After multiple interactions throughout weeks and months of engagement, a lot of students get genuinely good at talking to AI. They know how to phrase a prompt, give it context, push back on a weak first answer, ask it to try again a different way. That’s a real skill, and it’s worth having. Call it prompt engineering: getting better outputs by getting better at the ask.
But there’s a different skill sitting beneath it, and it doesn’t show up on any list of prompting techniques, because it happens before the prompt gets written and submitted.
Call it prompt discernment: knowing what you actually want the interaction to do for you before you optimize how you ask for it. If you’ve run into another framework that also uses the word discernment for AI use it probably means something narrower: checking whether an AI’s answer is actually any good after it comes back. That matters too. This post is about the moment before that one, the choice you make before you’ve asked for a specific outcome.
Prompt engineering makes the answer better. Prompt discernment decides whether you should be asking for the answer at all, or asking for something that makes you work harder to get there yourself. You can be excellent at one and careless with the other. In fact, the better you get at engineering, the easier it becomes to skip discernment entirely, because a well-crafted prompt gets you exactly what you asked for, fast, and fast rarely leaves room to notice you asked for the wrong thing.
Here’s the one-line version of prompt discernment. On its own it does nothing. Typed into a chat window with nothing else attached, “What would I miss out on learning if you just gave me the answer?” has no task riding with it, so there’s nothing specific for the AI to actually answer meaningfully. Prompt discernment only works pointed at something real, in the same message as the thing you were about to ask.
Take the calc problem set open in another tab. The prompt you were about to send was something like “solve this related-rates problem and show the steps.” Attach discernment to it instead, all in one message: “What would I miss out on learning if you just gave me the answer to this related-rates problem?” — then paste the problem underneath. The response tends to name something specific: setting up the relationship between the rates, deciding what’s fixed. Engineering the prompt better would still get you clean steps. Prompt discernment is what tells you those steps were the part you needed to do yourself.
Now the essay draft. The prompt you were about to send was “give me a strong thesis for this Reconstruction paper.” Attach prompt discernment the same way: “What would I miss out on learning if you just gave me a strong thesis for this paper?” A decent answer says plainly that the thesis is where you decide what you think, and that deciding is the actual assignment. No amount of engineering the prompt fixes a thesis you didn’t arrive at.
It won’t always find something to protect. Attach it to a syntax error or a factual lookup and the honest answer might be “not much.” Prompt discernment isn’t a password. It’s a check, and sometimes the check comes back clear. What it does every time is put the decision in front of you before the engineering takes over and makes the decision for you by default.
So before you spend the rest of this semester getting better at asking AI for things, spend this week getting better at asking yourself one thing first, attached to whatever you’re actually about to ask:
“What would I miss out on learning if you just gave me the answer to ____?”
Prompt engineering will keep improving on its own. It’s built to. This is the part that only improves if you practice it on purpose.