In plain English
Real data can be scarce, expensive or private. Synthetic data fills the gap with realistic made-up examples, like fake customer records or practice questions, so teams can train and test without exposing real people.
Example
A bank generates millions of fake but realistic transactions, including rare fraud patterns, to train its fraud checks without touching real customers' accounts.
In short
Made up on purpose. Realistic fake data used to train and test AI.
Why it matters
AI labs are running short of fresh real-world text, so synthetic data is a growing part of how new models are trained. It also helps with privacy rules and rare situations. The risk is that mistakes in the data generator get copied into the next model.
Common confusion
Synthetic data vs. hallucination. Both are "made up," but synthetic data is created on purpose, labeled as such and checked before use. A hallucination is an AI accidentally presenting something false as fact.