Two postings, same $500 fee. One wants a 30-second radio read, and once the files are delivered, you’re done. The other hands the client a license to build a synthetic copy of your voice and keep it talking, maybe for years. Same money on the screen. Wildly different deal underneath.
You’ll run into that split more and more when browsing voice over jobs. The big casting marketplaces now separate listings into traditional voice-over, voice data, and branded AI voice, and each one comes with its own kind of paperwork. Below is what’s worth reading closely, and what to push back on, before your signature goes anywhere.
Quick caveat: this is general contract information, not legal advice. Whether any clause holds up depends on governing law, union coverage, and the precise wording in front of you.
First, figure out what you’re actually selling
Skip the fee for a moment. The more useful question is what leaves the room with the client. Plenty of postings look like a plain commercial gig and still carry a clause allowing new lines to be generated long after you’ve packed up the mic.
Here’s a rough way to sort them:
| Project type | What the client walks away with | Your role afterward | Where the risk sits |
|---|---|---|---|
| Traditional voice-over | A recorded performance for stated uses | Agreed revisions or pickups | Usage scope, term, media |
| Voice-data session | Recordings used to build or evaluate AI models | Usually ends at delivery | Training purpose, retention, reuse |
| Licensed AI voice | Permission to create synthetic audio based on you | Possibly approvals, updates, renewals | Output limits, duration, exclusivity, pay |
A classic session is the familiar one. The contract names the script, files, media, and audience. A regional spot might be capped at one territory for 13 weeks; audiobooks follow their own conventions. Pay tracks the usage you agreed to, and that’s that.
Voice-data work is a different animal. Nobody’s going to hear your recordings as an ad. They’re feeding a model, and the buyers are often AI voice startups trying to test voice products in real conditions, so they want lots of clean, varied speech. Worth asking which system gets trained, who can open your files, whether your data gets blended with other people’s, and whether a second model would need your permission all over again.
Licensed AI voices go one step further. The U.S. Copyright Office’s 2024 report on digital replicas describes them as recordings digitally created or altered to realistically but falsely depict a person. License one of those, and it can say things you never said, inside whatever fences your contract builds. No new booking needed on their end.
Consent language is where most deals go wrong
“We may use AI” tells you almost nothing. Solid consent wording names the actual model or project, lists what outputs are allowed, pins down media and territory, says who gets to use the voice, and rules out contexts you’d never want attached to it. Political ads come up a lot here.
Union contracts give a decent yardstick. SAG-AFTRA’s 2023 TV/Theatrical deal brought in informed consent and compensation requirements for creating and using digital replicas, and the 2026 successor, ratified in June and running through June 30, 2030, pushes those protections further. Keep in mind they cover union work. Sign an independent contract on your own, and they don’t tag along.
Some states have filled part of that gap. In California, Labor Code Section 927 took effect on January 1, 2025. A digital-replica clause can become unenforceable there if it lets a replica stand in for work you’d otherwise do in person, if it lacks a “reasonably specific” description of intended uses, and if you had neither a lawyer nor a union representing you when you agreed. Illinois and New York passed laws along similar lines. Details differ, so check both your own state and wherever the client is based.
Training and output are two separate asks
People blur these together, and clients sometimes count on that.
Training rights let your recordings serve as model input. Output rights let the client produce brand-new synthetic audio. A contract might request one, the other, or both, and nothing stops you from pricing them separately.
The vocabulary gets slippery fast. Fine-tuning means training an existing model further. A derivative model is a modified version of one that already exists. Sublicensing means handing your material to someone else entirely. If the model can be transferred, the contract should say who’s allowed to receive it, and any change to that list ought to set off whatever approval step you negotiated.
How long is this thing going to follow you?
A fixed term has an end date. A perpetual license doesn’t stop on its own, ever, unless something in the contract ends it. If there’s renewal language, look for who kicks off an extension, how the new fee gets worked out, and whether “no” is an option for you.
