AI Voice Over for Videos: Can a Production Company Use It for Clients?

Yes, a video production company can use AI voice over for client videos, provided the voice is a licensed stock voice or the cloned voice belongs to someone who has given written, verified consent, the client agrees in writing, and platform labels are applied where required. Never clone or imitate a voice without the owner's own permission.

The tools make the technical part trivial, which is why the rules are worth settling before the first project. The risk to a studio is rarely the audio quality. It is a contract that never mentioned AI, a cloned voice nobody authorised properly, and a published video that breaks a platform's disclosure rule.

Follow me on Instagram@sagnikteaches

Stock voice or cloned voice: choose before you quote

Most AI voice work for clients falls into one of two types, and they carry different obligations.

Connect on LinkedInSagnik Bhattacharya

Stock voices

A stock voice is one of the tool's built-in or library voices. It does not belong to your client or to anyone you hired for the project. Your questions are about licence terms (commercial use, redistribution, whether you can use the output in client work and in paid advertising), about whether the voice might be mistaken for a real, identifiable person, and about disclosure. For many explainer, training and social videos this is the simplest and lowest-risk choice. Check the plan as well as the voice: on ElevenLabs the commercial licence starts with the paid Starter plan, and the free plan does not include it, so a client video voiced on a free account is a licensing problem before it is anything else.

Subscribe on YouTube@sagnikteaches

Cloned voices

A cloned voice reproduces a specific person. Clients ask for it so the founder "narrates" every video without recording time, or so a corrected line can be patched in without a re-record. This is where consent matters. ElevenLabs' documentation says a professional voice clone needs its Creator plan or above, requires a voice verification step before the fine-tuning request, and can only be made of your own voice: it states that even with someone's consent you cannot clone their voice on your account. The route it describes is for the person to create and verify the clone on their own account and share it with you through a private sharing link. Instant voice cloning, available from the Starter plan and built from a minute or two of audio, works differently: the app asks you to confirm that you have the right and consent to clone the voice. For client work, that tick box should rest on the signed form below, never on a conversation. Read the current pages before you promise a client either workflow, since plans and rules change.

Where a production company holds the account and the client's founder holds a clone on their own account, the founder keeps control, and you have a technical and a paper trail showing consent. That is a better position than recordings sitting on your server.

The written consent a cloned voice needs

A verbal "yes, that's fine" is not consent you can rely on. The agreement should be specific about scope, because broad open-ended permission is the thing people later regret. Cover at least these points, and have a qualified adviser review your own template.

• Who: the person whose voice it is, by name, and the company they speak for.
• What: creating a synthetic copy of the voice, and generating audio from it.
• Where and what for: named projects, channels and kinds of content.
• What not: no political, adult, medical, financial-advice or endorsement content, and no words the speaker has not approved.
• Approval: the speaker approves each script or a defined script category before generation.
• Duration: start and end dates, and what happens to existing videos at the end.
• Withdrawal: how the speaker can withdraw for future use, with notice.
• Deletion: who deletes the clone and training audio, by when.
• Payment: whether the speaker is paid, and whether fees are due on reuse.

Here is an illustrative short form for a founder's voice, to show the level of detail.

Voice consent. I, [name], of [company], consent to [studio] creating
a synthetic copy of my voice using a tool I verify myself, and to
generating audio with it for the videos named in Schedule 1 for
[client]. I will approve each script before audio is produced. The
copy will not be used for any other purpose, for any statement I
have not approved, or after [end date]. I may withdraw consent for
future use by giving [30] days' written notice. On withdrawal or at
the end date, the copy and any training audio will be deleted and the
studio will confirm in writing.

Sign it before recording the training audio, not after. A clone built first and papered later is the sequence that causes disputes.

Illustration: a six-minute training video for an HVAC contractor

Say an HVAC contractor's owner wants a six-minute safety-induction video for new technicians, narrated in her own voice, with updates twice a year as procedures change. Recording each update means a studio session every time. The production company proposes an AI voice built from her recorded voice, created and verified by her on her own account and shared with the studio.

