The 10 Best AI Training and Data Annotation Jobs in 2026: Legit Sites Compared on Real Pay

AI companies pay people to write and rate chatbot responses, grade code, check math, and label data so their models get better. These remote gigs usually go by AI training, AI trainer, or data annotation jobs. The industry is big and growing fast: Mercor, Handshake, and micro1 each reported hundreds of millions to billions of dollars in annualized revenue in 2026 (TechCrunch), and a Scale AI–commissioned study estimated about 200,000 Americans earned money this way in 2024 (Oxford Economics).

That doesn't mean every platform is a good fit. Below, we compare the 10 most established options for US workers using each platform's own published pay, payout, and tax information, plus what workers report on Trustpilot and in the press.

Quick comparison

Trustpilot scores checked October 7, 2026. Several platforms actively invite reviews, which tends to lift scores, so read the negative reviews too.

How we chose these platforms

We included platforms that hire US-based workers, publish how they pay, and have operated long enough to have a track record. We excluded general freelance marketplaces (like Upwork or Toptal) that aren't focused on AI training.

For each one we checked the official job board, help center, and payment and tax pages, and looked at Trustpilot reviews and news coverage. We list advertised pay, not guaranteed pay. Keeper has no affiliate or referral relationship with any platform on this list.

What to expect before you apply

The work comes and goes. Projects can run for weeks or months, then end. Workers call an empty task list an "empty queue" or the "dash of death." Breaking Even, a newsletter that tracks AI-training subreddits, found that about a third of DataAnnotation subreddit posts in June 2026 were about work droughts.

Accounts can be removed without much explanation. Sudden deactivations are one of the most common complaints across Outlier, DataAnnotation, and Handshake AI reviews.

Expertise pays, and generalist work is shrinking. In late 2025, Scale AI cut generalist contractors in Dallas, citing "an industry shift toward higher skill, expert data work," and xAI laid off about 500 generalist tutors (TechCrunch). If you have a degree or professional experience in a technical field, now's your time to shine.

Assessments are often unpaid. Expect to spend one to several hours on tests, interviews, or training before you earn anything. Lawsuits against Scale AI and DataAnnotation's parent company both allege unpaid onboarding.

You're a contractor, not an employee. That means flexibility, but no benefits and no tax withholding. More on that below.

The 10 best AI training platforms in 2026

1. DataAnnotation – best for generalists who want on-demand payouts

Apply at DataAnnotation

DataAnnotation (often called DataAnnotation.tech or DA) is the worker-facing brand of Surge Labs, the AI data company founded by Edwin Chen.

Pay: According to its FAQ, general projects pay $25 to $50+ an hour, multilingual $20+, coding $40 to $150+, and STEM and professional (finance, legal, medical) $40 to $125+ (DataAnnotation FAQ). Projects pay hourly or per task.

Getting in: A starter assessment of about an hour, plus optional specialized assessments for coding, STEM, or professional work. No degree is required for general projects. Rejected applicants usually don't hear back (TIME).

Getting paid: PayPal only. You request a payout and it typically arrives within a few days.

What workers say: Workers praise the rates, flexibility, and reliable PayPal payments. TIME interviewed a worker earning about $400 to $500 a week. The biggest complaints are unexplained droughts and accounts that go quiet without notice.

Pros: Some of the best generalist rates; you choose when to cash out; no set schedule.
Cons: No feedback if you're rejected; work can disappear for weeks; a pending misclassification lawsuit alleges unpaid assessment time (Bloomberg Law).

Is DataAnnotation legit? Yes. It's a real company that pays through PayPal, and it says it will never ask you for money (Trust & Safety). Apply only through its official signup page.

2. Mercor – best for experts chasing the highest rates

Browse Mercor roles

Mercor matches contractors with AI labs for expert projects. It raised money at a $10 billion valuation in October 2025 (TechCrunch) and says it pays out more than $4 million a day at an average contracted rate of $108 an hour (company figures, unaudited).

Pay: Listings range from about $20 to $50 an hour for quality analyst work to $100 to $200 an hour for data science and specialized expert roles. Most roles pay hourly, tracked with screenshot-taking time-tracking software.

Getting in: A roughly 20-minute AI video interview, then ID verification and a background check. Some roles include a paid work trial. Mercor accepts US citizens, permanent residents, and some visa holders.

