Your Voice. Their AI. Who Gets Paid?
Licensing your voice could add income without another recording session. But the check is only one part of the deal.
By Joe “Crash” Kelley
SonicAttention | September 26, 2026
A lot of my voiceover artist friends are understandably concerned that their talent will be replaced by AI-created voices. I understand the concern. When you have spent years developing your delivery, building client relationships and making a living behind a microphone, watching a computer turn a paragraph into a convincing performance can feel pretty threatening.
And let’s be honest about what these tools can do. Some AI voice models, carefully prompted and directed, can produce remarkable content. In the right creative hands, a talented producer can develop a wide range of commercial performances, characters and approaches. I use these tools in my own work. Pretending they are incapable of producing something good would not be an honest starting point.
But this does not have to be the end of the line for a talented voice. There may be another opportunity here: license an authorized version of your voice and collect revenue without personally spending hours in the studio reading every piece of copy. Arrangements that compensate performers for licensed AI voices already exist, although their payment structures and protections differ considerably.[1][2][3]
Some clients will still want the actual person in the session. They want your interpretation, your instincts, your ability to hear a direction and find something the scriptwriter didn’t quite put on the page. I would not casually throw that part of the business away.
Other clients may be comfortable with a creatively directed, authorized AI version of your voice. For the right work, under the right agreement, that could become additional revenue rather than simply another competitor for the work you already do.
Notice the words “under the right agreement.” That is where this discussion gets important.
I don’t have a problem with a performer finding another way to get paid for something they spent decades developing. I have a problem when everybody in the transaction gets paid except the person whose voice made it possible. And I have an even bigger problem when that person gets a check but loses control of what the voice is used to say.
He retired. His voice didn’t.
There is an actual radio example worth examining. In a September 23 Radio World article, KHYI in the Dallas–Fort Worth market described “Bot Dillon,” an AI version of longtime personality Brett Dillon. According to the account, Dillon participated, gave permission and receives a monthly licensing stipend. The station discloses the AI use and assigns the voice a two-hour weekday shift. It launched June 8, 2026.[1]
That is a different proposition from secretly feeding somebody’s airchecks into a machine after showing them the door. A person participated in the decision, agreed to a defined assignment and receives an ongoing payment.
There is an important qualification: the article is a Buyer’s Guide testimonial, not independent audience research. It does not disclose the stipend amount or full contract. We cannot evaluate the compensation, approval rights or exit terms from that account alone.[1]
Still, it presents a possibility worth considering. A retired broadcaster may welcome income that does not require another daily shift. A working performer might see value in licensing specific projects while continuing to record others personally.
The opportunity is not necessarily to stop being a voice professional. It could be to add another way of doing business as one.
“The actor gets paid” is the beginning of the conversation
This is not limited to one station. September coverage of BionicVO’s partnership with Yaman Media Group describes a platform featuring professional actors who have consented to voice cloning. The company says actors are paid when their voices are used. That is its stated model, not an independent accounting of what individual performers earn.[2]
ElevenLabs also documents a Voice Library payout system. Eligible performers can earn when paying users generate audio with their shared professional voice clones. Free-user generation does not earn those rewards, and default rates depend in part on the notice period the performer selects.[3]
Those examples show why the details matter. A monthly stipend, a payment per generated file and a fee for an advertising campaign are not interchangeable arrangements.
What counts as a use? Generating the audio? Downloading it? Putting it into a commercial? Running that commercial on one station—or distributing it across an entire company?
Imagine a hypothetical spot that is generated once and then used repeatedly on radio, streaming and social media. Compensation triggered by generating the file measures something very different from compensation tied to the campaign’s duration and distribution.
Before getting excited about the royalty, find out what the royalty meter is actually counting.
Price the permission, not just the afternoon in the booth
If somebody wants to license my voice, I would want to separate two conversations. What are you paying me to help create the model, and what are you paying me for the right to use it afterward?
Recording material is a service. Authorizing future performances is a different commercial opportunity. I would not want the second one quietly included because the first one only required an afternoon.
The 2025 SAG-AFTRA Audio Commercials agreement provides a useful example. For covered work, it addresses payment for time spent creating a digital replica and specifies compensation for replica-generated performances, including a 1.5-times-scale session payment subject to stated exceptions, plus use payments as though the performer had worked in person.[4]
That is not a universal rate card for every radio employee, freelancer or platform. It applies to the work and parties covered by the agreement. But it demonstrates an important principle: eliminating the need for another recording session does not automatically eliminate the value of the performance.
