Glossary · Updated Sep 2, 2026

Training data

The text, images, audio, or code used to teach a model, which shapes its capabilities, biases, and legal exposure.

Definition

Training data may include public web content, licensed datasets, synthetic data, and, for some products, user interactions. Its provenance affects copyright risk and commercial safety, which is why some vendors train only on licensed material and offer indemnification. Data quality also determines how well models handle specific domains and languages.

Why it matters when choosing a tool

Knowing what a model was trained on informs both output rights and whether your own data might be used to train it.

Where you will meet it

AI Image & Design, AI Music Generation, AI Legal & Contracts

Related terms

Data residency and privacy · Training opt-out · IP indemnification

Tools where this matters

Reviewed products in the related categories