Before You Give Me the Answer, Ask Me a Question

An AI assistant can hand you a beautifully organized explanation while you sit there feeling increasingly intelligent. Very considerate of it. The awkward question arrives when you close the window: can you explain the idea yourself?

Our earlier Conjugo article, “What If Getting the Wrong Answer Is Part of Learning?”, raised a question about what happens when instant answers bypass the work of understanding. Here is a practical companion to that conversation: change what you ask the AI to do first.

Instead of opening with “Solve this,” try: “Before you give me the answer, ask me a question that helps you understand how I am thinking about the problem.” That small instruction makes room for your own explanation. It gives the conversation something to examine besides an empty space waiting for the machine to fill it.

Imagine you are trying to understand why a marketing campaign drew plenty of clicks but few inquiries. An immediate answer might produce a plausible list of explanations. A thinking partner could begin by asking what the ad promised, what visitors saw after clicking, and what evidence would distinguish a weak offer from a confusing form. You still have to decide what you know. The AI helps make the missing pieces visible.

The next instruction is equally useful: “Give me one hint at a time, and let me try before you continue.” This is a proposed way to structure a learning conversation, not a guarantee that every model will become a brilliant tutor. If the question is too vague, ask for a more concrete one. If you are completely lost, ask for a worked example. There is no virtue in turning confusion into an endurance sport.

A third move is to ask the AI to inspect your attempt. “Here is my explanation. Which step is unsupported, and what would I need to show for it to hold?” That request invites a specific critique. It is more useful than asking whether your reasoning is good and receiving a flattering paragraph about your excellent insight.

The assistant also needs permission to admit uncertainty. Ask it to distinguish a factual correction from an alternative interpretation, and to identify claims that need checking. A confident machine can be wrong, and a questioning machine can lead you in the wrong direction. For questions with verifiable answers, compare the result with a reliable source, a calculation, or a test you can inspect.

Try this as a reusable opening prompt:

“Help me understand this topic. First ask what I already think and what I am trying to learn. Ask one focused question at a time. Let me attempt an answer before offering a hint. If I am stuck, show a small worked example and explain it plainly. Challenge unsupported steps without inventing objections. Flag uncertainty. At the end, give me a new example to work through on my own.”

The final part matters. Close the explanation and try a fresh example. Explain your choice in your own words. If you cannot do it yet, bring that attempt back into the conversation. The gap becomes the next thing to work on, rather than something a polished answer quietly conceals.

This approach also needs an off switch. Sometimes you need a direct answer, an accessible explanation, or help completing a task under time pressure. Tell the assistant which kind of help you want. A useful collaborator should not insist on conducting a Socratic seminar when you asked where the export button is.

For Conjugo, the larger question is what kind of human–AI relationship we choose to practice. We can judge a session by the artifact it produces, and we can also ask what it leaves us able to notice, explain, question, and do. Those are different outcomes, and both deserve attention.

The next time you open an AI conversation to learn something, try giving your own thinking the first turn. Ask the machine to meet you there. A finished answer can end a question; a well-placed question can open a much more interesting conversation.