Termination clauses are where sloppy contracts tend to unravel. An end date says nothing about stored recordings. Push for written answers on what happens to retained training data, whether the model gets deleted or switched off where that’s technically doable, and whether audio generated before the cutoff can keep running. If it can, the contract should list those files and how long they stay live.
Exclusivity: small print, big consequences
Promising not to voice a rival product during a six-week campaign is a modest favor. Agreeing to stay out of an entire industry, including affiliates nobody bothered to name, is a different matter altogether. Lay the category and territory next to the term and picture the bookings you’d have to turn down. Voice one bank’s AI assistant, and a competitor’s radio campaign could be off the table.
Getting paid for the booth and for what comes after
Your session fee covers your time behind the mic. What happens to your voice afterward is a separate thing of value, and it deserves separate money: a renewal fee, a usage-based payment, a premium for exclusivity, or some mix of those.
When pay depends on usage, insist on reporting terms and a clear trigger for each payout. Don’t assume ongoing involvement either. Get it in writing whether you approve outputs, what extra recording or model updates pay, who can request them, and how many revision sessions the base fee already covers.
Run these questions before signing
Read every page of the agreement, including the linked policies. A tidy two-line job post can leave out permissions buried in an attached license. Save copies of those policies on the day you accept, particularly if the client reserves the right to edit posted terms later.
Then go through this list:
- Which recordings and which model does my consent cover, and who may use them?
- Training, synthetic output, or both?
- Which topics, languages, media, and territories are allowed, and who gets to hear it?
- How long does the license last, and can I turn down a renewal?
- Does exclusivity block named competitors or a whole category?
- How are the session and the license paid, and what triggers later payments?
- What happens to the model, the data, and existing audio once the deal ends?
Got an answer by email or on a call? Ask for it in the signed contract. Verbal reassurances have a way of evaporating.
Starting doesn’t lower the stakes
A newer actor’s voice is every bit as useful to a model as a veteran’s. Before shopping for a mic or prepping an audition, find out whether the client wants a finished performance or raw training material, since that single answer reshapes the whole contract. With online postings in particular, weigh the pay against everything you’re granting, not just the hour you’ll spend recording.
Quick answers to things actors ask
Q. How do I become a voice actor from home?
Carve out a quiet space, pick up basic editing, cut a focused demo and audition for parts that fit you. A home setup changes nothing about contract risk, so you’d still check usage terms and any right to train on your files.
Q. Where can I find voice acting gigs in Chicago?
Casting platforms, local talent agencies and the city’s production scene are the usual places, with location filters where available. Plenty of remote roles take Chicago-based talent too. Illinois also has its own Digital Voice and Likeness Protection Act covering digital-replica clauses, which is handy to know about before signing.
Q. How much does voice over pay?
No standard rate exists. Workload, distribution, license length and exclusivity all push the number around. With synthetic-use deals, make sure the session fee and the payment for future output show up as two separate lines.
Q. What’s the best voice-over site?
Depends on you. Match the work on offer to your strengths, then look at payment procedures and how disputes get handled. You should be able to find the client’s intended use in both the listing and the agreement before saying yes.
Q. How do I get hired for voice-over work?
Build a demo for the specific role, complete your profile, and follow audition instructions to the letter instead of sending whatever sample you have handy. Before accepting, confirm the end use and ask outright whether your recordings will train or run an AI system.
Q. Can ChatGPT do voice overs?
It can produce spoken audio in supported conversation modes. Hiring a performer or getting permission to copy a particular person’s voice is a separate question entirely, and access to a generator doesn’t license anyone’s voice. Given how easily convincing voice deepfakes get made now, the consent terms in your contract carry real weight.
Before you sign
An old-school booking pays for one performance with set usage. An AI deal may tack on training access, future synthetic lines, exclusivity, and obligations that keep going, all while your time in the booth stays short. So it’s worth pausing on two things. Can you describe, in plain words, what the client may do with your voice and what remains yours? And does the money actually match how far those rights stretch?
Related: AI Wealth Concentration: Who Owns the $20 Trillion Boom?
