AI Training Jobs: What They Pay and Where to Apply (2026)
AI training jobs are real, remote and often well paid: advertised rates usually run about $15–25 an hour for generalists and $40–150 for credentialed experts. They are also gig work with no guaranteed hours, so treat them as side income, apply to several platforms, and never pay anyone to get started.
Disclosure: Some links on this page are affiliate links. If you buy through them, we may earn a commission at no extra cost to you. It never changes our verdicts or rankings. How we make money.
AI labs pay people to train their models, and more people find out every month. US searches for “outlier ai” run about 74,000 a month, up 95% in a year (DataForSEO, October 2026). More applicants means tougher assessments and more competition for each project, so the best time to get accepted is before the crowd arrives.
The work is real. A chemistry PhD might be paid to catch a model’s mistake in a reaction mechanism; a retired English teacher might rate two chatbot answers for tone. This page gives you the honest version: what the work is, what it’s advertised to pay, which platforms to try, how to get accepted, and how to avoid the scammers who have noticed how much people want these jobs.
What is an AI training job?
AI companies improve their models with human feedback. You are given a prompt and one or more AI answers, and you:
- Rate and compare answers for accuracy, helpfulness and safety, following a long guideline document.
- Write prompts and ideal answers, especially hard ones the model gets wrong.
- Find and explain mistakes in code, math, science, law or medicine (expert work).
- Label data: images, audio, video or text, for computer vision and speech models.
- Test agents: give an AI a multi-step task and grade how it did.
Job titles vary: AI trainer, AI tutor, data annotator, rater, model evaluator, “expert contributor”. It is almost always remote, flexible and done as an independent contractor.
How much do AI training jobs pay?
Pay depends far more on what you know than on the platform. These are advertised or widely reported ranges, not earnings promises:
| Work type | Typical pay (USD) | Who gets it |
|---|---|---|
| Search and content rating | About $10–15/hr, often per task | Anyone who passes the test; rate set by country |
| Generalist writing and rating | About $15–25/hr | Strong writers, often degree holders |
| Coding evaluation | About $40–150/hr | Developers who pass timed coding tests |
| STEM experts (math, physics, chemistry, biology) | About $40–125/hr | Graduate students, PhDs |
| Law, medicine, finance | About $40–125/hr, occasionally more | Licensed or credentialed professionals |
| Other languages | About $20+/hr | Fluent speakers, varies by language |
Two things the ranges hide. First, you are paid for time worked, not time available: if the project queue is empty, you earn nothing. Second, rates are set per project and can drop when a project ends and a new one starts. In November 2025, for example, Mercor ended a large Meta project and offered workers a replacement at a lower hourly rate (AOL/Business Insider).
The best AI training job platforms
Apply to several. Work comes in waves, and the people who earn steadily usually have two or three platforms going.
| Platform | Best for | Advertised pay | Notes |
|---|---|---|---|
| DataAnnotation | Generalists and coders | $25–50+/hr general; $40–150+ coding, STEM, professional | One unpaid starter assessment (about an hour), no retakes; paid via PayPal. Run by Surge AI. Official FAQ |
| Outlier (Scale AI) | Graduates, coders, STEM, languages | Not published; generalists report roughly $15–35/hr | Weekly pay. Onboarding needs ID and a résumé. Frequent complaints about empty queues and removal from projects. |
| Mercor | Domain experts (PhDs, doctors, lawyers, bankers, engineers) | Wide range; expert projects often $50–150+/hr | AI video interview to apply. Pays weekly via Stripe or Wise. Payment docs |
| Handshake AI (MOVE fellowship) | Grad students, PhDs, recent grads | About $50–125/hr in listings | Part-time, about 5–20 hours a week. Recruits through the Handshake careers network. |
| Alignerr (Labelbox) | Experts and professionals | Roles listed from about $20–50 to $40–120/hr | Some projects pay only for work a reviewer accepts. |
| Turing | Software engineers | About $25–55/hr agentic coding; more for experts | Timed coding tests; some roles need 30+ hours a week. |
| Micro1 | Senior engineers and PhDs | About $50–200/hr in listings | Very selective; expert-only. |
| Invisible Technologies | Coders and specialists | Wide range by location and skill | Picked up clients that left Scale AI in 2025. |
| TELUS Digital AI Community | Search and ad rating, any background | About $9–15/hr in the US, often per task | Formerly TELUS International and Lionbridge. Steady but low pay. |
| Appen / CrowdGen, OneForma, Toloka, Welocalize | Beginners, other languages | Mostly about $9–20/hr | Easy to join, inconsistent work. |
| Prolific | Paid research studies plus some AI tasks | Study minimums are modest; AI expert roles up to about $25/hr | Good filler between projects. |
Some AI labs also hire “AI tutors” directly. These jobs appear on their careers pages, not on gig platforms.
Full reviews, with pay sources, payout schedules and complaints: Outlier, DataAnnotation, Alignerr, Mercor and Handshake AI. Current open roles and posted rates are on the AI gig pay tracker.
