Hollywood writers, voice actors, and post-production artists are increasingly taking paid gigs rating, correcting, and feeding their own craft into AI models. That work trains the same systems studios already use to cut production costs. It's a direct trade: steady entertainment work has thinned out, and AI labs pay well for exactly the skills these workers already have.
Short answer: Screenwriters, editors, and voice actors are picking up freelance contracts on platforms like Mercor, Outlier, and Turing, rating and rewriting AI output for roughly $12 to $200+ an hour depending on skill, with some specialists billing $350. Separately, unions like SAG-AFTRA now require studios to notify and bargain before licensing a performer's voice or likeness to train an AI model. Both are real, current income paths, with real tradeoffs.
I test AI tools for a living. So when writer and showrunner Ruth Fowler published her account of doing 20 AI-training contracts across five platforms in eight months, calling it "the new waiting tables," I went and looked at what that work involves. I also tested the voice-cloning tools on the other side of this trade — the ones built from creative professionals' work. Cloning a convincing voice from a short sample takes minutes, not months. That gap, between how fast the technology moves and how slowly protections catch up, is the whole story here.
What's actually happening in Hollywood right now
Production work has been shrinking for years. Motion picture and video production shed about 49,000 jobs over the decade ending February 2026, a 21% decline, according to the Bureau of Labor Statistics. On-location shoot days in the Greater Los Angeles area fell again in the third quarter of 2025 alone, down 13.2% year over year, with television shoot days down 20.7%, per FilmLA's regional production reports. At the same time, studios are leaning harder on generative tools: Netflix disclosed in its Q2 2026 shareholder letter that GenAI workflows touched roughly 300 of its titles that year, concentrated in post-production work like crowd generation and de-aging.
Those two trends meet in the same person. A screenwriter whose script work has dried up, or an editor whose post-production hours got cut, is often the exact profile an AI lab wants for reinforcement learning from human feedback (RLHF). That's someone who can tell a mediocre AI-written scene from a good one. Someone who can judge whether a synthesized voice sounds human. The pay is real. So is the discomfort. Fowler put it plainly: the job of correcting and rating AI writing output amounts to teaching a system how to do the job you used to have.
How Hollywood creatives are training AI, step by step
1. Rating and rewriting AI output on gig platforms
The most common path is RLHF contract work through platforms like Mercor, Outlier, Turing, Handshake, and Micro1. Tasks range from rating chatbot answers for tone and accuracy to rewriting AI-drafted scenes so they read like a professional wrote them. Fowler described doing 20 such contracts in eight months, work she took on specifically because entertainment income had dried up. Screenwriter Robin Palmer told CBS News she now spends 30 hours a week on Mercor evaluating AI-written fiction for structure, characterization, and transitions, work she compares to "coaching a developing writer."
2. Licensing voice or likeness through a union-backed agreement
SAG-AFTRA's 2026 contract requires studios to notify and bargain with the union before licensing a performer's voice, image, or performance to train an AI model. That closes the door on a studio quietly selling a background actor's scanned likeness to a third-party AI company without consent or compensation. If a studio skips that step, the union can push the dispute to arbitration, and damages there aren't capped at what the actor would have earned for the original performance.
3. Feeding real production work into a studio's own AI pipeline
This one doesn't always feel like "training AI" from the inside, but it is. When a VFX artist's crowd shot gets run through a generative tool for a Netflix production, or an editor's cut gets used to fine-tune an in-house model, that work is shaping the system, even though the credit reads the same as any other production job.
4. Learning to direct AI tools instead of being replaced by them
A smaller group is going the other direction: paying to get fluent in AI filmmaking tools so they can direct the output rather than compete with it. Courses like Curious Refuge's AI Filmmaking program run $749 per course, or $149 a month for full membership access, according to the vendor’s current pricing. Roughly 10,000 people have gone through its courses and workshops, and the school says about 95% of current students already work in entertainment or advertising.
5. Testing where your own voice or face ends up
Voice-cloning tools sit downstream of all of this. In my testing, ElevenLabs’ Starter plan ($6/month) includes instant voice cloning from a short sample, and the $22/month Creator plan adds professional-grade cloning built for higher-fidelity use. None of that is illegal or automatically unethical — plenty of actors use it under their own consent — but it's worth knowing how little friction stands between "I recorded a sample" and "a usable clone of my voice exists."
Example prompts you can copy
Before signing any AI-training gig contract or licensing agreement, run it through ChatGPT or Claude with prompts like these:
- "Review this AI-training gig contract and flag any clause that lets the company reuse my writing, voice, or likeness beyond this specific task."
