Somebody has to teach the chatbots. Every time a model gets a little better at writing code or explaining a tax rule, there was a person somewhere who read its answer, decided it was wrong, and showed it what right looks like. That person is usually a freelancer working from home, and the contract calls them an AI trainer, an AI rater or a data annotator.
It's real work with real pay. It's also wildly uneven. On Outlier, which says it has more than 900,000 contributors in 52 countries, general English rating is advertised at up to $15 an hour. Someone with a PhD in quantitative finance can earn up to $150 an hour on the same platform. Most people end up much closer to the first number than the second, and that's worth knowing before you spend an evening on screening tests.
What the work actually is
Most of it is judging. A typical rating task shows you a request, say, "explain compound interest to a 14-year-old," and two answers the model wrote. You check whether the math is right, whether the answer is really pitched at a teenager, and whether anything in it is unsafe. Then you pick the better one, score both, and write two or three sentences explaining your choice. That explanation is the part the AI lab actually wants, so it matters more than the click.
Conversation training flips the task around: you write the request and the ideal answer yourself, sometimes across a long back-and-forth.Annotation jobs have you label images, video or audio.Voice evaluation means listening to AI-generated speech and scoring how natural and accurate it sounds.Outlier's Hindi voice role, for instance, asks for a native or highly fluent speaker to score conversations on naturalness, helpfulness and audio quality.
Then there is expert work, where a pharmacist checks a model's dosing logic step by step or a developer writes a coding problem the model fails and grades its attempted fix.The job titles vary (AI Rater, AI Conversation Trainer, AI Data Annotator, AI Voice Model Evaluator), but the core skill is the same: careful judgment, clearly explained.
What AI trainer jobs pay
Rates are posted, but they are almost always ceilings. These are the figures the platforms themselves were advertising in late October 2026:
- General rating and annotation on Outlier: up to $15 an hour. That is the figure in every one of its 5,136 listings on Joblet that state pay.
- Workada's AI and data labeling listings: $17 an hour.
- Hindi voice evaluation on Outlier: up to $7.50 an hour.
- Building and testing AI agents on Outlier: up to $35 an hour, for people with at least two years in backend engineering, AI automation or systems integration.
- Quantitative finance on Outlier: $40 to $150 an hour, PhD required.
- Mercor: an average contracted rate of $97 an hour across its expert network.
The higher-paying roles ask for a lot. Outlier's agent-building role wants a strong command of at least two programming languages, such as Python, JavaScript, Go or Java, plus SQL, and the work involves writing evaluation rubrics, debugging agent traces and stress-testing agents. The quant finance role asks for a PhD and prior experience as a quant researcher, trader or risk modeler.
micro1 doesn't publish rates at all. Outlier says contributors earn about 7.5% more on average through Missions, its bonus program, but also warns that lower rates may apply during onboarding and overtime phases.
My read on those numbers: generalist rating is a side income, not a living. Language work depends heavily on which language you speak. The serious money goes to people with credentials the AI labs struggle to find, like doctors, lawyers, engineers and quants.
How many of these jobs are out there?
When we pulled the AI training listings on Joblet on October 28, 2026, there were 6,628 of them, and 99% were remote. 4,547 were in the US and 2,000 in Canada, with a handful in the UK, Ireland, Australia and New Zealand.
Outlier accounted for most of them: 5,201 listings across 46 job titles. It posts each role in many cities, so the same job tends to appear over and over. You only need to apply once. The numbers move every day as projects open and close, so if you want to see what is open right now, the live AI trainer search on Joblet is the place to look.
Getting in: Outlier, Mercor and micro1
Outlier's process has four steps: build a profile, import your skills, verify your identity, then pass skill screenings that decide which projects you see.It says onboarding "typically takes 30 to 90 minutes." You will usually need undergraduate-level knowledge in your area, good written English, a valid ID and phone number from your country, and permission to work there.
According to its FAQ , pay goes out every Tuesday for the previous week's work, through PayPal, AirTM or ACH bank transfer, and everyone works as an independent contractor with no minimum hours.Outlier pays either hourly or per task, and says rates vary by expertise, project complexity and location, so two people doing similar work in different countries can see different numbers.
Mercor pitches itself to professionals. It describes a network of more than 5 million experts, including physicians, lawyers, engineers and consultants, posts each role with its rate on its listings page, and advertises daily payouts. The example roles on its homepage ranged from $60 to $70 an hour for a sales and marketing expert to $95 an hour for a consulting expert. If you have years of experience in a field, it's worth a look. If you are new to this work, it probably isn't where you start.
micro1 also focuses on domain experts, in areas such as engineering, finance, healthcare and law. Its process includes an AI-run interview (a testimonial on its site mentions an AI interviewer named Zara) and structured skill assessments, with human recruiters reviewing profiles too. The useful part: its interview prep page offers free practice interviews of about 18 minutes, with feedback. Do one before the real thing.
The part the job ads skip
You're a contractor, not an employee. That means no benefits, no overtime pay and nobody withholding tax for you. "No minimum hours" sounds like freedom, and it is, but it also means nobody promises there will be work. Projects open, fill up and pause, sometimes with little warning, and your income moves with them.
Taxes catch a lot of people out. In the US, contractors owe self-employment tax of 15.3% for Social Security and Medicare on net earnings, on top of income tax, and may need to pay quarterly estimates, according to the IRS. Put part of every payout aside from the first week. Payout timing matters too if you are relying on this money: Outlier pays weekly, while Mercor advertises daily payouts.
Getting through the screening, and staying on projects
Lead with whatever you can prove. A degree, a license or years in a specific field is what separates a $15 rating job from an expert project. When the screening starts, read the guidelines all the way through first. The tests are graded against those guidelines, and accuracy counts for far more than speed.
Once you are in, your quality score decides how much work you get. Platforms move low scorers off projects, and Outlier advertises "quality-based rewards." Short, specific explanations tend to beat long ones. It also helps to be active on two platforms, so one project pausing doesn't wipe out your week, and to track your hours and payouts from day one. That record makes taxes easier and shows you your real hourly rate, which is often lower than the advertised one.
How to spot a fake AI training job
The popularity of this work has attracted scammers. Walk away if you see any of these:
- A request to pay for training, equipment, software or "account activation." Real platforms don't charge you to work.
- A recruiter who contacts you on Telegram or WhatsApp, or from a personal email address, instead of the platform's own domain, such as outlier.ai, mercor.com or micro1.ai.
- High pay offered with no skills assessment at all.
- A request for your bank login, or to receive money and send it on. That's money laundering, not AI work.
- An ID verification link on a site that isn't the platform's own. Check the address before you upload anything.
Is it worth it?
For the right person, yes. If you write precisely, like finding mistakes, have a specialty you can document and can live with income that changes week to week, AI training is a flexible way to earn from home, and a well-paid one if you have expert credentials. If you need guaranteed hours, health benefits or a predictable paycheck, treat it as a side income or a bridge between jobs, not a replacement for one.
Either way, start with one platform, take the practice tests seriously, and keep a simple record of your hours and pay from the first day.