Make Money Training AI (2026): Pay, Platforms and How to Get In

AI training platforms pay people to write, rate and correct chatbot answers, usually $15 to $50 an hour for general work. Coding and expert roles list $60 to $150 or more. DataAnnotation, Outlier, Alignerr and Mercor are the main names. All are legitimate but selective, and work can dry up without notice. Apply to several, take the entry tests seriously and never pay anyone for access.

How to choose

• Your skills. General roles want strong written English or another language, careful reasoning and fact-checking. The best-paid work goes to programmers, people with STEM degrees, lawyers, doctors, finance professionals and native speakers of less-covered languages. Mindrift says 70% of its AI trainers hold a master's degree or PhD.
• Startup cost. Nothing. Most platforms need only a computer, a stable connection, ID and sometimes a resume or LinkedIn profile.
• Time to first payment. Approval takes days to weeks. After that, most platforms pay weekly or on request: Outlier every Tuesday, Mercor every Wednesday, DataAnnotation within a few days of a request, Mindrift every two weeks.
• Where you live. Pay often depends on location. Outlier says rates vary by expertise, project complexity and location. Some platforms accept only certain countries, and availability changes, as when Remotasks left Kenya, Nigeria and Pakistan in 2024.
• Risk. Projects end abruptly, accounts can be paused without much explanation, and several platforms treat you as a contractor with no guaranteed hours. Plan for irregular income.

What the work actually is

• Rating and ranking: comparing two AI answers and explaining which is better and why.
• Writing: crafting prompts and ideal answers, often in a specific subject.
• Red-teaming: trying to get a model to produce unsafe or wrong output so it can be fixed.
• Coding: writing, reviewing and debugging code that trains coding models.
• Annotation: labelling images, audio, video or text. This is closer to classic microtask work and usually pays less.

Most tasks come with detailed guidelines, a time limit and quality reviews. Low review scores reduce the work you are offered.

General AI trainer platforms: DataAnnotation and Outlier

DataAnnotation: lists $25 to $50 an hour for general roles, $75 to $150 or more for coding and $75 to $125 or more for specialists. You take a Starter Assessment only once, and approval usually comes within a few days. It pays through PayPal, arriving within a few days of a request. It says you can work as much or as little as you want, but long-running projects go to workers who qualify and perform well.

Outlier: asks for ID, a resume and a LinkedIn profile, then runs skill screenings. It shows the rate before each project and pays weekly on Tuesdays by PayPal, Airtm or ACH bank transfer. Outlier says it has paid more than $500 million to experts. It is run by Scale AI, which absorbed the older Remotasks platform. Scale AI has faced wage and worker-classification lawsuits filed from December 2024.

Expert networks: Alignerr, Mercor, Mindrift and micro1

Alignerr: advertises an $80 an hour average. Listed roles range from $20 to $40 an hour for some language and policy work to $40 to $120 for prompt writing and image description. It pays weekly.

Mercor: lists a $112 an hour average contracted rate and more than $4 million in daily payouts. It pays hourly contractors every Wednesday through Stripe or Wise, depending on your country. Standard Stripe payouts are free and arrive within 5 business days, while instant payouts cost 1%.

Mindrift: lists $15 to $30 an hour at entry level and $60 to $100 or more for domain experts. It pays per completed task every two weeks and shows the amount before you start. Toloka's former worker sign-up page now redirects to Mindrift.

micro1: recruits experts for AI training and data projects and publishes interview preparation material for applicants.

Compare these in best AI training platforms.

Large crowd vendors: CrowdGen, TELUS Digital and RWS TrainAI

CrowdGen (Appen's crowd), TELUS Digital AI Community and RWS TrainAI hire raters, search evaluators and annotators in many countries and languages. Our CrowdGen profile lists coverage of 170+ countries with PayPal, Payoneer, Airtm and bank payouts. These vendors are often the only option for people outside the countries the expert networks favour, but they usually pay less, and many projects require a specific language or residency.

The low end: microtasks

Classic annotation on microtask sites pays far less. A CHI 2018 study of Mechanical Turk found a median of about $2 an hour. If a "data job" pays per tiny task with no hourly rate shown, work out your real hourly rate before committing. See Mechanical Turk and Prolific, which requires researchers to pay at least $8 an hour.

What you can realistically earn

Listed rates are ceilings on availability, not promises of hours. At DataAnnotation's $25 an hour floor for general roles, 10 hours a week comes to $250 before tax, but only if projects are open to you that week. Mercor's $112 average reflects experienced professionals, so a new generalist should not plan around it. Because these platforms treat you as a contractor, no tax is withheld, so set part of each payout aside. For people outside the US, the payout route matters too: our guides to receiving payments with Payoneer and Payoneer vs Wise explain the fees.

Compare the options

First steps this week

1. Update your LinkedIn profile and resume to highlight writing, subject expertise, languages and any coding.
2. Apply to three platforms at once, such as DataAnnotation, Outlier and one expert network that matches your field.
3. Treat each assessment like an exam: read every guideline, write in full sentences, check facts and do not rush. DataAnnotation's Starter Assessment can be taken only once.
4. Set up the payout method each platform uses, such as PayPal, Wise or a bank account in your name.
5. Once accepted, start with small batches and aim for top quality scores before taking on volume.
6. Keep a second income source, because projects can pause for weeks.

What to avoid

• Paying for access. No legitimate AI training platform charges an application, training or "unlock" fee. The FTC says reported job scam losses rose from $90 million in 2020 to $501 million in 2024, and task scams that ask for deposits are common.
• Recruiters on WhatsApp or Telegram. Real platforms recruit through their own sites and email domains. Check the URL before entering ID documents.
• Buying or renting accounts. Accounts are tied to your ID and location, and shared or resold accounts get banned with earnings withheld.
• Assuming stable work. The Washington Post reported in 2023 that Remotasks pay fell sharply and delayed payments were common, and a 2022 Fairwork study found the platform met only one of ten fair work standards. Diversify across platforms.
• Using AI to do the tasks. Clients pay for human judgment, so pasting chatbot output breaks the point of the work and risks removal from projects. Read each platform's rules on AI tools before you start.

Sources

• DataAnnotation: home page
• DataAnnotation: FAQ
• Outlier: home page
• Outlier: FAQ
• Alignerr: home page
• Mercor: home page
• Mercor: contractor payments
• Mindrift: home page
• Wikipedia: Scale AI
• Hara et al. (CHI 2018): Mechanical Turk earnings
• Prolific: researcher pay guidance
• FTC: job scams
• FTC: 2024 fraud loss data