A free preview, generated by MasteryLoopHQ.
Getting an answer from an AI chatbot takes seconds. Understanding it is still your job. When you are learning, a fast answer can skip the part that builds understanding: trying, getting stuck and working out why. The fix is not to avoid AI. It is to ask it different questions, ones that keep the thinking on your side. This guide covers six habits: ask it to explain, ask why, show your own attempt first, be specific, check what it says, and explain it back in your own words. One note: if a class, teacher or employer sets rules for using AI, follow them. These habits are for understanding a subject, not for handing in AI’s work as your own.
“What’s the answer?” gets you a result. “Explain how to get there, one step at a time, and stop so I can try the next step” gets you a way of thinking. You can also ask for a hint instead of the solution, ask for a simpler example first, or ask it to quiz you on the idea before it shows more. Try asking: “Don’t give me the answer yet. Explain the idea behind this problem, then give me one hint.”
Asking why turns a fact into a reason. “Why does this step work?” “Why this formula and not that one?” “Why is the other option wrong?” Reasons are what you can reuse on a new problem. “Why not?” questions are especially useful because they show the edge of an idea, the point where it stops working. Try asking: “Why does this work? And when would it not work?”
Before you ask for help, write down what you tried, even if it is wrong, and paste it in. Then ask AI to find the first step that went off, and to explain why, without fixing the rest. This lines up with one idea from tutoring research. In a 2011 review, Kurt VanLehn compared human tutoring, computer tutoring and instruction without tutoring. He sorted computer tutors by how small a piece of the learner’s work they responded to, “into answer-based, step-based, and substep-based tutoring systems.” Most intelligent tutoring systems, the review notes, respond at the step or substep level, while most other computer tutors respond only to final answers. That review covers human tutors and computer tutoring systems built before today’s AI chatbots, and it did not study today’s AI chatbots. The takeaway for a learner is modest: help that responds to your steps is a different kind of help from a reply to a final answer, and you can ask for that kind of help by showing your steps. Try asking: “Here’s my work. Don’t redo it. Tell me which step, if any, is the first one that’s off, and why.”
A vague question gets a generic lecture. Compare “Explain derivatives” with “I get that a derivative is a slope, but I don’t see why a slope at a single point makes sense. Explain that part, assuming I know algebra but not limits.” A good question says what you are learning and why, your level, what you already understand, the exact sentence or step that loses you, and how you want it explained (an example, a picture in words, an analogy). Try asking: “I understand ___ but I’m lost at ___. I already know ___. Explain only the part I’m stuck on, with one example.”
AI can be wrong while sounding sure. NIST, the U.S. standards agency, names this risk “confabulation” and defines it as “The production of confidently stated but erroneous or false content (known colloquially as ‘hallucinations’ or ‘fabrications’) by which users may be misled or deceived.” There is no trick that reliably catches these errors. The real check happens outside the chat. Ask where a claim comes from, then look at that source yourself. Compare it with your textbook, class notes or the original document. Work one example by hand. Asking AI to argue the opposite can show you a weak spot, but it is not proof either way. Checking AI is partly checking yourself. In his 1974 Caltech commencement address, Richard Feynman said: “The first principle is that you must not fool yourself—and you are the easiest person to fool.” An explanation that sounds right can feel like understanding when it is not. Try asking: “Which part of your answer are you least sure about? What source could I check it against?” Then actually check, and make sure the source exists and says what AI claimed.
Close the chat. Explain the idea out loud or in writing, simply, as if to a friend who is new to it. This is often called the Feynman technique, and it is also called teach-back. The places where you get vague, skip a step or reach for jargon are where your understanding has a gap. Then use AI as a reviewer, not a writer. Paste your explanation and ask what is wrong or missing. Do not ask it to rewrite your explanation, because that swaps your thinking for its own. Try asking: “Here’s my explanation in my own words. Don’t rewrite it. Tell me what’s wrong, what’s missing, and one question that would test whether I really get it.”
Here is one simple loop. First, try the problem yourself. Second, show your attempt. Third, say what you know and where you are lost, then ask for an explanation and a hint, not the answer. Fourth, ask why. Fifth, check one claim against a source you trust. Sixth, explain it back and ask for feedback. Seventh, come back to the idea on another day. A copy-and-paste checklist, one question per habit: 1. Did I try it first and show my attempt? 2. Did I ask for an explanation or a hint instead of the answer? 3. Did I ask why, and when it would not work? 4. Did I say what I know and where I am lost? 5. Did I check one claim against a source I trust? 6. Did I explain it back in my own words and ask what I missed? What AI cannot do for you is the trying, the checking and the explaining. Those are the learning. For a different angle, see the related study at /explore/learning-mode-prompts-and-what-they-cant-do, which looks at ready-made learning prompts and what they cannot do.
When you are learning, the question you ask AI decides who does the thinking. Asking it to explain, asking why, showing your attempt, being specific, checking what it says, and explaining it back yourself all keep the learning on your side, and give AI the job of finding your gaps instead of filling them.
In your own words: (1) Rewrite this weak question into a better one using at least three habits from this guide: “What’s the answer to problem 4?” (2) Why is showing your own attempt first different from asking for the answer, even if both get you to the right result? (3) You explain an idea back and AI says it’s “perfect.” What is one way to check that for yourself? Answer key. (1) One example: “Here’s my work on problem 4. I think I went wrong at step 2 because ___. Don’t give me the answer; tell me which step is off and why, assuming I know ___.” It shows the attempt, is specific, and asks why. (2) Your attempt shows exactly where your thinking went wrong, so the help responds to your steps. A final answer can be right without showing you where you were stuck. VanLehn’s review separated help that responds to final answers from help that responds to steps, and it did not study today’s chatbots. (3) Check it against a source you trust such as notes, a textbook or the original document; try the idea on a new example; or explain it to a person and see where they get confused. AI praise is not a check.
Reading a good explanation feels like understanding it. Usually it isn't the same thing — and you don't find out which one you've got until someone asks you to explain it back.