AI training gigs: get paid to train AI on your own schedule
AI training gigs are flexible, contract-based tasks that pay $15–$150/hr. Here's what gigs exist, what they pay, and how to land your first one.
An AI training gig is a short, contract-based assignment where you help teach an artificial intelligence model to be more useful, accurate, or safe — and get paid for it on your own time. There is no boss, no fixed shift, and no minimum number of hours. You log in when you want, pick up tasks from a queue, and get paid weekly for the work you complete. For experts with a real specialism, it is one of the most flexible ways to earn serious money from home in 2026.
This guide explains what AI training gigs actually look like, who is buying the work, what the gigs pay at each level, and how to land your first one. If you want the full picture of this kind of work first, start with what AI training work is; if you are starting with zero experience, read our entry-level guide alongside this one.
Gigs vs jobs: why the distinction matters
Most people search for "AI training jobs" expecting employment. In practice, the vast majority of this work is gig work: you sign on as an independent contractor, take tasks when they are available, and stop whenever you like. That is a feature, not a bug. AI labs do not need a standing army of annotators; they need bursts of human judgement on specific models, in specific domains, for specific weeks. The gig model matches that demand perfectly.
What this means in practice:
- You set the hours. Platforms describe the work as fully remote and asynchronous. Data Annotation's FAQ describes it as a flexible, task-based contractor role; Alignerr's listings say you control when and how much you work, from 5 to 40 hours a week depending on your availability.
- There is no guaranteed volume. Work arrives in projects. A project can run for months or end with little notice — reviewers report projects closing on the lab's timeline, not yours. Gigs are feast-and-famine by design.
- You are a contractor, not an employee. In the US that usually means 1099 status: no benefits, and you handle your own taxes. (We will be publishing a dedicated tax guide for AI trainers.)
- Pay is by the task or the hour, per project. A listed rate is not a promise of income. As Data Annotation puts it, the work you see depends on demand for your skills and on your past performance.
If you want predictable income, a gig alone will not give it to you. If you want a second income stream you can turn on and off around a job, a degree, or a family, few things beat it.
What the gigs actually look like
AI training gigs fall into a handful of task types. Different platforms skew toward different ones, and your background decides which pay you the most.
Response rating and ranking. The bread-and-butter gig. You read two or more AI-written answers to the same prompt and pick the better one, or score a single answer against a rubric for accuracy, clarity, and safety. Generalists can start here with no special credentials.
Prompt writing. You write questions, instructions, and edge-case scenarios for the model to attempt. Good prompt writers think adversarially: what would a real user ask that might trip the model up?
Expert review. This is where the money is. A doctor evaluates medical answers, a lawyer checks legal reasoning, an engineer reviews code. Platforms pay a large premium for credentials because a wrong answer in these domains is expensive for the lab. See our profession-specific guides for doctors and nurses, lawyers, finance professionals, and software engineers.
Red teaming. You deliberately try to make the model misbehave — produce harmful, biased, or leaking outputs — so the lab can patch the weakness. Creative, unusual, and well paid when it appears.
Data annotation and labelling. The original gig category: tagging images, transcribing audio, classifying text. Lower paid than expert review, but the easiest on-ramp. Our data annotation jobs hub covers this corner of the market in depth.
What AI training gigs pay in 2026
Pay varies far more by task and platform than by anything else, so think in tiers rather than averages. All figures below are drawn from platforms' own listings and recent independent reviews (September–October 2026), and they describe what the platform advertises — your actual earnings depend on the projects you get matched to.
Entry-level gigs: roughly $15–$30 per hour. Basic response rating, simple annotation, and generalist tasks on platforms like Outlier and Data Annotation sit here. Data Annotation's own FAQ lists $20 or more per hour for multilingual projects and $25–$30 for general work; Outlier's generalist track is reported around $25–$30 an hour.
Skilled gigs: roughly $30–$75 per hour. Coding review, STEM problem-writing, and multilingual specialist work. Data Annotation lists $50–$100 for coding, STEM, and professional projects that need a degree or licence; Outlier's STEM raters are reported at $50–$60 an hour.
Expert gigs: $75–$150+ per hour. Licensed physicians, senior engineers, data scientists. Data Annotation's September 2026 job board shows $60–$125+ for physicians and $75–$150+ for software engineers and data scientists. Alignerr's current listings span $15–$150 per hour depending on domain and task complexity. Mercor lists roles like a financial-crime analyst gig at $75–$100 an hour.
