It should be unsurprising that the professor's email got flagged because the highlighted areas are fairly boiler plate. People keep trying to convince me that ChatGPT and other LLMs aren't stochastic parrots, but if they are, they will produce lots of boiler looking text, and thus detectors will be far more likely to false flag any text along these lines. Plagiarism detectors frequently flag these same types of sentences as there is so low variance in these types of messages that people will often write the exact same thing, word for word. LLM detectors can't actually give you accurate confidence indicators. They indicate how confident the model is of the result, not how likely the prediction is to be accurate.