The suggestion was personalized, not different. ChatGPT could have the exact same set of questions to ask. All students are independently taking the same exam

The suggestion was personalized, not different. ChatGPT could have the exact same set of questions to ask. All students are independently taking the same exam at the same time, responding conversationally. ChatGPT asks all of them the same question, and then probes based on weaknesses in the responses. Grades the answer to that question, then moves on to the next. Everyone got equal questions. Is there going to be inconsistency in how minor wording trips different followups? Sure. But it isn't clear that this randomness is worse than the subjective differences in how humans grade an essay response to a question.