In plain words
Generative AI refers to systems that generate content by using patterns learned from data. The output can be a sentence, picture, sound, program, or another structured object.
A closer look
Different model families generate in different ways. An autoregressive language model produces a sequence by predicting successive tokens. Many image models use a denoising process that transforms noise into an image guided by a prompt. “Generative” describes the capability, not one specific architecture.
Generation is shaped by training, the current input, and decoding or sampling settings. A model can combine patterns in novel ways, but it can also reproduce learned material or make unsupported claims. Whether an output is useful depends on the task: invention is welcome in fiction, while a factual summary must stay grounded.
In practice
A model drafts three versions of a product description from a list of features. The text is generated, but a reviewer still needs to check that every claimed feature was actually supplied.
A useful distinction
Generated does not mean verified, original in every respect, or based on live information. A fluent answer may come entirely from learned patterns and the prompt, without any search or source checking.