Data annotation is the process of labeling or tagging raw data, such as images, text, audio, or video, so machine learning models can learn from it. Annotations add the meaning a model needs, like marking objects in an image, categorizing a document, or noting the sentiment in a sentence.
Annotation can be manual, semi-automated, or a mix, and usually includes quality checks because label errors flow straight into model behavior. For example, a team building a defect detector might annotate thousands of product photos, marking each flaw so the model learns what one looks like.
Annotation prepares training data by adding labels. Whether the resulting dataset is usable and reproducible in a specific AI run is a separate readiness question.