What is AI training work?
Updated 31 Oct 2026
AI companies need humans to teach their models what a good answer looks like. AI training platforms pay people to provide that feedback. You do not build AI and you do not need to code (unless you apply for coding roles).
Typical tasks
- Rating and ranking: compare two or more AI answers and choose the best, then explain why.
- Writing prompts: write questions that are hard for AI, usually in your field of expertise, plus the correct answer.
- Creating rubrics: define what a complete, correct answer must contain so other answers can be graded.
- Labeling data: tag images, audio, text or video so models learn to recognise things.
- Red-teaming: try to make a model produce unsafe or wrong answers, and document what happened.
- Evaluating agents: check each step an AI agent took to complete a task.
Three levels of work
- Microtasks (Remotasks, Clickworker): simple and open to almost everyone, but low pay.
- Generalist AI training (DataAnnotation, Outlier): good writing and attention to detail, usually $15-$50/hr.
- Expert AI training (Mercor, micro1, Alignerr): professionals and graduates in coding, STEM, law, finance or medicine, often $50-$150/hr.
Language skills are a fourth path. Bilingual speakers are in demand on micro1, OneForma and TELUS Digital, especially for less common languages.
Is it a real job?
It is contract work. You are an independent contractor, you choose your hours, and projects can start and stop without warning. Treat it as flexible side income, and sign up to two or three platforms so you are never without work.