How to Get Paid to Train AI

💰 $0 to start ⏱ First dollar: 1–3 weeks, if approved 🤖 Write and rate for AI

Updated

The 30-second version

AI companies pay contractors to write prompts, grade chatbot answers and fix their mistakes. You apply on a platform like DataAnnotation, Outlier, Alignerr or Mercor, pass an unpaid test, then take paid tasks when there are any. Advertised rates run from about $15–$50 an hour for generalists to $40–$200 an hour for most coding and credentialed-expert tracks. The rate is real; the hours are not guaranteed, and passing the test does not mean you will get work.
Startup cost
$0 (you need a computer and internet)
Advertised pay
$15–$50/hr generalist, $40–$200/hr expert
First dollar in
1–3 weeks if approved; never if no work shows up

▶Watch this first

"40 Hours of Data Annotation on Outlier AI || How Much Can You Make?" — Peterson Hustle. Picked from a real creator, not a GigWiz production.

✓What you actually need to start

Legit platforms never charge you to join, buy training, or unlock tasks.

👣The steps

Pick your laneGeneralist writing and rating is the easiest door but the lowest pay. If you code, have a STEM degree, or work in law, medicine or finance, apply to those tracks first; their advertised top rates are often double or more.
Apply on the official sites onlyGo directly to dataannotation.tech, outlier.ai, alignerr.com, mercor.com or prolific.com. Ignore texts, WhatsApp or Telegram “recruiters” offering AI-training jobs.
Treat the assessment like an interviewThe entry test is unpaid. Read instructions twice, write carefully, and check facts. DataAnnotation says approval usually comes within a few days of its Starter Assessment; some people never hear back.
Apply to two or three platformsWork on any one platform comes and goes with client projects. Spreading out is the only real hedge against an empty queue.
Protect your quality scoreDataAnnotation says the work you see depends on demand for your skills and your past performance. Rushing early tasks can cost you access later.
Track hours and set aside taxesLog paid and unpaid time so you know your real hourly rate. If you expect to owe $1,000 or more in tax, the IRS expects quarterly estimated payments.

💵The money, honestly

TrackAdvertised rate (platform pages)What it looks like at 10 hrs/week
Generalist (writing, rating)$15–$50/hr (Outlier Polish-language roles up to $15; DataAnnotation $25–$50; Mercor puts entry work at $12–$25)$150–$500/week gross, if work is there
Coding$23–$150+/hr (Outlier technical docs up to $23; DataAnnotation $40–$150+)$230–$1,500/week gross
Math / STEM$40–$125+/hr (Outlier math up to $50; DataAnnotation STEM $40–$125+)$400–$1,250/week gross
Law, medicine, finance experts$40–$200+/hr (DataAnnotation $40–$125+; Mercor $75–$200+)$400–$2,000/week gross

Rates are what the platforms advertise on their own pages, and every one says the actual rate depends on the project, your skills and location. Outlier says lower rates apply during onboarding. Independent reporting is lower: TIME found U.S. labeling gigs typically around $20/hr, and Business Insider described an Outlier worker whose effective rate fell from $50 to about $20 an hour.

Out of revenue: 15.3% self-employment tax on 92.35% of net earnings, plus federal and state income tax. Example: $250 for 10 hours at $25 leaves about $215 after self-employment tax, before income tax. Unpaid assessments and onboarding time also drag your true hourly rate down.

⚠ The reality check

  • Passing the test does not mean you get work. TIME reported workers who passed assessments and were never offered tasks. Platforms don't publish pass rates or how many approved workers actually get projects.
  • Projects dry up, sometimes fast. Business Insider reported Scale AI, which runs Outlier, lost OpenAI and Google work after Meta's 2025 investment, and activity among Outlier taskers dropped. The New York Times, via Moneywise, reported that once experts teach a model a skill, projects in that specialty dry up.
  • The hourly rate only covers task time. Tests, onboarding and waiting are mostly unpaid. Business Insider described one tasker who spent close to 40 hours in a month in unpaid onboarding without landing any paid work.
  • Accounts can close with no clear explanation. TIME reported users whose accounts were deactivated, including one claiming $2,869 of work unpaid, and support emails that went unanswered. Cash out often so less is at risk.

❔Questions people actually ask

How much do AI trainers really make?Advertised rates are $15–$50 an hour for generalist work and $40–$200 an hour for most coding and credentialed-expert tracks. What you actually earn depends on how many hours of work you get, which can be zero. One DataAnnotation worker told TIME they made $400–$500 a week part-time; others passed the test and got nothing.
Is DataAnnotation or Outlier a scam?Both are real and pay real money, but work is unpredictable. The scams are the copycats: anyone who asks you to pay a fee, buy training or deposit crypto to unlock tasks. DataAnnotation says it will never ask you for money. The FTC says reported job scam losses grew from $90 million in 2020 to $501 million in 2024.
Do I need a degree or coding skills?Not for generalist work. DataAnnotation says no coding background is needed for its general track. The better-paid tracks do want proof: Outlier's math work asks for a bachelor's in math or a related field (current students can apply).

The GigWiz verdict

✓ Good if
  • You have a degree, coding skills or professional credentials and want flexible work that uses them.
  • You want a $0-to-start side income and can live with weeks of no work.
The numbers
Startup cost$0
Advertised pay$15–$200/hr by track
First dollar in1–3 weeks if approved
DifficultyEasy to apply, hard to get steady work
⚠ The catch

The hourly rate is real, but nobody promises you hours: projects end, queues empty, and you find out when it happens.

Sources: DataAnnotation FAQs; Outlier math expert page; Outlier Polish-language role page; Mercor: AI trainer hourly rates; TIME: Side hustle or scam? Data annotation work; Business Insider (via AOL): Inside the fallout from Meta's Scale AI deal; Moneywise: White-collar workers paid to train AI; IRS: Estimated taxes; IRS: Self-employment tax; FTC: 2024 fraud loss data.

Spot a wrong number, a dead link, or a claim that's aged badly on this guide? ⚠ Found a mistake? Let us know