Get paid to train AI: what it actually pays

The big AI labs need people to judge their models: read two answers and mark which is better, write a correct reply for the model to learn from, or check its working on a maths or coding question. They pay for this, and they reach ordinary people through a handful of platforms you can join from home.

This is the one way to earn from AI that needs no audience, no clients and no product. You do need to pass an unpaid test, write clearly, and go in with open eyes about the one real catch. The rest of this page is what to expect and how to start safely.

What the work actually is

You are the human check on an AI. A task usually looks like one of these:

• Compare two answers. The model gives two replies to the same question and you mark which is better, and say why.
• Write the model answer. You write the reply a good assistant should have given, so the model can learn from it.
• Check its working. On a coding, maths or specialist question, you confirm the steps are right and flag where they are wrong.

General tasks need no specialist skill. DataAnnotation asks for a bachelor's degree or equivalent experience, plus clear writing (read 7 October 2026). The better-paid ones want a real skill you already have, such as coding, maths, law or a second language.

What you can expect to earn

Here is the honest version, with the stated rates kept separate from what you will actually take home.

So the headline hourly figure is true and it is also not the whole story. A good rate across a week with no tasks pays nothing, which brings us to the catch.

The one real catch

You are paid per hour you actually work, and the amount of work goes up and down. Tasks arrive in projects, and when a project you qualify for ends, your dashboard can sit empty for days. People report this constantly, and it is normal, not a sign you did something wrong.

The clearest outside look comes from Annotation Academy, an independent site that reviews this work and says it earns nothing if you join any platform. For its DataAnnotation review (updated 15 September 2026, read 7 October 2026) it read 7,514 comments across 41 threads.

It deliberately split the company's own forum from outside ones, because "a company's own forum is not neutral ground." The complaint it found most often was this: long stretches with nothing on the dashboard.

How to check a platform is legit before you join

The work is real, and the scams sit around it. Before you give any site your details or your time, run the same simple check that honest reviewers use.

1. You should never pay to start. A real platform pays you. If a site asks for a joining fee, a deposit, or to buy a course or "starter kit" first, close the tab.
2. Read reviews from outside its own pages. A company's own site and forum will always look positive. Search for the name with words like pay, work dried up, or account, and read what people say elsewhere.
3. Check how it pays and when. A genuine platform tells you the rate, how hours are counted, and how and when you are paid. If that is vague, treat it as a warning.
4. Never hand over money details beyond what gets you paid. A payout method is normal. Card numbers, "verification" fees, or crypto are not.

The real platforms, and what to confirm

These are the two most talked about, and they are not the only ones. Whatever you join, the four checks above come first.

What the unpaid assessment is really like

Everyone starts with an unpaid assessment, and it is more than a quick quiz. Going in knowing the shape of it stops you giving up halfway or misreading what happens next.

Work out if it is worth your time, then plan week one

Before you sign up, it helps to be honest with yourself about whether this fits your week and which tasks you could actually get. This prompt walks you through that, then lays out a careful first week. Paste it into a new chat in Claude or ChatGPT, fill in the lines in brackets, and send.

You are a blunt, practical adviser helping me decide whether paid AI-training work (rating and writing AI answers on platforms like DataAnnotation or Outlier) is worth my time, and if so, how to start safely. What is at stake: this work pays per hour actually worked and the amount of work swings up and down, so the wrong expectation wastes weeks. Be honest, not encouraging. If it is a poor fit for me, say so plainly.

FILL THIS IN
- Where I live (country, and state or region if it has one): [fill in]
- Hours a week I could realistically give this: [a number]
- Why I want it: [e.g. steady side income, flexible top-up, trying it out]
- Skills I actually have that a model could be tested on: [e.g. coding in a named language, maths, law, nursing, fluent in a named language, or "none specialist, general only"]
- The lowest amount that would make the time worth it for me this month: [a number and currency, or "not sure"]
- My situation: [e.g. need reliable income to pay a bill, or have spare time and no pressure]

IF ANYTHING KEY IS MISSING
If my hours, my skills line, or my "worth it" number are blank, ask me for them and wait. Do not guess them for me.