Getting paid: Weekly through Stripe, every Wednesday, for the previous Saturday through Friday. The first payout has a seven-day hold (Mercor docs).

What workers say: The rates for specialists are hard to beat, but projects can end suddenly. In November 2025, Forbes reported that a large Meta project ended abruptly and workers were offered a new one at $16 an hour instead of $21 (Forbes). Mercor disputed the reports.

Pros: High expert rates; weekly pay; clear tax documentation.
Cons: Screenshot monitoring while you work; projects and rates can change quickly; a March 2026 data breach exposed contractor IDs and Social Security numbers, followed by several lawsuits (Staffing Industry Analysts). It also faces pending misclassification suits.

Is Mercor legit? Yes, it's a well-funded company that pays weekly. Go in knowing the work is project-based and can end without much warning.

3. Handshake AI – best for college students and recent grads

Apply to Handshake AI

Handshake, the college career network, runs an AI training program (its "fellowship") that recruits students and graduates from associate's degree through PhD. Handshake says it has paid more than $100 million to over 750,000 fellows (company figures).

Pay: $40 to $125 an hour depending on the role (Handshake AI). Optional meetings, Slack time, and downtime aren't paid, and onboarding is paid only if the project terms say so (Handshake support).

Getting in: You must be based in the US with valid work authorization. Some F-1 students on CPT or OPT may qualify. Expect an identity check, training, and an assessment.

Getting paid: Weekly through Stripe. Payouts go out by Wednesday for the previous Monday through Sunday (Handshake support).

What workers say: It's one of the few options open to current undergrads at high hourly rates. But Business Insider reported in March 2026 that several contractors were suspended and not paid for completed work, with no appeal process (Business Insider via AOL). Handshake also faces federal and state misclassification lawsuits filed in 2026.

Pros: Open to students; high hourly rates; weekly pay.
Cons: Reports of suspensions with unpaid work; caps on paid time; pending lawsuits.

Is Handshake AI legit? Yes. It's run by an established company. Screenshot your hours and completed work in case of a pay dispute.

4. Outlier – best for getting started with no degree

See Outlier opportunities

Outlier is operated by Scale AI. Meta bought a 49% stake in Scale in June 2025, after which several major AI customers reportedly pulled back (TechCrunch). Outlier says it has more than 900,000 contributors and has paid out over $500 million.

Pay: Generalist work pays "up to $15/hr," US personal finance experts up to $90 an hour, and machine learning experts up to $150 (Outlier). You see the rate before each project.

Getting in: You need to be 18 or older and fluent in English. No degree is required for generalist work. You'll verify your ID and pass skill and project assessments.

Getting paid: Weekly on Tuesdays through PayPal, Airtm, or ACH (Outlier FAQ).

What workers say: On-time weekly pay is the most common praise. The most common complaints are empty queues and sudden deactivations without a human appeal. Account bans were the top topic on the Outlier subreddit in June 2026 (Breaking Even).

Pros: Low barrier to entry; weekly pay; expert tracks pay well.
Cons: Low generalist rates; frequent empty queues; deactivations. Scale AI agreed to a $12.5 million settlement with California contractors over misclassification claims (final approval hearing October 30, 2026) (King & Siegel).

Is Outlier legit? Yes. It's a real platform that pays weekly. Treat it as a way in, not a reliable income.

5. Alignerr – best for weekly pay across many role types

Apply to Alignerr

Alignerr is the contractor network of Labelbox, an AI data company. It says it pays out $1 million a week to more than 100,000 experts.

Pay: It advertises an $80 average hourly rate, with writing roles around $30 to $40, AI training and language $40 to $120, and trading $55 to $110. Some projects pay hourly, others per task, and some only for accepted tasks (Alignerr payment FAQ).

Getting in: Upload a resume or LinkedIn, complete an AI video interview, and pass assessments. ID verification happens once you're matched to a project.

Getting paid: Weekly, the Friday after each Monday-to-Sunday cycle.

What workers say: Reviews are mostly positive about the interview and pay speed. Complaints, including eight Better Business Bureau complaints in the past year, involve non-payment and deactivations (BBB).

Pros: Weekly pay; wide range of roles; clear US tax page.
Cons: Work is cyclical; pay-per-accepted-task projects mean rejected work earns nothing.

Is Alignerr legit? Yes. It's backed by an established company and publishes detailed payment and tax policies.

