Glossary entry

Data annotation

The process of adding labels, judgments, corrections, or other structure to data used for training or evaluation.

Definition

What it means

Data annotation turns raw or generated material into examples that an AI system can learn from or be measured against. Depending on the task, annotators may classify content, write demonstrations, compare responses, apply a rubric, correct an output, or review a sequence of actions.

Why it matters

Role in the system

Annotation makes human and domain expertise explicit enough to guide training and evaluation, while annotation quality directly affects the reliability of the resulting signal.

Example

In practice

A software engineer reviews an agent's terminal session, marks whether each action is valid, and records a corrected solution for supervised fine-tuning.