- "Summarize in plain English what rights I'm giving up in this agreement, and whether there's any compensation if my work is used to train a commercial model later."
- "Compare this contract's payment terms — hourly rate, kill fees, ownership of output — against typical freelance or union gig terms and flag anything unusual."
- "List every place in this document where 'derivative works' or 'training data' appears, and explain what each clause actually allows."
None of these replace a lawyer or your union rep for anything with real money or rights attached, but they catch the obvious red flags before you sign.
Common mistakes to avoid
The biggest mistake I see: treating an AI-training gig's hourly rate as directly comparable to an old union rate. Gig platforms rarely offer residuals, pension contributions, or health credits the way a SAG-AFTRA production does. Second, signing broad IP-assignment language because the contract is framed as "quick freelance work." Read it the way you'd read a licensing deal, because that's often what it actually is. Third, uploading a voice or face sample to a tool without checking its data-retention policy first. Once a clone exists, you can't un-train it out of a model. And fourth, assuming a single verbal assurance from a hiring manager protects you the way a formal, bargained agreement does. When I tested how these gig postings are worded, the more reputable ones spelled out scope and reuse rights clearly. The vague ones were the ones worth walking away from.
Where the money and the control actually sit
| Path | What it actually is | Typical pay or cost | Who controls your work |
|---|---|---|---|
| RLHF gig work (Mercor, Outlier, Turing, Handshake) | Rating or rewriting AI output on contract | $12–$200+/hr by skill level, some specialists at $350/hr; no residuals | The platform, and whichever AI lab buys the data |
| Union-bargained AI licensing (SAG-AFTRA agreements) | Formal, consent-based licensing of your voice or performance | Negotiated fee per use, per contract | You, through your union — studio must notify and bargain first |
| Studio-side GenAI production use (e.g., Netflix's ~300-title 2026 slate) | Your production work reviewed or regenerated with in-house AI tools | Your normal production fee | The studio, under its own AI-use guidelines |
| AI upskilling courses (Curious Refuge) | Paying to learn AI filmmaking or screenwriting tools yourself | $749/course or $149/month | You — you're the customer, not the training data |
Tools that make this easier
If you're weighing whether to take an AI-training gig at all, my AI jobs apocalypse piece and what’s actually happening to jobs breakdown cover the broader employment picture beyond entertainment, which helps put any single gig platform's pay in context. For finding legitimate freelance and gig work without wading through scam postings, my best AI tool for job searching guide covers the tools worth using. If your income is shifting toward writing or editing AI output rather than original scripts, my best AI writing tools roundup is a good next stop for judging which tools are actually worth learning. And if you're an artist worried about your own work being mistaken for (or scraped into) AI output, when online commenters detect your art as AI covers that flip side directly. Before trusting any AI tool with your voice, likeness, or unpublished work, my AI tool ratings guide explains what to check in a vendor's data policy before you sign up.
My take
Taking AI-training gig work isn't a betrayal of your craft. It isn't free money either. It's a real, if uneven, income option in an industry where shoot days and production jobs have both been shrinking for years. The people who read the contract first, and know exactly what rights they're licensing, tend to come out fine. The ones who get burned treated it as just another gig-app task. It usually isn't. It's licensing your voice, your writing style, or your judgment to train a system that competes with you later. Read the contract like it's a licensing deal, because it is one.
Frequently Asked Questions
Is it legal for a studio to use my performance to train an AI without asking?
Under SAG-AFTRA's 2026 agreement, no — a signatory studio has to notify and bargain with the union before licensing a performer's voice, image, or performance for AI training. Non-union or independent work may not carry the same protection, so check your specific contract's language on "derivative works" and "training data" before you sign.
How much does AI-training gig work actually pay?
It varies widely by platform and skill level. Mercor’s own published rates run from $12–$25/hour for entry-level annotation up to $75–$200+/hour for specialized expert work, and CBS News reported in May 2026 that some Mercor contributors, including specialists, were billing as much as $350/hour. There's no standard rate the way there is on a union call sheet, and the work is inconsistent week to week.
What's the easiest way to check if a contract lets a company reuse my work?
Search the document for phrases like "derivative works," "training data," "perpetual license," and "third party," then ask an AI assistant like ChatGPT or Claude to summarize what rights each clause grants in plain English before you sign anything.
Do I need a union to have any protection here?
No, but it helps significantly. SAG-AFTRA membership brings a bargained AI framework with notice requirements and arbitration; without it, you're relying entirely on whatever a given platform's or studio's contract says, which varies a lot and is worth reading closely every time.