How a CS2 TA Screens ChatGPT-Generated Python Before Grading

ChatGPT-generated Python doesn't announce itself with a watermark. It announces itself in small repeated habits: a docstring that restates the function name, a broad except Exception wrapped around ten lines, variables named result_dict and student_data, comments that narrate the obvious. In my CS2 section last fall, a 20-minute screen using Codequiry's AI code detector plus a manual read of the top 25 flagged files caught 18 of 22 confirmed cases before grading started.

The fastest triage is a two-layer pass. First, run every submission through an AI code detector that looks at statistical and structural signals, not just keywords. Then read the highest-probability files like a code reviewer, not a grammar checker. You aren't looking for proof in the first pass. You're deciding which ten files deserve fifteen minutes of human attention.

The