I would also look carefully at exclusivity. If an arrangement prevents me from working for competing advertisers or stations, that restriction belongs in the compensation discussion. A check can look attractive until you discover how many other checks you agreed not to pursue.
One station using approved imaging is not the same proposition as a company using your vocal identity across multiple formats, markets and advertisers. Those should not become the same deal simply because somebody calls both of them “AI licensing.”
Permission to clone is not permission to say anything
Suppose I approve a digital version of my voice for a station’s imaging. Does that also authorize commercial endorsements? A customer-service chatbot? A different language? Another station the company acquires next year?
I would want those questions answered before the agreement is signed—not when somebody sends me a recording and asks when I started working for a business I have never heard of.
The covered SAG-AFTRA commercial provisions require a reasonably specific description of the intended use. They also require additional consent when a proposed use goes beyond that description.[4] That is a useful distinction even for performers whose work falls under a different arrangement: permission should describe something, not merely waive everything.
There is another distinction worth understanding. The U.S. Copyright Office’s digital-replicas report explains that a person’s voice and a copyrighted recording of that voice are not the same thing. Rights involving a recording, a person’s identity and contractual permission should not be treated as one interchangeable package.[5]
In everyday language, “We already have your audio files” should not end the conversation. Neither should “I own my voice” substitute for having a qualified attorney examine what an agreement actually permits.
The National Association of Voice Actors identifies consent, usage limits, duration, opt-out provisions, compensation, exclusivity and secure storage and tracking as important contract issues.[6] None of that needs to become a vocabulary test. It comes down to familiar questions: who can use it, for what, for how long, and what happens when somebody crosses the line?
Find the exit before you walk through the door
Getting your voice into a system is one question. Getting it out is another.
And “getting it out” can mean several things. Preventing new customers from selecting it is not necessarily the same as preventing existing customers from generating more audio. Stopping new generation is not necessarily the same as ending use of recordings already created.
ElevenLabs’ March 6, 2026 Voice Library Addendum makes those distinctions explicit. Existing users can retain access during the selected notice period, and audio generated before that period ends remains available for use afterward. The notice period cannot be shortened after sharing, although it can be extended.[7]
That does not automatically make the arrangement unacceptable. It does mean the performer needs to understand what continues after pressing the removal button.
The same terms say optional live-moderation tools are not guaranteed to block prohibited categories.[7] An automated filter is not the same thing as personally approving a script.
I would want a plain-English explanation of what happens to the model, existing audio, access and payments when the relationship ends. I would also want to know what happens if the station is sold or the technology vendor changes.
“Don’t worry, you’re in control” needs to survive a more specific question: What, exactly, can I stop?
You do not have to choose the broadest possible deal
Even within one technology company, the arrangements can differ. ElevenLabs’ Iconic Marketplace describes a process in which customers submit project requests, talent or authorized rights holders review them, and the parties agree on licensing terms. Submitting a request does not guarantee approval.[8]
That is different from broadly sharing a voice in a self-service library. It does not mean every announcer can command a celebrity’s terms. It does demonstrate that “AI licensing” is not one unavoidable business model.
A station personality might prefer a limited arrangement with a known employer. A freelance performer might prefer individually approved campaigns. Someone else might accept broader availability because the potential volume fits their business.
The sensible question is whether the performer understands the trade-offs and has a meaningful choice. I would not let “passive income” do all the selling. The revenue might arrive without another session, but the agreement could still affect an active career.
It is perfectly reasonable to decide that one opportunity fits and another does not.
Your clone didn’t buy a truck from that dealer
Here is where I draw a hard line.
A cloned voice must never be used to deliver a personal endorsement for a product or service the actual person has not tried or used. A licensing agreement should expressly prohibit that—not bury permission for it in the fine print.
That is the standard I would insist on for my own voice. A license authorizes specified production work. It does not authorize somebody to manufacture my personal experiences.
Imagine a salesperson submits a script saying, “I bought my last three trucks here, and I wouldn’t go anywhere else.” The cloned voice delivers it beautifully. The client loves it. The spot is ready to run.
Except the person behind that voice has never bought a truck there.
That is not a creative interpretation. It is an invented endorsement delivered in the sound of a real person. Being able to generate the sentence does not make it acceptable to use.
