AI Training Jobs 2026: 8 Platforms Ranked by Real Pay

Start cost$0

TimeA few hours a week to 10+ hours a week

First moneyAbout a week after you are accepted

RiskSome risk

IncomeComing soon

Platforms compared

Fee to join a real one

$8/hr

Prolific's pay floor

Oct 5

Rates checked

What AI training work actually is

Chatbots learn partly from people. Companies hire remote contractors to write questions for a model, compare two answers and pick the better one, explain what is wrong with a reply, fix code the model wrote, or check that a medical or legal answer is correct. You may also see it called data annotation, RLHF work (reinforcement learning from human feedback), AI rating or AI tutoring.

It is contract work, not a job. You are an independent contractor, you pay your own taxes, and the platform decides which projects you see. Most platforms give you an unpaid entry test first. If you pass, you pick tasks from a dashboard when they exist. When a project ends, your work can stop overnight.

There are two kinds of work. Generalist work asks for strong English, careful reading and good judgment. Expert work asks for a degree, a license or real professional experience in coding, math, science, medicine, law or finance, and it pays more per hour.

How we checked these numbers

Every pay figure comes from the platform's own page or from two independent sources, read between September and October 2026. Where the platform's advertised rate and what workers report do not match, we show both. Rates are per hour or per task as published. We do not quote monthly totals because hours depend on how much work is on offer. Read our methodology.

AI training platforms compared

Ranked for a reader who wants to start without special credentials first, then the expert platforms. The pay column shows the advertised rate, then what reviewers and workers report in 2026.

Platform	Type of work	Published pay range	Who it is for	Payout	Getting in	The catch
1. DataAnnotation	Writing prompts, rating answers, coding checks	Site: $25 to $50/hr generalist, $40 to $150+ coding. Reviews: $14 to $20 general, $25 to $45 coding	Generalists and experts	PayPal, after a 7 day hold	Hard: one unpaid test, often no reply	Work can dry up; misclassification suit pending
2. Outlier	Rating, writing, coding and expert tasks	Site: about $25/hr average. Reviews: $12 to $45, mostly $18 to $28	Generalists and experts	Weekly on Tuesdays: PayPal, ACH or Airtm	Medium: unpaid onboarding and quizzes	Reported pay cuts since 2025
3. Prolific	Paid research studies plus invite only AI tasks	$8/hr floor on studies. AI tasks: up to €25/hr general, up to $150/hr for ML specialists	Anyone 18+, experts for AI tasks	PayPal from $6	Waitlist, then a short skills test	The waitlist can last months
4. TELUS Digital AI Community	Search and ad rating, annotation, language tasks	About $11 to $20/hr in the US	Generalists, bilingual people	PayPal, Payoneer or Hyperwallet, the following month	Medium: qualification exam	Slow pay, often capped near 20 hrs a week
5. Alignerr	Expert evaluation, coding, some language and voice work	Site: $40 to $120/hr most roles. Reviews: $15 to $60	Mostly experts, some open roles	Weekly; PayPal, Payoneer or ACH reported	Hard: unpaid evaluation phases	Months between projects for some
6. Handshake AI	Expert prompts and grading for AI labs	$40 to $125/hr stated; listings cluster $75 to $100	Students, grads and grad level experts in the US	Weekly via Stripe	Hard: ID check, background check, assessment	Pay disputes and a 2026 lawsuit
7. micro1	Expert AI training contracts	Reviews: $22 to $40 typical, $50 to $150+ for specialists	Professionals and PhDs	Deel, twice a month	Hard: AI video interview	Rejections come with no reason
8. Mercor	Expert AI training contracts	Roles posted from $60 to $250/hr; general tasks reported from about $20	Credentialed experts	Weekly via Stripe or Wise	Hard: AI interview plus tests	Headline rate swings; projects end

Sources: each platform's own site (DataAnnotation, Outlier, Prolific, Alignerr, Handshake, micro1) and independent 2026 reviews named in each card below. Scroll the table sideways on a phone.

Each platform, in plain terms

1. DataAnnotation

Best for generalists

The best known generalist platform. You write prompts, judge chatbot answers and explain your reasoning. Its parent company is Surge Labs (Surge AI).

Published pay: the site lists $25 to $50 an hour for generalists, $25 to $40 for language roles, $40 to $150+ for coding and data roles, and $40 to $125+ for finance, legal, medical and other specialist roles (dataannotation.tech, read October 2026).
What workers report: independent 2026 reviews put general work at about $14 to $20 an hour and coding at $25 to $45 (theairankings.com; talentsforai.com). Treat the site's numbers as the top of the range.
Payout: PayPal. Earnings sit for about 7 days, then you can move them (paidtaskers.app; secondtalent.com).
Who can join: 18+, in the US, UK, Canada, Australia or New Zealand. Some reviews add Ireland.
The catch: the starter assessment is unpaid, takes about an hour, and many people never hear back. Work can stop with no notice. A worker misclassification class action against Surge was filed in May 2025 in San Francisco Superior Court (Bloomberg Law; Clarkson Law Firm). We found no ruling as of October 2026.

