I don't think criticising AI on grounds of it not 'understanding' is a strong argument, since we have neither a definition for, nor a way to measure, what understanding is.
But the best SE answers actually convey real understanding to the reader; they go beyond the brief provided by the question, and explain a subject with conciseness and lucidity. Nobody could mistake such an answer for ChatGPT output.
[I have a suspicion that such really good answers may be mostly several years old; I haven't ever thought to try and quantify that]