The FTC’s endorsement guidance is direct: people cannot discuss an experience with a product they have not tried, imply regular use after trying it once, or claim enthusiasm they do not honestly hold.[9]
My contract would go further than simply saying, “Follow the rules.” It would explicitly prohibit fabricated personal experiences, purchases, satisfaction, recommendations and results. A genuine personal endorsement would require separate written approval of the specific claim and intended use, along with review of the finished performance.
And trying something would not give the advertiser permission to invent the rest. “I sampled it” does not mean “I use it every day.” Permission to use my voice does not mean permission to invent my opinion.
Not every commercial read is a personal endorsement. An announcer communicating store hours is different from a recognizable personality claiming to trust a business with their family. The context and what consumers reasonably understand matter.[9]
The FTC’s guidance also makes clear that using AI avatars does not create an exemption for false testimonials. AI-generated marketing is not categorically prohibited, but false underlying testimonials and misleading representations can violate the rules.[10]
For me, the distinction is simple: you can license the sound of my voice. You cannot license a fictional version of my life.
That prohibition needs financial teeth
I would not sign an agreement that treats a fabricated endorsement as a minor production error. Any advertiser or licensee permitted to use my clone should be bound by that prohibition, with substantial financial consequences for violating it.
Not a polite reminder. Not merely a refund of a small licensing fee while the performer deals with the damage.
I would want my attorney to address immediate suspension of access, removal of unauthorized material, preservation of usage records, responsibility for corrective communications and meaningful financial remedies. I would also want the agreement to address downstream use, so the licensee cannot simply pass the voice to somebody else and act as though its responsibility ended there.
There is a legal detail worth getting right. What we casually call a “severe financial penalty” needs to be drafted as something a court can enforce. Courts can reject contractual amounts that are punitive rather than reasonably connected to the anticipated harm. Properly structured liquidated damages—an agreed amount or formula for difficult-to-measure losses—are different from an arbitrary punishment.[11]
That is not an argument for going easy on the offender. It is an argument for having an attorney build serious protection under the law that governs the deal, rather than inserting an impressive number that falls apart when challenged.
I would want the remedies to reflect the foreseeable harm from attaching my identity to something I never said or believed. I would want counsel to address enforcement costs and ensure any limitation-of-liability language does not quietly undermine the protection we just negotiated.
Paying damages should not purchase permission to keep using the false endorsement. The use must stop.
The objective is a prohibition with enforceable consequences—not an optional fee for borrowing somebody’s reputation without permission.
Sounding human and being trusted are different tests
We have spent a lot of time asking whether listeners can hear the difference. That matters to production quality. It does not settle every question about audience acceptance.
Crowd React Media’s July 2026 study tested two scripts with 1,326 weekly U.S. radio listeners ages 18–45. The company reported that listeners did not reliably identify which versions were AI and that most performance ratings were comparable. After disclosure, 48% of those hearing a human voice felt more favorable, compared with 25% of those hearing AI. Among the AI listeners, 20% felt worse and a majority reported no change.[12]
That is not evidence that everybody rejects AI. It also was not a test of a familiar local personality licensing a clearly disclosed clone. We should not pretend those results answer that particular question.
What it tells me is that a convincing performance and an accepted business practice are not necessarily the same thing. My response would not be to hide the source more effectively. It would be to explain the arrangement honestly and make sure the actual person has meaningful involvement.
There is a difference between a performer approving a tool-assisted production and a system independently creating new statements in that person’s voice. I would want the description given to the audience to match what is actually happening.
And I would want a named human responsible for reviewing the output. “The computer wrote it” would not satisfy me as an explanation for something aired under my name.
A royalty check does not answer every employment question
There is another wrinkle. A deal can compensate the owner of a recognizable voice while raising separate questions about other performers’ work.
In May 2025, Epic introduced a conversational AI Darth Vader in Fortnite with support from James Earl Jones’s estate and a statement from his family. Days later, SAG-AFTRA announced an unfair-labor-practice charge against Llama Productions, alleging a failure to provide notice and bargain over replacing performers’ work.[13][14]
Those were not the same dispute. The union acknowledged members’ and estates’ rights to control their digital replicas while raising a separate bargaining issue. A filed charge is an allegation, not a finding; I am citing the historical disagreement, not claiming an outcome.
