When you ask an AI assistant a question - whether it's about history, science, or how to fix a stubborn piece of code - the first step is to pause before accepting the answer. AI excels at fluency, and fluency can be disarming. A paragraph that reads smoothly can feel true even when it isn't. So the discipline begins with slowing down and asking: What claims is the AI actually making? Extract the concrete statements from the prose. Dates, names, causal relationships, statistics - these are the pieces most likely to drift.