6. micro1 – best for licensed professionals

Browse micro1 roles

micro1 recruits doctors, attorneys, engineers, analysts, and other professionals for AI training. Its gross revenue run rate grew from $100 million to $500 million in eight months (TechCrunch).

Pay: Roles include hospitalists at $70 to $100 an hour, litigation attorneys at $80 to $150, QA engineers at $90 to $175, and equity research at $180 to $220. Medical coding and voice roles started around $20.

Getting in: An unpaid AI interview, which may be proctored with your webcam on and your screen shared. Passing doesn't guarantee immediate work.

Getting paid: Twice a month through Deel (micro1 payroll policy). Work from the 1st to the 15th is paid around the 20th.

What workers say: Trustpilot reviews are strongly positive, though most describe the interview rather than ongoing work.

Pros: Professional-level rates; many specialized roles; open to US workers.
Cons: Only two paydays a month; time tracking and proctoring; certification doesn't guarantee projects.

Is micro1 legit? Yes. It's a fast-growing, venture-backed company with published pay policies.

7. Snorkel Expert Community – best for advanced-degree experts

Join the Snorkel Expert Community

Snorkel AI recruits people with advanced degrees, typically a master's or PhD for math and a JD for law, and pays them through an agent of record, HireArt.

Pay: Per accepted task, with the price shown up front. Snorkel says top experts earn $3,000+ a week, but it doesn't publish an hourly equivalent.

Getting in: Screening, assessments, and a background check covering identity, criminal history, and education. It's US-only, and California residents are excluded on several tracks.

Getting paid: Weekly for accepted work from the previous week.

Pros: Built for experts; upfront task prices.
Cons: Rejected work isn't paid; not open to Californians; requires an advanced degree.

Is Snorkel legit? Yes. Snorkel AI is an established AI company.

8. Prolific – best for short studies and quick tasks

Join Prolific

Prolific started as a platform for academic research studies and now also runs AI evaluation tasks. It's a UK company.

Pay: Researchers set the pay for each study. Prolific requires at least $8 an hour and recommends $12 (Prolific). Participants who pass AI skill assessments can access AI tasks.

Getting in: Join the waitlist, verify your ID, and connect a PayPal account that matches your name.

Getting paid: PayPal, with a $6 minimum cash-out and no Prolific fee.

What workers say: Highly rated for reliable payment. Common complaints are not enough studies and occasional bans or rejections without human review.

Pros: Easy, short tasks; fast cash-outs.
Cons: Lower pay than expert platforms; limited study availability.

Is Prolific legit? Yes. It's an established research platform with a large, well-reviewed participant base.

9. Mindrift – best for fixed-price tasks shown up front

Apply at Mindrift

Mindrift is the AI training platform of Toloka, a Netherlands-based company that raised $72 million in a round led by Bezos Expeditions in 2025 (Business Wire).

Pay: About $15 to $30 an hour for entry-level work and $60 to $100+ for specialists. Each task shows a fixed reward before you start.

Getting in: Experts in more than 90 fields can apply, and US workers are eligible. Expect an ID check.

Getting paid: Twice a month, on the 5th and 20th, through Payoneer or PayPal.

Pros: You know what a task pays before you start; wide range of fields.
Cons: Paid only twice a month; mixed reviews.

Is Mindrift legit? Yes, it's operated by an established, venture-backed company.

10. Turing – best for software engineers wanting longer contracts

See Turing jobs

Turing provides coding and technical experts to AI labs, including OpenAI, and raised money at a $2.2 billion valuation in 2025 (TechCrunch).

Pay: Not published on its listings. Watch out: the $100 to $1,000 figures on some Turing pages are referral bonuses, not pay.

Getting in: An unpaid timed assessment, then interviews. Many roles require at least 20 hours a week for 12 weeks and four hours of overlap with Pacific time.

Getting paid: Through Deel.

What workers say: Trustpilot complaints describe long waits in the talent pool after passing tests and delays withdrawing money.

Pros: Longer, more structured contracts; strong for experienced engineers.
Cons: No published rates; long waits for placement; set hour commitments.

Is Turing legit? Yes. It's a well-funded company, but expect a slower, more traditional hiring process.

Also worth knowing

• xAI AI Tutor roles. Elon Musk's xAI posts AI tutor jobs that can be full-time, part-time, or contract. Some eligible US roles come with health insurance and a 401(k), making them the rare AI training job that may come with employee benefits. A writing tutor listing offered $35 to $75 an hour (xAI careers). xAI cut about 500 generalist tutors in September 2025.

