who treats their voice as something to license on their own terms rather than something to protect only after it has been copied. Read your contracts for AI clauses, keep a record of your own work, and join Equity if you are in the UK. Collective agreements are where the real protections are being won.
Questions people ask about AI and voice acting
Will AI replace voice actors?
For functional work like phone menus, internal training and placeholder tracks, it largely has. For work where the voice carries meaning, brand films, narration, character and adverts, human voice actors are still the norm, because clients are paying for interpretation and direction, not just sound.
Can AI copy someone's voice?
Yes, and it needs surprisingly little audio. Whether it is allowed is a separate question. In the UK there is no single right that protects your voice itself, only a patchwork such as protection for recordings of your performances, plus passing off and data protection in some cases, and the government has promised a consultation on whether to change that.
Is it legal to use an AI voice in an advert in the UK?
Yes, if it does not mislead. The UK advertising codes apply however an ad was made. Using a voice that imitates a real person without permission, or presenting a synthetic voice as a real customer or expert, can mislead and can also create legal risk under passing off or data protection.
Do I have to disclose an AI voice?
If your audience includes people in the EU and the voice could pass as a real person, yes, under Article 50 of the AI Act, which has applied since August 2026. In the UK there is no single labelling law yet, but the test I use is whether a reasonable listener would assume a real person was speaking. If they would, say so.
Does a voiceover contract cover AI cloning?
Not unless it says so. A standard agreement licenses the recording you made. Cloning that voice, or using it to train a model, needs separate consent in writing, covering what is used, where, for how long, and how the artist is paid.
Can my employer use my voice in AI content?
Only with care, and ideally with a separate written agreement. Your voice recording is personal data, and because of the power imbalance at work, UK data protection guidance treats employee consent as hard to rely on. Before recording, agree where it will be used, for how long, what happens when you leave, and that it will not be cloned or used to train a model without your separate written agreement.
What happens to my voice if I leave the company?
It depends on what was agreed, which is why it should be agreed in writing before you record. Without terms, content featuring your voice may stay live long after you have gone. You can ask for it to be reviewed or removed, and you can withdraw consent where consent was the basis, but content already shared publicly cannot always be pulled back.
What will AI mean for traditional artists?
The volume work is going synthetic, human performance is becoming something clients choose deliberately, and a licensed middle ground is growing where artists are paid for approved uses of their voice or style. The biggest risk is to new entrants, because the small jobs where people learned the craft are the first to go.
How do I know a voiceover I paid for is really human?
You often cannot tell by ear any more. The practical checks are a live directed session, a short custom test read with a specific change, and a written promise that the recording is a live performance by the named artist with no synthetic voice or voice conversion. If a voice sold as human turns out to be synthetic, that is likely to be a misrepresentation and a breach of most platforms' terms.
Where are AI voices genuinely useful?
Scratch tracks during editing, timing guides, accessibility tools, fast updates to information-only content, and openly synthetic projects. Also licensed pickups, where an actor has agreed that a copy of their voice can be used for small line changes.
