Mostly not like that. an AI like chatGPT isn't a guy who reads manuals then learns from them and then can answer questions. So training it on the manuals wouldn't have a lot of responses past what a search of the documents would do, but worse.

What it really would need is to train on like, message boards where people answered those questions again and again, until it could predict what answer goes with what sort of question.

(but also, a major flaw in this sort of statistical language generation is they generate sentences and don't really know what anything is. So if you asked the minimum width of some cable or something it would do a great job creating an answer that LOOKED like an answer with some numbers in it, but wouldn't necessarily be anything real. ChatGPT is real bad on like, if you ask for a phone number it'll create a phone looking number that is just random numbers, because that gives a sentence that looks correct)