Visit DataAnnotation →

2. Outlier

Most projects

Outlier is the contributor platform of Scale AI. It runs generalist writing and rating projects plus coding and subject expert work.

Published pay: Outlier's English generalist page advertises an average of about $25 an hour (outlier.ai, read October 2026).
What workers report: $12 to $45 an hour depending on the task, with most standard tasks at $18 to $28 (breakingeven.online 2026; aitrainer.work 2026). Pay is per task, so your speed changes your real hourly rate.
Payout: weekly on Tuesdays through PayPal, ACH or Airtm, depending on your country.
The catch: after Meta's 2025 investment in Scale AI, Scale laid off about 14% of its staff and ended contracts with 500 contractors in July 2025 (TechRepublic; UPI). Workers report effective pay falling from roughly $40 to $50 an hour to about $15 to $20 on many projects. Separately, Scale agreed to a $12.5 million settlement in 2026 over claims that it misclassified California contributors (openclassactions.com; King Siegel). Onboarding courses are often unpaid.

Visit Outlier →

3. Prolific (studies and AI tasks)

Most reliable payouts

Prolific is a research study platform. Most studies are academic surveys, but AI companies also use it, and it invites people with the right background to longer AI training tasks.

Published pay: researchers must pay at least $8 (£6) an hour, and Prolific recommends $12 (£9) (Prolific researcher help center). For AI tasks, Prolific's own recruiting posts say researchers tend to pay up to €25 an hour for general AI training work and up to $150 an hour for machine learning specialists.
Payout: cash to PayPal once you have $6 (£6). After your first four cashouts, transfers are instant (prolific.com/participants).
Getting in: join a waitlist. Invites depend on which demographics researchers need, so it can take hours or months. AI tasks add a 10 to 15 minute skills test.
The catch: study slots fill in minutes, and you can't control how many you see.

Visit Prolific →

4. TELUS Digital AI Community

Steady, lower pay

The old search engine and ad rater work, plus annotation and language projects. It has been around longer than the chatbot boom.

Pay: most US workers report about $11 to $20 an hour, lower in other countries (selfmadesuccess.com, June 2026). ZipRecruiter put the typical US rater wage near $17 an hour in mid 2026. Some expert projects pay $20 to $25.
Payout: PayPal, Payoneer or Hyperwallet, paid the month after you work.
Getting in: apply to a listed project, then pass a qualification exam, compliance checks and ID verification.
The catch: many projects have an informal cap of about 20 hours a week, and the exam study material can be long.

Visit TELUS Digital AI Community →

5. Alignerr

Experts and coders

Alignerr is run by Labelbox. It recruits PhDs, researchers and working professionals, with a few roles that need no experience.

Published pay: $40 to $120 an hour for most roles, an advertised average of $80, and $25 to $45 for language roles (alignerr.com, read October 2026).
What workers report: $15 to $60 an hour, averaging around $37 (breakingeven.online; annotation.academy 2026). Some projects pay per approved task instead of per hour.
Payout: weekly. Reviewers report PayPal, Payoneer and ACH.
The catch: evaluation phases are often unpaid, some workers go months between projects, and on per task projects a rejected task pays nothing.

Visit Alignerr →

6. Handshake AI

Students and grads

Handshake, the college job site, runs an AI training program (once called MOVE) for students, graduates and researchers who write and grade expert level prompts.

Pay: about $40 to $125 an hour stated, with tracked listings clustering at $75 to $100 (aitrainer.work, October 2026). University career offices have posted the program at $50 to $125 an hour.
Payout: weekly through Stripe for approved hours.
Getting in: US based with work authorization, plus ID verification, a background check and a role assessment. Expect weeks, not days.
The catch: workers have reported deposits below their logged hours, and a proposed class action, Boggs v. Stryder Corp., was filed in federal court in California on July 20, 2026, alleging misclassification (aitrainer.work; breakingeven.online). Those are allegations, not findings.

Visit Handshake AI →

7. micro1

Professionals

micro1 matches vetted experts in engineering, finance, health, law and languages with AI labs, from early career to PhD level.

Pay: micro1 does not publish rates on its expert page. Reviews report $22 to $40 an hour as typical and $50 to $150+ for specialists (aitrainer.work; annotation.academy 2026).
Payout: through Deel, twice a month, in US dollars.
Getting in: apply to a role, then do a 20 to 30 minute webcam interview with micro1's AI interviewer, Zara.
The catch: rejected applicants get no reason, and some reviewers dislike the monitoring software used during work.

Visit micro1 →

8. Mercor

Highest posted rates

Mercor hires doctors, lawyers, bankers, engineers and PhDs to train and test AI models for large labs.

