A model producing a citation is not looking anything up. It is producing text that has the shape of a citation, which is why invented ones look so convincing: the author is plausible, the journal exists, the year is sensible, and the paper is not real.
The check takes about a minute. Open the source yourself rather than asking the model whether it is real, because it will tell you yes. Search the exact title in quotes. If nothing comes back, search the author plus the year. If the paper exists but the claim is not in it, that is the more common failure and the more dangerous one.
Do this before the number reaches your document, not after. Once a figure is in a board pack or a funding application it gets repeated, and the person who repeats it inherits the problem.
We put this in the free course rather than behind the paywall, because the people most likely to be burned by it are the ones who have not paid for anything yet.
The lesson this comes from
Telling a good answer from bad
Everything so far gets you a better answer. None of it tells you whether the answer is true. That is a separate job and it takes about a minute once you know how.