You follow an explanation and think, “That makes sense.” Later, a similar problem appears and you cannot get started.

Understanding a worked answer while it is in front of you and producing an answer yourself are different tasks. If your goal is to learn, build time for both.

An AI conversation can be one source of assistance. It can also reveal the answer too early, misunderstand your attempt, or give incorrect feedback. A prompt asking it to tutor carefully does not guarantee good teaching. You remain responsible for choosing reliable source material and checking what you are learning.

Here is a practical routine for a small skill you need at work. Each session should end with something you attempt yourself. If you prefer to avoid AI, the same routine works with a textbook, a colleague, or a course exercise.

Start with a skill, not a vague subject

“Learn spreadsheets” is difficult to practice in one sitting. “Explain why this formula changes when I copy it down” is manageable. “Understand business” is enormous. “Calculate a percentage increase and explain the denominator” gives you something observable.

Choose a trusted reference before the conversation: official documentation, course material, a textbook, or an expert-reviewed explanation. For a workplace process, the authoritative reference may be your organization's current procedure.

Write a goal in plain language: “At the end, I want to calculate the increase between two amounts and explain each step without the example open.” That sentence will tell you whether the session accomplished what you intended.

Make an attempt the tutor can inspect

Try the task before requesting an explanation. Show the relevant steps, including the place where you became uncertain. This gives you a specific question and makes it easier to examine feedback.

For example, imagine a fictional monthly cost rises from $80 to $100. Your attempt might be:

The difference is $20. I divided 20 by 100 and got 20%. I am unsure which amount belongs underneath the fraction.

That is more useful than “I don't understand percentages.” It identifies a choice you can reason about. Do not include private work data; use invented numbers that preserve the structure of the problem.

Request the smallest useful hint

Try this instruction:

I am practicing percentage change. Here is my attempt: [steps]. Ask one question about the first incorrect or unclear step. Wait for my response. Do not give the complete solution. If you are uncertain, say so. I will check the explanation against my reference.

A useful next question might be, “Which amount represents the starting position you are comparing the increase with?” You can then reason that the $20 increase is being compared with the original $80. Calculate $20 ÷ $80 = 0.25, or 25%.

The tool might instead give you the whole answer. If it does, stop that example. Reading a revealed solution is allowed, but it no longer tests whether you could produce the next step. Choose a fresh example before judging your progress.

Explain the correction in your own words

Do not stop at “25% is correct.” Explain why: “The original $80 is the baseline. The $20 increase is one quarter of that starting amount.”

Check the explanation against the trusted reference. If the feedback contradicts it, inspect the exact disagreement. You can ask the tool to identify its assumption, but another confident answer does not settle the issue. Consult the documentation, answer key, teacher, or knowledgeable colleague as appropriate.

A corrected result with no explanation may finish a worksheet. It does not tell you whether you understand the choice that produced it.

Close the conversation and try a related task

Now calculate the increase from $120 to $150 without looking at the previous working. Write the difference, choose the baseline, calculate the ratio, and explain it. Then check: $30 ÷ $120 = 0.25, or 25%.

Try a variation too. A value falls from $150 to $120. The decrease is $30 against the starting $150, so the percentage decrease is 20%. Returning to an earlier amount does not necessarily reverse the percentage with the same number.

If this exposes confusion, return to the specific step. You have useful evidence about what to practice next. A fresh attempt is a personal check, not a standardized measurement of learning.

What the research supports—and what it does not

A 2025 field experiment studied nearly 1,000 high-school mathematics students. A standard AI interface helped during assisted practice but was followed by weaker unaided exam performance than the control condition. A tutor with teacher-designed safeguards largely mitigated that effect; it did not show an unaided performance advantage over the control group. Read the PNAS study.

The specialized tutor used problem-specific teacher input, including correct solutions and guidance about mistakes. An ordinary “give me hints” prompt is not equivalent. The study also does not establish that every adult work task will show the same result.

My practical recommendation is narrower: when learning matters, distinguish the work completed with assistance from the work you can later do and explain yourself.

Your next 15-minute session

Choose one small skill and a trusted explanation. Spend five minutes trying a problem, three minutes getting one targeted hint, five minutes on a fresh problem, and two minutes recording what remains unclear. Extend the session when checking requires more time.

Without AI, use a worked example with the final steps covered, a textbook hint, or a colleague who agrees to ask a question before explaining.

Finish with this reflection: What can I now do without the answer visible, and which step still needs help? Keep that note. It gives your next session a better starting point than another broad request to explain everything.