Generative AI is artificial intelligence that creates new content (text, images, audio or code) from a request called a prompt. It does not pick a stored answer: it builds one on the spot using patterns learned from huge amounts of examples. ChatGPT is the best known case.
A question from the app. Pick an answer.
Ask the same thing twice: same answer?
Not always. Many apps pick each piece with some randomness, so it can vary.
Many apps pick each piece with some randomness. So two answers can come out different.
Repeating does not force the AI to refuse. The app may have usage limits.
Picture an email from a customer complaining about a late delivery. A classic AI looks at it and gives it a label: "complaint", "urgent", "not spam". A generative AI instead writes you a draft reply that did not exist before. Both are AI; only the second one creates.
Inside, it is all numbers: a text model picks one piece of text (a token) after another. What you see at the end can be a summary, an image, a song or a piece of code. If it also understands photos or audio, it is multimodal AI.
The prompt is everything you give the AI to work with: your question, the instructions, the examples and any text you paste. Compare "write a message to my boss" with "write a short, formal message telling my boss I will be in at 10 tomorrow because I have a doctor's appointment". The second leaves much less to chance.
A detail many people forget: say which variety of the language you want. If you write in English and do not say it, you may get British spelling when you needed American, or the other way around. Asking for "in US English" usually does it.
"What it generates is true." It can invent facts, quotes, laws or links in the same confident tone; that is called a hallucination. Check what matters. "It copies and pastes from the internet." It generates new text, though it sometimes repeats phrases it has seen. "A longer prompt is always better." What helps is clarity: what you want, for whom and in what format.