How do AI training jobs work after you are accepted?

You are added to a project, complete its onboarding and guidelines, then take tasks from a queue. Reviewers score a share of your work, and you are paid for approved hours or tasks.

Why does work stop on an AI training project?

Usually because the project has no tasks left, or because your quality score fell below the project threshold. A project can also end when the client has enough data.

How do AI training jobs work? Projects, tasks, reviews and pay, day to day

How-to · By Nikolaos Petridis

Published 31 Oct 2026

Once you are accepted, AI training work runs on projects: you onboard, take tasks from a queue, get reviewed, and are paid for approved time or tasks. Here is how each part works day to day, and what decides whether you keep getting work.

The short version

AI training jobs work on projects. A project is one client's piece of work, such as rating coding answers or writing hard physics problems. You are accepted onto a platform once, then added to projects one at a time. Each project has its own guidelines, its own queue of tasks, and often its own rate.

If you have not applied yet, start with how to get started and what AI training work actually is. This guide picks up after you are accepted.

Step 1: onboarding to a project

Each new project starts with reading. You get a guideline document, examples and sometimes a short test. Read all of it, even if the training is unpaid. Most early mistakes are rules you skimmed, and early scores stick.

Step 2: taking tasks from the queue

Work arrives as tasks in a queue. You open one, complete it in the platform's tool, submit it and take the next. A task can take two minutes (a quick label) or two hours (a full expert solution with chain of thought reasoning).

The queue is shared and it runs out. When it is empty, there is no work until the client sends more, a gap workers call an empty queue. This is why hours vary from week to week, and why many people join more than one project or platform.

Step 3: review and your quality score

As a tasker, you do not mark your own work. Reviewers check a share of your tasks against the rubric, and hidden gold tasks test you against a known answer. The results make up your quality score.

That score decides what happens next. Keep it high and you get more tasks, sometimes better paid projects, and in some cases an invitation to review. Let it slip and you get fewer tasks, then none. Our guide to how QA reviewers score your work explains what they look for. Most lost scores come from edge cases, so check the guidelines before you guess.

Step 4: tracking time and getting paid

Of the 1,331 listings on this site at the end of October 2026, 1,141 pay by the hour and 91 pay per task. Hourly projects track your time in the tool or ask you to log it, and many set an expected time per task; going well over it can be flagged. Per-task projects pay only for approved tasks. Our guide to hourly vs per-task pay compares the two.

Pay then goes out on the platform's schedule through a payout service. How you actually get paid covers the methods and timing.

Step 5: when a project ends

Projects end when the client has enough data. That is normal, not a sign you did badly. A good record on one project is what gets you onto the next, so the quality habits above matter more than speed.

A typical week

Monday: onboarding a new project, two hours of reading, unpaid or paid at a training rate
Tuesday to Thursday: two to four hours a day of tasks, while the queue has them
Friday: queue empty; you check another project or browse open roles
Following week: reviewer feedback arrives; you adjust to it

For the vocabulary used on these projects, see the glossary.

Questions

How do AI training jobs work after you are accepted?: You are added to a project, complete its onboarding and guidelines, then take tasks from a queue. Reviewers score a share of your work, and you are paid for approved hours or tasks.
Are AI training jobs paid hourly or per task?: Mostly hourly. Of the 1,331 listings on this site at the end of October 2026, 1,141 pay by the hour and 91 pay per task.
Why does work stop on an AI training project?: Usually because the project has no tasks left, or because your quality score fell below the project threshold. A project can also end when the client has enough data.