AI Data Annotation Jobs: What the Work Is, What It Pays, and Who Actually Gets In
Every AI model you have heard of was trained on work done by people. Someone ranked two answers and said which was better. Someone drew a box around a pedestrian. Someone wrote the ideal reply to a tricky question so the model could learn from it. That work is called data annotation or AI training, it is entirely remote, and the advertised hourly rates are high enough that it looks too good to be true.
Some of it is. This guide separates the part that is real from the part that is not.
What the work actually is
- Preference ranking. You are shown two AI answers to the same prompt and you choose the better one, then explain why. Most common task type, lowest barrier.
- Writing model responses. You write the answer a model should have given. Pays more, and rewards anyone who writes clearly.
- Domain review. A nurse checks medical answers, an accountant checks finance answers, a developer checks code. This is where the high rates live, and it needs a real qualification.
- Image and audio labelling. Drawing boxes, tagging objects, transcribing speech. The most mechanical, the lowest paid, and the fastest to be automated away.
What it pays
DataAnnotation, one of the larger platforms, advertises on its own homepage a general range of $20–60+/hour, with language specialists at $25–40/hour, and coding, finance, legal, medical and STEM work at $75–150+/hour.
Those are the rates. They are not the monthly income, and the difference between the two is the whole story of this job.
| Hours you actually get | At $20/hr | At $30/hr | At $75/hr (specialist) |
|---|---|---|---|
| 5 hrs/week | $400 | $600 | $1,500 |
| 15 hrs/week | $1,200 | $1,800 | $4,500 |
| 30 hrs/week | $2,400 | $3,600 | $9,000 |
Read that table with the top row in mind, because for most people the top row is the real one. The rate is not the limiting factor — task availability is. Work arrives in batches. A project closes and your queue is empty for a week. You are not an employee, there are no guaranteed hours, and no platform promises you any.
That is why the realistic figure for someone starting out is $250 to $1,500 a month, and why it swings so hard month to month. Treat it as a strong second income, not a salary you can plan rent around.
The other filter nobody mentions
Most platforms make you pass an unpaid assessment first — usually a writing test, sometimes a domain test. A lot of applicants do not pass it, and most platforms will not tell you why. Budget an hour or two for it, do it when you are fresh, and treat the writing quality as the thing being tested, because it is.
Where to sign up
Go to the platform's own site. There is no application fee anywhere on this list, and no legitimate middleman.
- Appen — the oldest of these, and historically the most open to contributors outside North America and Europe.
- Outlier — expert-level work, strong on coding and STEM.
- Mindrift — project-based AI training work.
- DataAnnotation — the rates quoted above are theirs.
- Remotasks — more image and mapping work than text.
Check the country list at signup, not on a blog. Eligibility on these platforms changes with the projects they win, and every "countries accepted" list you find elsewhere — including lists published by people trying to sell you a course — goes stale within months. DataAnnotation's own FAQ, for example, states only that workers need reliable internet and fluent English, and does not publish a country list at all. The signup form is the only source that is current.
What makes the difference between $250 and $1,500
Register on three or four platforms, not one. When one queue dries up another usually has work. This single habit is most of the gap between people who earn steadily here and people who quit after a month.
Lead with a specialism. Generic tasks pay generic rates. If you have a nursing qualification, an accounting background, a law degree or working code on GitHub, say so at signup — that is the difference between the $25 tier and the $75 tier, and it is the same effort per hour.
Sort out payment before you start. Most of these platforms pay through PayPal, and PayPal's support for receiving money varies by country. Confirm you can actually receive and withdraw before you put in fifty hours.
Write like a person, not a template. Every one of these platforms scores your work quality, and your score decides which projects you are offered. The people earning at the top of the range are not faster — they are better, and the system notices.
Two things to be honest with yourself about
This work is genuinely open to people in Africa in a way most remote jobs are not, and it pays in dollars. It is also unstable by design, the assessments reject a lot of people, and the volume of the simplest tasks is falling as models improve. Go in with a second income in mind and it is one of the best options available. Go in expecting a salary and it will disappoint you.
And the usual rule holds here too: nobody legitimate charges you to apply, to be "verified", or to unlock tasks. If a site asks for money before you have earned any, it is not one of these platforms.
