I find AI-generated work pretty easy to detect for many of the reasons cited above (polished but generic, no citations, not tailored to the question or course materials). But I don’t accuse a student of turning in Chat GPT-generated work until I’ve run it through three or four detectors and get consistent results. I never use just one. I’ve given out a handful of zeroes for using AI-generated responses this semester, and not a single student has protested the grade.
I do tell students that the best protection they have against a false plagiarism accusation (and I include using AI as plagiarism) is their own notes, revisions, and thought work. If they can show me this material so I can see their work-in-progress, then I am very likely to reverse the plagiarism accusation (unless I can prove it without a doubt – but this is only possible with actual plagiarism, when I have the original source in hand). Never has a student challenged a plagiarism accusation in this way. Not once. I’ve gotten a few weak protests, but in those cases, I was able to show the text those students had copied. As I said above, ChatGPT is new, but I’ve not yet gotten a single protest when I’ve given zeroes for AI-generated work.
All of this is to say that your son’s best defense is to (a) run his own work through several generators to see if there are inconsistent results, (b) show his preparatory work for the assignment, and/or (c) show other written assignments to demonstrate that this is, in fact, his writing style.
If his writing style is similar to AI-generated work, then he would be wise to run his writing through detectors in the future and revise if necessary to make sure his prose is less likely to be identified as AI.