Pay: individual roles are posted from about $60 to $250 an hour. Mercor's own "average contracted rate" has moved between $80 and $141 an hour during 2026 and sat near $108 in late September (usagepricing.com tracking; herohunt.ai). Reviewers report general evaluation work from about $20 to $30 an hour.
Payout: weekly through Stripe or Wise.
Getting in: an AI interview of about 20 to 30 minutes, then role tests.
The catch: most roles need real credentials, and projects can end with little warning.

Visit Mercor →

How to get accepted

The entry test decides almost everything. Your resume matters much less than how carefully you work during it.

Apply to three or four at once. Acceptance is slow and silent at most platforms, so don't wait on one.
Take the test when you have a quiet hour. DataAnnotation gives you one attempt with no retake. Read every instruction twice.
Explain your reasoning. Graders want to see why one answer beats another. Specific, short, correct explanations win over long ones.
Fact check everything. If a task asks you to judge a claim, look it up. Catching a confident wrong answer is the core skill.
Write clean English. Spelling, grammar and following the format exactly count. Many tests are failed on format alone.
List your real expertise. Coding languages, degrees, licenses and second languages unlock higher paying projects. Never claim skills you lack; tests check.
Do the first tasks slowly. Early quality scores decide which projects you see next.

Red flags and scams

Fake "AI training" and "task" jobs are everywhere. The FTC says reported losses to job scams passed $220 million in the first half of 2024, driven by task scams, with about 20,000 task scam reports in those six months versus about 5,000 in all of 2023 (FTC Data Spotlight, December 2024).

Walk away if you see any of these

A recruiter messages you first on Telegram, WhatsApp or a text. You are asked to pay to apply, buy a starter kit, pay for training or "unlock" your earnings. Your balance grows in an app, but you must deposit crypto before you can withdraw. Someone sends you a check and asks you to send part of it back. A job promises a fixed weekly income for simple clicking or rating.

Real platforms never charge you. DataAnnotation, Outlier, Prolific and the rest are free to join. Paying to get paid is the clearest sign of a scam.
Apply on the official site only. Type the address yourself. Scammers copy platform names and logos in direct messages.
Protect your ID. Real platforms verify identity inside their own app or through a known provider, never by a photo of your ID sent over chat.
Report it. Report job and task scams to the FTC at ReportFraud.ftc.gov. The FTC's alert on task scams shows how they work.

The honest downsides

Unpaid assessments. Most entry tests and many onboarding courses pay nothing, and some take hours.
Silence is common. Many applicants never hear back, and platforms rarely explain a rejection.
Work dries up. Projects end when a client's contract ends. Your dashboard can go from full to empty overnight, and accounts are sometimes removed with no warning.
Rejected tasks can pay $0. On per task projects, work a reviewer rejects may not be paid, and time spent on it is lost.
Pay rates change. Workers report Outlier pay falling on many projects since 2025, and Mercor's headline rate has swung by tens of dollars in 2026.
You are a contractor. No benefits, no overtime and no paid breaks. That is exactly what the lawsuits against Surge, Scale and Handshake are about; the Surge and Handshake cases are still open.

Taxes

This is self employment income, so you may owe income tax and self employment tax, and platforms may send a 1099. Our tax guide covers how side income is reported.

AI training jobs FAQ

Are AI training jobs legit?

Yes, the platforms on this page are real companies that pay contractors. But no one is promised work, entry tests are usually unpaid, and pay depends on how many tasks are available. Fake AI training jobs offered by message on Telegram or WhatsApp are scams.

How much do AI training jobs pay?

Published rates run from Prolific's $8 an hour study floor to $25 to $50 an hour for DataAnnotation generalists, and up to $150 to $250 an hour for credentialed experts on Mercor or Alignerr. Workers often report less than the advertised rate, for example about $14 to $20 an hour for general DataAnnotation work.

Do I need a degree to train AI?

No. DataAnnotation, Outlier, Prolific and TELUS Digital all have generalist work that needs strong English and careful judgment. Degrees, licenses and coding skills unlock the higher paying expert projects.

Do you have to pay to start AI training work?

No. Legitimate AI training platforms never charge a fee to apply, train or withdraw your pay. Anyone asking for money or a crypto deposit is running a scam.

How fast do AI training platforms pay?

Once accepted, most pay weekly. DataAnnotation holds earnings about 7 days before PayPal transfer, Outlier pays on Tuesdays, Prolific pays to PayPal from $6, and TELUS Digital pays the following month. Getting accepted can take weeks or never happen.

Why did my AI training work suddenly stop?

Projects end when a client's contract ends, and platforms move workers based on quality scores. It is common for a dashboard to go empty for days or months. Apply to several platforms so one slow period does not stop all your work.

Pay rates and policies change often. Figures were checked between September and October 2026 against each platform's site and independent reviews, and may be out of date when you read this. Nothing here is an offer of work or a promise of income. Results vary. See our affiliate disclosure.

AI Training Jobs,Ranked for 2026