A few important caveats. First, these are advertised or reported rates, not guarantees: project-based work means your effective monthly income swings with availability. Second, per-task arrangements usually pay less per effective hour than hourly ones — always clarify the pay structure before committing time. Third, higher rates concentrate on platforms that vet for expertise; the open-door platforms pay less but accept almost anyone who passes their assessments.
Where the gigs come from
A handful of platforms dominate the market, and they work differently:
- Outlier (operated by Scale AI). Large project marketplace with generalist and specialist tracks. Reviewers report $25–$60 an hour depending on track, weekly PayPal payouts, and an onboarding skill exam that takes two to six unpaid hours. Projects can end abruptly.
- Data Annotation. Task-based and flexible, paid through PayPal. Known for a famously strict starter assessment and for paying coding and STEM work well above general tasks.
- Mercor. Matches experts to AI labs through an AI-led video interview (about 15 minutes) plus a written assessment. Fully remote and asynchronous, weekly payouts via Stripe or Wise. Strongest on professional and technical roles.
- Alignerr. Works with major AI firms and lists expert contributor roles from $15 to $150 an hour with 5–40 flexible hours a week, paid weekly.
We compare several of these head to head — see Mercor vs Outlier and Mercor vs Micro1 — and our pay rates by field guide breaks down what different professions can expect.
How to land your first AI training gig
The application process is closer to a driving test than a job interview: platforms screen for skill, not CV polish. Here is the usual sequence.
- Pick a platform that matches your level. If you have a degree, licence, or professional background, aim at the expert-facing platforms first — their rates are multiples of the generalist ones. If not, start with a generalist platform to build a track record.
- Apply and complete the screening. Expect an onboarding quiz, a written assessment, or an AI-led interview. Higher screening scores unlock higher-paying task batches, so take the assessments seriously — they are the single highest-leverage hours you will spend.
- Start with lower-complexity tasks. Your early approval rate and quality scores are your reputation on the platform. A strong early record unlocks premium projects; a sloppy start can lock you out of them.
- Follow the rubric exactly. Every project ships guidelines that tell you precisely what a good answer looks like. The workers who earn the most are not the cleverest — they are the most consistent at matching the rubric.
- Go multi-platform. Once approved on one platform, apply to others. No single platform has steady volume, so diversification is the closest thing to income stability this market offers.
The fine print: what nobody tells you upfront
Availability is the real pay ceiling. The difference between a $500 month and a $3,000 month is rarely the rate — it is whether projects were open in your domain that month. Treat gigs as variable income and never budget against the top rate.
Quality scores decide your future. Platforms quietly grade your work. Fall below the bar and tasks dry up or your account is deactivated; stay above it and you get first pick of new projects. Read every guideline twice.
Some gigs ask for a weekly commitment. "Flexible" usually means you choose which hours, not how many. Some project roles expect a minimum — Mercor listings, for example, sometimes ask for around 20 hours a week for the project's duration. Check before you accept.
Watch for scams. Legitimate platforms never charge you to apply, never ask for payment to unlock tasks, and never conduct hiring over messaging apps. If a "gig" asks for money upfront, it is not a gig — it is a scam. Our entry-level guide includes the full red-flag checklist.
You are training your possible replacement. It is the industry's open secret: the expert feedback you sell today makes the model better at your domain tomorrow. Most experts treat gigs as well-paid bridge income — excellent while it lasts, not a thirty-year career plan.
Is an AI training gig worth it for you?
It is worth it if you have expertise a lab will pay for, a few flexible hours a week, and realistic expectations about variability. A software engineer picking up evening review gigs at $75–$150 an hour can add thousands a month without changing jobs. A student doing generalist rating gigs at $20–$30 an hour gets genuinely useful income with zero commute and zero schedule.
It is not worth it if you need a fixed paycheque next month, if you cannot tolerate projects vanishing without warning, or if the assessment process sounds like a chore — because the assessments are the easy part, and the rubric-following is the job.
The market for human expertise in AI training is still growing fast — the category's biggest search terms are up roughly ninefold year over year — but it rewards the prepared. Pick the platform that fits your background, pass the screening properly, protect your quality scores, and spread yourself across more than one marketplace. Do that, and AI training gigs become what they are at their best: the most flexible high-skill side income on the internet.