What is data annotation? Tasks, pay and how to start

What is data annotation

Data annotation is adding labels to raw material so a machine learning model can learn from it. The raw material might be audio, video, images, documents or text; the labels say what is in it, where, and what it means. There is usually a correct answer, set out in a written guideline, and the job is to apply that guideline accurately and consistently. How annotation relates to evaluation and expert work is covered in what AI training work actually is.

What one task looks like

Tasks are small and repetitive, but the guidelines are detailed. Three examples from current listings:

• Audio and video labelling. On micro1's Data Annotator listing you label clips against a project guideline, go back over existing annotations to check they agree, flag ambiguous samples, and record why you made each call so it can be audited later.
• Document annotation. Mercor's PDF Annotation and Transcription (Korean) project has you screen each page, draw a box around every region (title, paragraph, table, figure, caption), assign a type and reading order, link captions to their figures, transcribe the text exactly, and add metadata. Every task gets a second-expert review.
• Transcription. micro1's Danish Transcription Expert role turns speech into accurate text, following rules for hesitations, overlaps, background noise and unclear passages, then checks transcripts for consistency.

Robotics footage is a notable source of this work: several micro1 listings label actions, objects and events in footage of robots, and score clips for quality against a scoring guide.

Who hires for it

Of the 699 listings on this site on 29 September 2026, 27 have annotation, labelling or transcription in the title. micro1 posted 19 and Mercor 8. Counting listings that name annotation as a skill, the figure rises to 121, because annotation is often one step inside a broader evaluation or expert role.

The 27 annotation-titled roles skew junior: 18 are pitched at junior level and 4 at entry level. Ten are micro1 transcription roles, each in a specific language.

What it pays

All 27 quote an hourly rate, and most sit well below the rest of the board. For the 22 entry and junior roles the median band runs from about $13 to $23 an hour, against a site-wide median floor of $50.

• General labelling: micro1's data annotator roles pay $6–10, its video clip roles $7–10.
• Document annotation: Mercor pays $12.68 on its Bengali, Gujarati, Malayalam, Odia and Telugu versions and $33.58 on Korean, for essentially the same work.
• Transcription: $10–35 across micro1's language roles, depending on the language.
• Specialist annotation: $50–90 on micro1's robotics video roles, which ask for engineering or annotation experience, and $200–400 for Mercor's radiologist imaging role.

The ceiling is set by what you know: the top of the radiologist band is forty times the top of a general clip labeller's.

How to start

If you have no specialist background, start with transcription in a language you speak natively, or with general labelling, and keep a record of your accuracy. Our guide to remote data labeling jobs with no experience covers applications, screening and red flags in detail.

Browse current roles under Languages & Linguistics, Audio & Voice and Video, Image & Design.