The steps run like this. First, the proposal states the voice approach and the owner signs the consent form above. Second, she records the training audio. ElevenLabs recommends 30 to 180 minutes of good audio for a professional clone, recorded cleanly with no background noise, room echo or second speaker, so the studio books half a day for it rather than twenty minutes. Third, she completes the verification and shares the voice. ElevenLabs says fine-tuning usually takes three to six hours after verification and can take up to 24, so the studio does not promise same-day narration. Fourth, the studio generates the narration from the approved script. Fifth, she listens to the first full version and signs off on the script and audio before publication. Sixth, the studio records in the project file the date, the script version and the approval.

When procedures change in six months, the studio updates two lines of script, generates the new audio, and sends it for approval. The saving is a repeat studio session; the cost is the half day at the start and the discipline of approvals. If the owner leaves the company, the consent form already says what happens next.

What to put in your contract with the client

Most production contracts were written before AI voices, and silence is the usual source of trouble. Add four short clauses.

1. Permission to use AI voices: the client agrees that AI-generated voice may be used for the named deliverables, and whether stock or cloned.
2. Responsibility for consent: if the client supplies a voice (its own staff, a presenter), the client warrants that it has the person's written consent, and you will ask to see it.
3. Disclosure and labels: the client agrees that AI voice will be disclosed where a platform or law requires it, and that you may add labels.
4. Approval: the client approves script and audio before publication, and it is on the client once approved. If the client edits the script after approval, you regenerate from the new version.

You can fold these into the proposal. Our tutorial on AI-assisted video production proposals shows where scope and revisions sit, and pricing AI services a small studio can add shows how to charge for AI voice work so that the saving is shared fairly rather than silently absorbed.

Platform labels and disclosure

Disclosure is a separate duty from consent. Three layers apply.

To the client. The contract and proposal tell them. They should never discover the voice is synthetic from the invoice.

On the platform. YouTube asks creators to disclose realistic altered or synthetic content, including content that makes a real person appear to say or do something they did not. Its guidance lists exceptions, including minor edits, production help such as AI-drafted scripts and, relevant here, cloning one's own voice to create voiceovers or dubs, and it states that disclosure alone does not limit a video's audience or its eligibility to earn money. A synthetic voice making a real person speak words they never said would count as disclosable. The disclosure itself is the "AI use" setting under Attributes when you upload in YouTube Studio; AI-generated audio that is not music is labelled in the expanded description. YouTube may also add a label automatically, for example to content carrying C2PA metadata, and says creators who consistently fail to disclose can face penalties up to removal of content or suspension from its Partner Program. If the studio uploads on the client's channel, agree who sets the AI use answer. Other platforms have their own AI-label tools and policies that change often, so check the current page of each platform where the video will run, and let the client know which labels you will apply.

To the audience and the law. Some jurisdictions are introducing transparency duties for synthetic audio and video. If the client has viewers in the EU, ask a qualified adviser about the AI Act's transparency provisions and their timing. Article 50 requires deployers, meaning the businesses using the AI system, to disclose that a deep fake was artificially generated or manipulated, and the Act's definition of a deep fake includes AI-generated audio that resembles existing persons and would falsely appear authentic; how and when that bites on a given video is a question for the adviser. For advertising, general rules against misleading viewers apply regardless of labels: a voice that claims to be a customer, a doctor or an expert it is not is a problem whether or not it carries a tag.

For wording and examples of labels in practice, see disclosure rules for AI presenter videos and AI video disclosure for travel clips.

A check before delivery

Add this to your delivery checklist. It takes five minutes per video and catches the mistakes that cause real disputes.

[ ] Voice type recorded: stock / cloned (whose?)
[ ] Written consent on file for any cloned or hired voice
[ ] Tool plan and its commercial-use terms checked this month
[ ] Script approved by the client in writing (version: ___)
[ ] Audio approved by the client (date: ___)
[ ] Pronunciation of names, products and numbers checked
[ ] Numbers, prices and claims in the script match the client's sources
[ ] Platform AI labels applied or ruled out (platforms: ___)
[ ] Description or credits mention AI voice where we agreed to
[ ] Source files, consent forms and approvals stored in the project folder

Realistic mistakes and how they show up

The three cases below are illustrations of patterns studios run into, not reports of particular projects. In the first, a security installer's video includes the line "installed in under two hours". The AI reads "two hours" correctly but mispronounces the company's trade name, a name made of two joined words. The client only notices after publication, and the fix is a re-render and replacement upload. A pronunciation check against a written phonetic guide would have taken a minute. Make it part of the script sign-off: list the names, then listen for each.