The shift you need to know about: experts are in, generalists are out
The easy money for generalists is shrinking. In 2025 xAI laid off about 500 generalist annotators and said it would grow its specialist team (TechCrunch). Scale AI closed a generalist contractor team, citing a shift toward “higher skill, expert data work”. After Meta bought 49% of Scale AI in June 2025, several big customers pulled back (TechRepublic).
Meanwhile expert networks grew fast. Handshake’s AI training business reportedly passed a $1B annual run-rate in 2026 (Dealroom).
The lesson: your degree, license or professional skill is the product. A nurse, an accountant or a Python developer earns several times what a generalist does for similar hours.
How to get accepted
- Lead with your strongest credential. List degrees, licenses, publications and languages clearly on your profile and résumé. These platforms filter on them.
- Treat the assessment like an exam. Most give you one attempt. Read the instructions twice, take your time, and write complete, specific answers. Sloppy grammar fails generalist tests.
- Don’t use AI to do the assessment. Platforms check for it, and it gets you rejected or banned.4. Apply in your real location. Use your own ID, phone number and bank account. VPNs and borrowed accounts end in bans.
- Read the guidelines properly once you’re in. Quality scores decide whether you keep getting work. The people who get removed are usually the ones who rushed.
- Expect silence. It is normal to wait weeks after applying, or to be accepted and then see no work for a while.
The downsides nobody puts in the ad
- No guaranteed hours. DataAnnotation, for instance, says plainly that there’s no minimum and no guarantee of work. Empty queues can last weeks.
- Unpaid onboarding. Assessments and some training are often unpaid.
- Sudden removals. Workers report being dropped from projects or deactivated with little explanation.
- Contractor status. No benefits, no paid time off. Misclassification lawsuits have been filed against Scale AI, Surge AI and Mercor, arguing workers should be employees. We found no rulings or settlements yet.
- Your data. You hand over ID and sometimes video and biometric checks. In 2026 Mercor suffered a data breach that led to several class actions (TechCrunch).
- The work can be grim. Safety projects may involve reading or rating disturbing content. You can usually opt out of these projects.
How to spot an AI training job scam
The “task scam” is everywhere. OpenAI has published research on one such network (OpenAI). It usually goes like this:
- An unsolicited message on WhatsApp, Telegram, SMS or LinkedIn offers easy money “rating AI” or “training algorithms”.
- A friendly “mentor” walks you through simple tasks, and you get small payouts that build trust.
- Then you’re told to deposit money, usually crypto, to unlock better tasks or withdraw your “earnings”. The money is gone.
The rules:
- Real platforms never charge you. As the FTC puts it, honest employers don’t make you pay to get a job.
- Apply only through the platform’s own website. Type the address yourself; don’t follow a recruiter’s link.
- Real platforms pay out through PayPal, Stripe, Wise or bank transfer. They don’t use crypto deposits or Telegram groups.
- “Earn $500 a day, no experience” is not an AI training job. It’s the scam.
Taxes (US)
You’re an independent contractor, so nobody withholds tax. Set aside a share of every payout for income tax and self-employment tax. For payments made in 2026, platforms generally only have to send a 1099-NEC once they pay you $2,000 or more in a year (it was $600). The income is taxable whether or not you get a form. Keep records, and talk to a tax professional if this becomes a meaningful part of your income.
Is it worth it?
As a flexible side income, yes, especially if you have a degree, a license or coding skills: few remote gigs pay $50+ an hour with no client hunting. As a full-time career, be careful. The work is uneven, the rules change overnight, and the work exists to make models need less human help over time.
If you want income you control, pair this with something you own. See AI side hustles and how to make money with AI. If you’re wondering whether your day job is safe, check how exposed it is to AI, or browse the jobs AI is least likely to replace. Your next step: pick two platforms from the table above and apply to both this week.
Tools in this guide
Questions people ask
Are AI training jobs legit?
The big platforms (Outlier, DataAnnotation, Mercor, Handshake AI, Alignerr, TELUS Digital and others) are real companies that pay real money. The scams are the copycats that message you on WhatsApp, Telegram or LinkedIn and later ask you to pay a fee or deposit to unlock tasks.
How much do AI training jobs pay?
Generalist rating and writing work typically pays about $15–25 an hour. DataAnnotation advertises $25–50+ for general work and $40–150+ for coding, STEM and professional work, and expert networks such as Mercor and Handshake AI advertise $50–125+ for PhDs, doctors, lawyers and engineers.
Do I need a degree for AI training jobs?
Not always, but it helps. Many generalist roles want at least a college degree or strong writing, and the best-paid projects now go to people with graduate degrees or professional credentials. Coding work is judged by tests rather than degrees.
Can I do AI training jobs with no experience?
Yes, for entry-level rating and writing projects, if you pass the assessments. There is no AI experience requirement; the skills tested are careful reading, clear writing, following detailed guidelines and subject knowledge.
Can I do AI training jobs outside the US?
Some platforms accept many countries, but rates are often set by location and some projects are US-only. You usually need ID and a bank or payment account in your own name in the country you apply from. Using a VPN to fake your location gets accounts banned.
Do I pay taxes on AI training income?
Yes. In the US these platforms treat you as an independent contractor, so you owe income tax and self-employment tax and should set money aside. You may get a 1099 form, but the income is taxable even if you don't.