I changed one thing about the way I use ChatGPT, and it changed what I get out of it. When I say something vague or unfinished, I don't want it to decide what I must have meant and write a polished version. I want it to ASK me. Sometimes I answer two questions and say, “No, that's enough. Let's move forward.” Other times we keep going. One question leads to another, and suddenly I'm considering options I didn't even realize were options when I started. I remember wanting to comment on a post about an experience that overlapped with my own in some ways, but differed in important others. I wanted to support what people were sharing without putting my own perspective in front of theirs. My first thought was basically: how do I write the right comment? But I had ChatGPT ask me questions instead. What was I actually trying to accomplish? Was I hoping to add my perspective, or help their message reach more people? Did supporting the post require commenting at all? We got to alternatives I hadn't originally considered: liking it, saving it, sharing the original message where appropriate. Supporting without making myself the center of the conversation. I didn't end up with a better comment. I ended up reconsidering whether a comment was what I needed in the first place. That experience keeps coming back to me as I think about AI in education. There's also a new study, published October 23, of 83 university students using generative AI. The groups guided through structured question sequences showed more learning-monitoring behaviors, greater reported metacognitive awareness, and higher course scores than the free-questioning group. It's a small, nonrandomized study, and it can't tell us which part of the approach made the difference. Still, I'm interested in what it suggests we should study next. I don't want AI to finish my thinking. I want it to help me see where that thinking can go. And I want to be able to say “enough questions” and make my own decision. How are we helping students stay in charge of their thinking when they use AI?

Prompt Engineering