STAGE 1: THE HONEST FIT
Tell me straight whether this suits me, based on what I wrote. Rules you must apply:
- If I need reliable income to cover a specific bill, warn me plainly that this work cannot be relied on, because work availability is not steady, and suggest I treat it only as a top-up.
- If I have a specialist skill, say that the better-paid tasks will suit me and why.
- If I have very few spare hours, be honest about whether it is worth the unpaid assessment and setup for that little time.

STAGE 2: WHAT TO EXPECT FOR MY HOURS
Using only the hours I gave you, lay out a realistic picture in plain words: that pay is per hour worked, that some weeks may have little or no work, and that I should not count on a fixed monthly figure. Do not invent an income figure or a rate. If I ask "how much will I make", explain why no honest person can promise that, and point me to check the platform's own current rate.

STAGE 3: THE SAFETY CHECK
Give me the checklist to run before I join any platform, in my own situation:
- never pay a fee, deposit, or buy a course to start;
- read reviews from outside the platform's own pages;
- confirm the stated rate, how hours are counted, and how and when I get paid;
- confirm the platform accepts people where I live;
- never give money details beyond a payout method.
Tell me to confirm the current rate and accepted countries on the platform's own site, because these change, and mark that line [CHECK ON THE SITE TODAY].

STAGE 4: MY FIRST WEEK
Lay out a careful first week for the hours I have:
- Day 1: pick one platform, read its own rules and an outside review, and start the unpaid assessment only if the safety check passes.
- Across the week: how to approach the assessment (take care, follow instructions exactly, write clearly), and a note that the assessment is unpaid and passing it does not guarantee work.
- A simple way to track my actual hours and actual pay, so after two weeks I can judge the real rate for myself rather than trusting any advert.

STAGE 5: THE TWO-WEEK DECISION
Give me three plain questions to answer after two weeks that tell me whether to keep going, treat it as occasional, or stop. Base them on my own "worth it" number and hours, not on any promised figure.

NEVER
- Never invent a pay rate, an hourly figure, a monthly income, or how much work I will get.
- Never tell me a platform is safe or legit from memory; give me the checks to confirm it myself.
- Never talk me into it. If it does not fit what I told you, say so.

LAST CHECK
Re-read your answer. Confirm you have not stated any income figure as fact, that every rate or amount is marked for me to check on the platform, and that your fit verdict matches what I told you about my hours and my need for reliable income.

The honest bit

Start this week

Run the prompt above, pick one platform, and read an outside review before you touch the sign-up. If it passes the four checks, start the assessment this week and track your real hours. In two weeks you will know the true rate for you, from your own numbers.

If you would rather earn from a skill you build once and sell again, the six side hustles worth your time is the better place to start.

Questions people ask

Is getting paid to train AI real or a scam?

The work is real. Big AI labs pay people to rate and write example answers, and reach them through platforms like DataAnnotation and Outlier. The scams are the courses and "systems" sold around it that promise easy passive income. The work pays per hour you actually work, and the amount of work goes up and down.

How much can you make training AI?

DataAnnotation lists about 25 to 50 US dollars an hour or more for general work, and from about 20 for multilingual work (read 7 October 2026). Those are rates for time worked, with no set number of hours. Monthly pay depends on how much work is available, so treat it as occasional income you cannot count on.

Do you need experience to train AI?

DataAnnotation asks for a bachelor's degree or equivalent experience for general work, plus an unpaid assessment, and clear writing helps. The better-paid tasks want a real skill you already have, such as coding, maths, law or a language.

Why did my work dry up?

This is the most common complaint, and it is normal. Work is handed out in projects, and when one you qualify for ends, your dashboard can sit empty for days. It is not a sign you did anything wrong, and it is the main reason this cannot be a reliable full-time income.