For radio, the distinction is useful. Even when the person whose voice is cloned participates willingly, somebody still has to consider the rest of the operation. Who writes, produces, checks and updates the material? What work is being added, and what work is being removed?
I see a legitimate opportunity for licensed voices to generate income. I also want to know whether the overall use of the technology is improving the product or simply reducing the number of people responsible for it.
Those questions can exist in the same conversation.
Yesterday’s announcement is not today’s contract
Here is something I found during the research that should make anyone slow down before signing. SAG-AFTRA’s pages for its previously announced Narrativ and Replica Studios agreements now carry September 24, 2025 notices saying those specific contracts are no longer in effect.[15]
Those notices do not, by themselves, explain why or establish the status of every other arrangement involving those companies. They do mean that an old announcement is not proof that the same protection applies today.
Before relying on phrases like “union approved,” “fully protected” or “talent controlled,” I would want the terms currently covering the specific work. Not a headline from two years ago. Not a logo on a sales presentation.
This is useful for station management, too. A producer who knows exactly what is permitted is in a better position than one operating on somebody’s recollection of a demonstration.
Good paperwork is not the enemy of getting the spot finished. Unclear permission is how a finished spot becomes another problem.
Start with a business plan, not an upload button
My advice to a voice professional considering this would be to start with a limited opportunity you can actually evaluate. One advertiser or station. A defined assignment. A clear term. An understood payment arrangement. Named people who can access the model and approve the work.
I would want to know whether the project supplements existing income or undercuts work I already sell. I would also want to understand what happens when the customer wants a change the model cannot deliver properly. Does the actual performer get called in—and paid—for that additional work?
For a client who wants the live session, keep offering the live session. For a client and project suited to an approved digital performance, offer that under terms you can live with. The opportunity may be two products, not the abandonment of one for the other.
I would insist on useful records of what was generated and where it was used. It is difficult to evaluate a royalty arrangement when the only explanation is that the system calculated a number.
NAVA provides a free Synthetic Voice/AI rider as a starting point and recommends professional review for the individual agreement.[16] A template is not a substitute for advice, but a performer does not have to begin with a blank page, either.
And choosing not to license is a legitimate decision. Nobody should have to surrender broad control of a professional identity to prove they understand new technology.
The voice still belongs to a person
I am interested in what a talented producer can do with these tools. I am equally interested in whether talented voice professionals can participate in the value being created.
A carefully licensed clone could handle suitable work while the performer concentrates on higher-value sessions, develops other business or simply enjoys more time away from the studio. Those are possibilities worth examining—not promises of effortless income.
But the opportunity should be built around the person, not merely around a library entry with a payout attached.
That means permission someone actually understands. Compensation that matches what is being licensed. Limits that remain meaningful after the model is created. And consequences when those limits are violated.
It also means remembering that the quality of the work still matters. A convincing voice cannot rescue a weak idea, fix a dishonest claim or replace the judgment needed to understand the advertiser and the audience. The technology is a production tool, not a substitute for having something worthwhile to say.
So to my voiceover friends: I would not dismiss licensing out of hand. There may be a way to turn a development that understandably concerns you into an additional source of income. But I would not rush into a broad agreement simply because the first demonstration sounds impressive.
Hear the demo. Explore the opportunity. Then read the deal.
I have said that AI should make radio better, not smaller. Here is another part of that argument: making a performance easier to generate should not make the person who built its value easier to disregard.
Let your voice work when you’re not in the booth. Just don’t let somebody else decide what you stand for.
Sources and reporting notes
Publicly available reporting and terms reviewed September 26, 2026. Company descriptions are identified as company claims, and undisclosed compensation is not presented as verified earnings. The contract standards advocated here are the author’s recommendations, not a claim that every existing agreement supplies them. This is reported business commentary, not individualized legal advice.
[1] KHYI and “Bot Dillon.” RW Staff, “‘Bot Dillon’ Picked Up Where Brett Dillon Left Off,”Radio World, September 23, 2026. The publisher identifies the article as a Buyer’s Guide testimonial. It reports Dillon’s permission, direct participation, monthly stipend, disclosed AI use and limited weekday assignment. The full contract, payment amount and independent audience results are not provided.
[2] BionicVO’s stated compensation model. Kevin Carter, “BionicVO Partners With Yaman,”RAMP, September 21, updated September 24, 2026. The consent-and-payment description comes from the company’s announcement. The report does not establish individual royalty rates or earnings.