In the second, a bicycle repair shop's owner asks for a short promo using a voice that "sounds like a well-known presenter". The studio declines. Imitating a recognisable person's voice invites claims about their voice and likeness, whatever the tool permits, and it can mislead viewers into thinking the person endorses the shop. The studio offers a stock voice with a similar warm pace instead, and the shop is content.

In the third, a solar installer's sales manager has been cloned by a freelancer using recordings from a webinar. The manager was never asked. When the freelancer leaves, the studio discovers the clone exists and has to find out who holds it. Nothing malicious has happened, but there is no consent, no record of where the audio came from and no deletion process. The simple rule that the person creates and verifies their own clone, and shares it, prevents this.

Hired voice actors and AI: where the contract bites

If a human voice actor is already on the project, switching to AI midway is a contract question before it is a technical one. Read the booking terms for three things: whether the fee covers the final audio only or also reuse, whether the actor's recordings may be used for anything other than the named video, and whether the booking can be cancelled and on what terms. Using an actor's recordings to build a clone, even of a voice you paid for, needs their separate written permission, and many actors expect extra payment for it. Never feed a hired actor's session audio into a cloning tool as a convenience.

Some actors now offer a licensed AI version of their own voice as a product. That is the clean route when a client wants a recognisable professional voice: the actor sets the terms, the fees and the permitted uses, and you add the licence to the project file.

A script prompt and what you fix in the output

Scripts written for AI voices need different care from scripts for human readers. A prompt that produces a usable first draft:

Write a 90-second narration for a safety induction video. Audience:
new HVAC technicians. Short sentences, maximum 18 words. Spell out
numbers as they should be spoken ("two point five metres"). Write
abbreviations in full on first use. Put a pause marker [pause] between
sections. Do not invent regulations or figures; use only the facts
below. Facts: [paste approved procedure text].

An illustrative first output opens: "Welcome aboard. Before you climb a ladder or open a unit, take two minutes to read this. [pause] First, isolate the power." What you would fix: replace "climb a ladder" with the exact wording from the procedure, check "isolate the power" matches the approved term, and add phonetic spelling for any product name. The model followed the structure but could not know the company's approved terminology, so a human edit against the procedure is the step that matters. Send the corrected script to the client for approval before generating any audio.

Dubbing and translated versions

Many clients ask for the same video in several languages. AI voices make this affordable, but consent and quality checks multiply. A cloned voice speaking a language the person does not speak is a new use of their voice, so the consent form must say so, and the person should hear and approve at least a sample. Translations need a fluent human reviewer, because a wrong sentence in a language nobody in the studio speaks is invisible until a viewer complains. Pronunciation of names and product terms is the usual failure. Budget a reviewer's time per language, and list it in the quote.

A quick sum for pricing the change

Say a studio used to charge for a two-hour voice-over session per video, with a freelance actor at a fixed fee plus an engineer for an hour, and produced eight updates a year for one client. With a licensed AI voice, the studio spends one hour on the script check, 30 minutes generating and listening, and 30 minutes on the approval round, about two hours per update instead of roughly four plus the actor's fee. Over eight updates that is about sixteen hours saved, against a one-off half day setting up consent and the voice, and a monthly tool cost in the tens of dollars. The numbers are illustrative, and the point is that the saving exists but is smaller than "voice generated in seconds" suggests, because the checking and the paperwork are where the time goes.

When the client says no to disclosure

Occasionally a client asks you to leave labels off or to say the voice is real. Decline in writing. Misleading viewers about a real person's words is the scenario platform rules and advertising rules are designed to stop, and a studio that goes along with it shares the consequences. Offer alternatives that meet the client's concern: a brief unobtrusive label in the description, a stock voice that does not suggest a particular person, or a human recording for the one video where authenticity matters most.

