How to start AI training jobs: a beginner's guide
If you are a nurse, a lawyer, an accountant or an engineer, there is a good chance an AI lab wants to pay you for a few hours of your professional judgement. AI training jobs are among the few remote roles where a mid-career professional can earn a strong hourly rate without retraining, and where the thing being paid for is exactly the expertise you already have.
The catch is that the field is badly explained. Half the articles about it quote a single made-up hourly figure, and the other half treat it as generic data entry. We track live listings from these platforms, so this guide uses what the roles actually advertise — and is honest about the parts that are worse than the hype suggests.
What AI training work actually is
Large language models are trained partly on human judgement. After a model is pre-trained on text, humans compare its answers, rank them, correct them and write reference solutions. That feedback loop is called RLHF — reinforcement learning from human feedback — and Hugging Face's technical explainer on RLHF is the clearest free write-up of how it works if you want the mechanics.
In practice you will be doing one of five things. The type matters enormously, because it is the single biggest driver of what you get paid.
| Type of work | What you actually do | Typical rate |
|---|---|---|
| Model evaluation | Judge whether a model's answer is correct in your field, and explain why | $80 – $105/hr |
| Coding tasks | Write reference solutions, build test environments, review generated code | $50 – $100/hr |
| Research & writing | Author expert questions, reference answers and grading rubrics | $65 – $90/hr |
| Data annotation | Label images, video or text against a written guideline | $30 – $50/hr |
| Writing feedback | Rate tone, fluency and style of generated text | $15 – $30/hr |
Those figures are medians from the 341 listings we track, verified in July 2026. The full breakdown, including pay by profession, is in our AI training job pay report.
The pattern is consistent: the more your judgement is hard to replace, the more it pays. Evaluation work in a regulated profession pays roughly three times what general writing feedback pays for the same hour.
If you would rather see the sign-up flow than read about it, the walkthrough above covers account setup on the three platforms most beginners start with.
Who actually gets hired
You do not need AI experience. You need verifiable expertise in something else. The platforms are explicit about this — Mercor's own guide to freelance AI training work describes the model as matching credentialed experts to projects that need their domain knowledge.
Across the listings we track, these backgrounds are in the strongest demand:
- Licensed professionals — doctors, nurses, lawyers, accountants. The highest-paying bracket, because credentials are slow to fake and slow to replace.
- Software engineers — especially in Python, C++, Rust and competitive programming. Consistently well paid and the highest listing volume in tech.
- PhDs and academics — physics, chemistry, biology, mathematics. Labs need people who can pose genuinely hard problems, not just solve easy ones.
- Bilingual speakers — particularly outside the major European languages. Rates vary widely, so read each listing.
- Media professionals — video editors, voice actors, designers. Real demand, but the lowest typical rates of the specialist categories.
If you are not in one of those groups, generalist annotation work is still open to you. Just calibrate: that end of the market pays closer to normal remote-work rates.
The five steps to your first paid task
The process is broadly the same across platforms, and most people who fail drop out at step three.
- Pick two platforms, not one. Acceptance does not guarantee work, so applying to a single platform and waiting is the most common mistake. Start with our Mercor review and micro1 review to decide which two fit your background.
- Fix your CV before you apply. These platforms screen with AI, and a resume written for humans often fails it. Our guide on passing AI resume screening covers the specific formatting that gets people cut.
- Prepare for an AI interview. Both major platforms screen with a recorded AI conversation that probes your CV claims. This is where most qualified people fail — see how to prepare for the Mercor interview.
- Complete the profile properly. Matching is driven by what you declare. A half-filled profile quietly sinks in the queue.
- Take the first task, even a small one. Early throughput and quality scores affect what you get offered next. The first week matters more than it looks.
What to expect realistically
Here is the part most guides skip. Passing screening puts you in a pool; it does not create a job. Work arrives in bursts tied to what labs are currently building, and a good week can be followed by three quiet ones.
“Treat AI training work as strong supplementary income on irregular volume. It is a poor fit for anyone who needs a guaranteed number of hours next month.”
Three specific things worth knowing before you start:
| Expectation | Reality |
|---|---|
| Guaranteed hours | None. Volume follows lab demand and can stop abruptly. |
| Instant acceptance | Screening is genuinely competitive and rarely explains rejections. |
| One flat rate | Rates in our data range from $2 to $300/hr. Always read the individual listing. |
| Employment | Contractor work. No benefits, and you handle your own tax. |
None of that makes it a bad deal. A qualified professional billing $90 an hour for six flexible hours a week is doing well by any remote-work standard. It just is not a salary.
Getting paid
Payment runs through the platform, not the end client. Most pay on a fixed cycle by bank transfer or a provider like Wise or Payoneer, and the practical friction is currency conversion rather than getting paid at all.
If you are outside the US, check the payout method before you invest time in screening — it varies by platform and country. Our guide to getting paid from India walks through fees and timelines for one of the largest markets.
Common questions
- Do I need a degree? Not always, but credentials set your rate. The highest brackets are effectively closed without one.
- Is this the same as data entry? No. Annotation is the closest, and it is the lowest-paid category. Evaluation and research work is professional judgement.
- Can I do it alongside a full-time job? Yes — most listings are part-time and asynchronous, commonly around 10-15 hours a week.
- Will AI replace this work? The work exists precisely because models need human judgement they cannot generate themselves. Demand has grown as labs have scaled, though no one can promise that holds indefinitely.
- How fast can I start earning? Screening usually takes days to a few weeks. Getting your first task after approval is the unpredictable part.
Where to start today
Pick the two platforms that best match your background, fix your CV for AI screening, and apply to both this week rather than sequentially. The people who do well here are rarely the most qualified — they are the ones who applied to more than one place and were ready for the interview.
Browse what is currently open by profession or by type of work, or compare the platforms on our platform directory.
Some outbound links to platforms are referral links. This does not affect the pay figures above, which are taken from the listings themselves.