[3] Voice Library payouts. ElevenLabs, “Payouts” documentation. Describes eligible professional voice-clone rewards, paid-user versus free-user usage and the relationship between default rates and notice periods. Participation does not establish demand or guarantee income.
[4] Compensation and consent for covered commercial work.2025 SAG-AFTRA Audio Commercials Contract Memorandum of Agreement, digital-replica provisions, particularly Sections C and D. Addresses reasonably specific consent, creation time, generated performances, session payments, use fees and exceptions. These provisions apply to covered work, not universally to every employment or licensing arrangement.
[5] Voice, identity and recording rights. U.S. Copyright Office, Copyright and Artificial Intelligence, Part 1: Digital Replicas, July 2024. Explains distinctions between vocal identity and copyright in particular recordings. The report is not presented as a comprehensive September 2026 survey of every applicable law.
[6] Performer contract considerations. National Association of Voice Actors, “Synthetic/AI Voices.” Identifies consent, usage limits, duration, opt-out provisions, compensation, exclusivity, storage and tracking as important issues. These are recommended protections, not assurances that a particular contract includes them.
[7] Removal, continuing use and moderation. ElevenLabs, Voice Library Addendum, last updated March 6, 2026, especially Sections 4 and 6. Explains notice periods, continued availability of previously generated output and limitations of optional moderation. Other terms and applicable law also matter.
[8] Project-specific licensing. ElevenLabs, Iconic Marketplace and licensing FAQs. Describes project requests, rights-holder approval and individually negotiated terms. This offering is distinct from the general Voice Library and does not establish identical terms for all performers.
[9] Truthful endorsements and actual experience. Federal Trade Commission, The FTC’s Endorsement Guides: What People Are Asking, particularly the section explaining that endorsements must represent accurate experiences and opinions. Whether an advertisement conveys a personal endorsement depends on its context. The article’s proposed blanket contractual prohibition on fabricated personal endorsements is the author’s recommended standard.
[10] AI avatars and false testimonials. Federal Trade Commission, The Consumer Reviews and Testimonials Rule: Questions and Answers. Addresses AI-generated avatars, false underlying testimonials and unauthorized celebrity avatars. It does not impose a blanket prohibition on AI-generated marketing.
[11] Enforceable damages versus punitive contract penalties. Cornell Legal Information Institute, “Liquidated Damages” and “Penalty Clause”; see also the New York Court of Appeals’ discussion in JMD Holding Corp. v. Congress Financial Corp. The distinction depends on applicable law and the agreement’s circumstances. An arbitrary punitive amount is not made enforceable simply by calling it “liquidated damages.” Counsel should structure appropriate remedies and their interaction with other contract provisions.
[12] Listener response to human and AI voiceover. Katie Miller, “Can Listeners Tell the Difference Between AI and Human Voiceover? We Tested It,”Crowd React Media, July 6, 2026. Fieldwork involved 1,326 weekly U.S. radio listeners ages 18–45 in May and June 2026. Results concern two tested scripts and reported reactions—not long-term audience behavior or a disclosed clone of a familiar local host.
[13] Estate-supported character-voice use. Epic Games, “This Will Be a Day Long Remembered: Speak with Darth Vader in Fortnite,” May 16, 2025. Acknowledges James Earl Jones’s estate and includes his family’s statement supporting the collaboration. Financial terms are not disclosed.
[14] Separate labor allegation. SAG-AFTRA, “Statement on Fortnite’s Use of A.I. Darth Vader Voice and ULP Filing,” May 19, 2025. Documents the union’s allegations and distinguishes its bargaining claim from control of a digital replica. This article does not assert an adjudicated violation or report the case’s ultimate outcome.
[15] Changes in agreement status. SAG-AFTRA’s Narrativ agreement page and Replica Studios announcement carry September 24, 2025 notices stating that those specific contracts are no longer in effect. The notices do not establish the reason or the status of other arrangements.
[16] Practical contract-review resource. National Association of Voice Actors, Synthetic Voice/Artificial Intelligence Rider. The rider is available free. NAVA describes it as an educational resource and recommends review of individual agreements by the performer’s attorney or agent.
Joe “Crash” Kelley is a radio operations manager, programmer, on-air personality and the creative force behind SonicAttention.