Costs and time

List prices change, so check the current pricing pages. ElevenLabs lists Starter at $6 a month with 30,000 credits and Creator at $22 a month with 121,000 credits, and sometimes discounts the first month. Its pricing page puts text to speech at about one credit per character, with every product drawing on the same pool. A quick, illustrative sum: a 90-second script of about 220 words is roughly 1,300 characters, so with three or four regenerations for pronunciation fixes each video uses around 5,000 to 6,500 credits: about 20 such videos a month on a Creator allowance, and four or five on Starter. The cost of the subscription is small against the studio time it saves on updates, but the real cost lines are the consent paperwork, the approval round trips and the pronunciation and quality checks. If you charge by the finished minute, price AI voice as a service with its own rate rather than silently cutting the studio-session line to zero: clients value the speed, and the checking is real work.

When a human voice is still the better choice

Choose a human voice actor when the video is brand-defining and the voice carries emotion, such as a flagship film, a charity appeal or a story with real stakes. Choose one when the audience expects a real person (a testimonial, an interview, a founder's apology), when the content is regulated or sensitive, or when the client's own contract with a presenter requires it. AI voice earns its place in training, explainers, product walkthroughs, internal updates and any content that changes often. For the full set of tools in a studio, AI filmmaking tools by job shows where voice sits among them, and ElevenLabs versus Speechify compares two voice tools directly.

Your first client conversation

When a client asks for AI voice, say three things: which type of voice you propose and why, what you need from them in writing, and how the video will be labelled. Then send the clauses above with the proposal. If you want help setting this up across a studio, including the contract wording and the delivery checklist, that is what my AI implementation consultation covers, and you can reach me through the contact page. For a wider view of AI across the business, the AI hub lists every tutorial in the series.

Voice-over consent questions production companies ask

Can I clone a client's director or founder for their video?

Only with that person's own written consent. For an ElevenLabs professional clone, the person must create and verify it on their own account and share it with you. Instant clones ask you to confirm you have the right and consent to clone the voice. Their employer's permission is not enough, and neither is a verbal yes on set.

Is a stock AI voice safer than a cloned one?

Generally yes. A stock voice carries no individual who can object to being cloned. You still need to read the tool's commercial-use terms, check the voice's licence conditions and avoid claiming it is a real person.

Do I have to say a voice over is AI on YouTube?

YouTube asks creators to disclose realistic altered or synthetic content, such as making a real person appear to say something they did not. Its guidance exempts cloning your own voice for voiceovers. Check the page for the current wording, because it changes.

Can I replace a voice actor with AI mid-project?

Not without checking your contract with them. If they were hired for the project, a change may breach it, and if you hold recordings of their voice, using those to train or build a clone needs their separate written permission.

What if the client later withdraws consent for a cloned voice?

Put the answer in the agreement before the project starts: whether existing videos stay live, whether the clone is deleted, and the notice needed. The person whose voice it is should be able to withdraw for future use.

Further reads

• AI Shot List Generator: Build a Shoot-Day Plan From the Script — Plan the shoot day before the voice work.
• AI Storyboard Generator From a Script: A Video Producer's Workflow — Draft the script the voice will read.
• AI Video Production Software Compared for a Small Studio — Compare AI software across the whole studio.
• Best AI for Filmmakers Selling Video Services to Local Businesses — Selling video services to local businesses.
• Create a Marketing Video With AI: A Shop Owner's Short-Clip Workflow — A short-clip workflow clients can follow.
• How to Label AI-Generated Images and Video on Social Media — Practical wording for AI labels on social posts.
• Best AI for DJ Drops: Custom Voice Tags for Event DJs — How an event DJ makes custom voice tags and client name drops with AI, from script to sampler pad, with the licence and consent checks to pass first.
• AI Subtitle Generator for Client Videos: Accuracy and Delivery Checks — Measure an AI subtitle generator on your own audio, then run line-length, reading-speed, export and QC checks before a client video ships.
• AI Video Script Generator for Production Companies: Brief to Draft — Use a two-pass brief-to-script process that protects client facts, runtime and approval history.
• AI Video Translator for Corporate Video: Dubbing, Subtitles, Checks — Choose subtitles or dubbing, cost each language and review the translated video scene by scene.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ElevenLabs pricing page and its documentation on professional and instant voice cloning; YouTube Help on disclosing AI-generated content; Article 50 of the EU AI Act; all checked in November 2026, plus general contract practice. Platform rules and laws change, so check the current pages and take advice on contracts.