• Appen CrowdGen. One of the oldest names in the space, now paying monthly. Its non-China revenue fell 27% in the first half of 2026 (Investing.com), and reviews frequently mention onboarding with no work afterward.

• TELUS Digital AI (formerly Lionbridge AI). Hires US search and ad evaluators. It requires three years of US residency and allows one worker per household.

• Invisible Technologies (Meridial). Recruits professionals for project-based work under a statement of work and pays every two weeks.

• Remotasks. Scale AI's original platform, now focused on image and video labeling. Its generative AI work moved to Outlier.

How to avoid AI training job scams

Fake "AI rating" and "task" jobs are a fast-growing scam. Americans reported losing $501 million to job scams in 2024, up from $90 million in 2020, and task scams made up nearly 40% of job scam reports in the first half of 2024 (FTC).

The FTC's warning signs (FTC consumer alert):

• You were contacted out of the blue by text, WhatsApp, or Telegram.

• You're paid a few small amounts early, then asked to deposit money or crypto to "unlock" tasks or withdraw earnings.

• The job is rating or liking products to boost them.

Two more rules of thumb: apply only through the platform's own website (the links above), and don't share your Social Security number or bank details until you're inside a real platform account. Legitimate platforms collect your tax information (Form W-9) after you're onboarded. As the FTC puts it: never pay anyone to get paid.

Tips for getting accepted and finding more work

• Lead with your expertise. Mention degrees, licenses, coding languages, and professional experience. Expert tracks pay more and seem to have steadier demand.

• Take assessments seriously. Many platforms give you one shot. Read the instructions twice and give yourself uninterrupted time.

• Join two or three platforms. Droughts are normal. Spreading out protects your income.

• Keep your own records. Screenshot hours, task IDs, and payouts. They help if a payment is disputed, and you'll need them at tax time.

• Read each project's pay terms. Some projects pay only for accepted work, and some don't pay for training or meetings.

How taxes work for AI training jobs

Every platform on this list pays US workers as independent contractors. You owe income tax plus 15.3% self-employment tax on your profit, and if you'll owe $1,000 or more in taxes for the year, you're expected to make quarterly estimated payments.

Whether you get a tax form depends on how you're paid. For 2026 payments, a platform paying you directly only has to send a Form 1099-NEC once you've earned $2,000 (IRS). Mercor, Handshake AI, and Alignerr (for workers paid through Deel) say they issue a 1099-NEC.

If you're paid through PayPal, as on DataAnnotation and Prolific, you may only get a 1099-K above $20,000 and 200 transactions, or no form at all. Either way, all of it is taxable.

A good habit to have is to move 25% to 30% of every payout into a separate savings account. We advise opening a high-yield savings account (HYSA), where you'll be able to earn interest on those savings!

How to lower your taxes with write-offs

As an independent contractor, you can deduct the ordinary and necessary costs of doing the work. Every write-off lowers both your income tax and your self-employment tax. Common ones for AI trainers:

• 💻 Your computer and gear. A laptop, monitor, headset, or webcam you use for tasks. You deduct the business-use share, and items under $2,500 can usually be written off in full the year you buy them. See how to deduct a computer.

• 🌐 Internet and phone. Deduct your business-use percentage: if 25% of your internet use is task work, write off 25% of the bill. See how to deduct internet and your phone bill.

• 🏠 A home office. If you have a space you use only for work, the simplified method lets you deduct $5 per square foot, up to $1,500 a year.

• 🤖 Software and AI subscriptions. A paid AI model you need to compare outputs, a grammar checker, or coding tools. More on software deductions.

• 🎓 Courses that sharpen your skills. Training that improves skills you already use for the work, like a Python or writing course. See which education expenses are deductible.

• 💳 Payment fees. PayPal or transfer fees taken out of your payouts.

Want a full breakdown on how to handle your taxes? Check out Keeper's free tax guide for AI training and data annotation jobs.

Keeper helps you handle your income, expenses, and taxes all in one app. Try it free for 14 days.

Do AI training work? Share what it really pays with an anonymous review on Keeper's Side Hustle Index.

Pay, payout, and rating information was checked on October 7, 2026, and changes often. Confirm current terms on each platform before you apply.