Use AI to quiz you, then check what you remember without it

The answer should disappear before the practice begins

An AI explanation can feel wonderfully clear while it is on screen. Every step follows from the last, unfamiliar terms are defined, and the final answer seems obvious. Then the chat closes and the same problem becomes difficult again. The useful question is not whether the explanation sounded good. It is whether you can reconstruct the reasoning when the explanation is gone.

I would use AI as a question setter and a feedback assistant, not as a continuous stream of finished answers. That changes the shape of a study session. Instead of asking for another summary whenever something feels uncertain, attempt a small question, inspect the gap, and ask for targeted help with that gap.

Retrieval practice means trying to bring information to mind instead of only rereading it. Cognitive scientist Pooja Agarwal explains it as a learning strategy, not merely an assessment. That principle supports practicing recall; it does not prove that any particular chatbot or the schedule in this article improves grades. The routine below is a practical suggestion, not a tested tutoring product.

There is still a place for explanations, reading, and worked examples. Someone encountering a topic for the first time needs material to learn from. The problem is allowing explanation to occupy the entire session. I would alternate a brief explanation with an attempt that exposes what is understood and what is only familiar.

Sources: Pooja Agarwal: What is retrieval practice?

Start with one lesson and one small objective

Choose a reliable source before opening the chat: an assigned textbook section, a teacher's notes, or a course lesson that you are permitted to use. A chatbot's general knowledge should not silently replace the material your course expects you to learn. If two sources use different definitions, that difference belongs in the session rather than being smoothed away.

Keep the objective small. Understand statistics is too broad for a short session. Distinguish a percentage change from a change in percentage points is manageable. It gives the question setter a boundary and gives you a concrete way to check the result. A narrow objective also makes it easier to notice when the chatbot wanders into an unrelated explanation.

For the worked example here, imagine a short lesson that says a percentage-point change is the difference between two percentages. Relative percentage change divides that difference by the starting percentage and multiplies by one hundred. Moving from twenty percent to thirty percent is an increase of ten percentage points, or fifty percent relative to the starting value.

Those are different descriptions of the same change, not competing answers. The starting value matters for the relative calculation. A move from forty percent to fifty percent is also ten percentage points, but its relative increase is twenty-five percent. This distinction gives us enough material for definitions, calculations, explanations, and error diagnosis without needing a large document upload.

Before using your own material, remove personal information and check your school's rules for AI use. Having access to a document does not necessarily give permission to upload or redistribute it. When the source cannot be shared, write your own short concept summary and keep the original available for manual checking.

Tell the chatbot not to reveal the answer yet

A normal request to explain a topic usually produces an explanation immediately. For practice, specify a different interaction: one question at a time, no solution until an attempt, and a short hint only when requested. This is not a security guarantee. It is a clear instruction that makes the intended learning activity more likely.

Ask the chatbot to keep its questions within the supplied lesson. If it needs outside knowledge, it should say so rather than adding that knowledge to the answer key without warning. This is particularly important when preparing for a course that uses a specific notation, method, or vocabulary. An alternative method can be valid and still be unhelpful at that stage of the course.

I would begin with three question types: explain the distinction in plain language, calculate a fresh example, and diagnose a mistaken answer. Each exposes a different weakness. Repeating the definition does not demonstrate the calculation, and performing a calculation does not necessarily show that the result has been interpreted correctly.

Keep the answer key separate from the question text. If the interface shows both at once, copy the questions into a blank document and hide the chat. That small practical adjustment matters more than writing a complicated prompt while leaving every solution visible beside the exercise.

Use only the lesson below to prepare a practice session. Ask one question at a time. Do not reveal the answer before I attempt it. Mix a plain-language explanation, a new numerical example, and an error-diagnosis question. If I request help, give one hint before the full explanation. After my attempt, show the correct reasoning and flag any uncertainty in your answer key. Do not treat your grading as authoritative.

LESSON:
[Paste a short lesson you have permission to use.]

Write an answer before deciding how much help is needed

Start with a question such as: A survey result moves from thirty percent to thirty-six percent. What is the change in percentage points, and what is the relative percentage increase? Put the chat aside and write the calculation. The difference is six percentage points. Dividing six by the original thirty gives zero point two, or twenty percent.

Include a sentence describing the denominator. That is where a common misunderstanding becomes visible. The relative increase is measured against the original thirty percent, not against the new thirty-six percent and not against one hundred respondents unless the question specifically provides counts that support that calculation.

An incomplete attempt is useful. You might correctly subtract the percentages but be unsure which value belongs underneath the fraction. Write that uncertainty down. It gives the chatbot a precise question to answer and prevents a correct final number from concealing a guess. A study record that says unsure about denominator is more actionable than a general note saying bad at percentages.

Do not spend the whole session staring at a blank page to prove independence. After a genuine attempt, request a small hint. For this example, a hint might ask which quantity the phrase relative increase uses as its reference. The aim is to expose and repair a gap, not turn practice into an endurance test.

Mark whether the answer was independent, hint-assisted, or completed after reading the solution. These are different outcomes. Counting all three as equally correct makes progress look cleaner while removing the information needed to choose the next exercise. You can improve without pretending every success happened unaided.

Ask for feedback on the reasoning, then verify it

After answering, ask the chatbot to identify the first incorrect step rather than merely announce correct or incorrect. If the subtraction is right but the denominator is wrong, a complete lecture on percentages may be unnecessary. The useful feedback points to that decision and explains how changing it affects the interpretation.

Check the feedback against the source lesson and a calculator where appropriate. AI can misread an answer, accept an incorrect explanation, or produce an inconsistent answer key. A friendly response is not validation. When the chatbot disagrees with the original material, stop and compare the definitions before continuing the quiz.

For the thirty-to-thirty-six example, both calculations can be checked directly: thirty-six minus thirty equals six, and six divided by thirty times one hundred equals twenty. If the chatbot labels six percent as the relative increase, its explanation has collapsed two different quantities. That is an error in the feedback, not evidence that the learner should change a correct answer.

When an answer is partly right, preserve the correct part in the error log. Write: Subtraction correct; used new value as denominator; next question should test starting value. That note is specific enough to guide practice tomorrow. A red cross beside the whole exercise would discard the useful distinction between what is stable and what still needs attention.

For subjects where you cannot independently verify the feedback, use the chatbot more cautiously. Ask a teacher, tutor, or knowledgeable reviewer about unresolved disagreements. Adding more confident AI explanations does not settle a question that requires subject expertise. The study tool should help locate uncertainty, not conceal it.

Change the numbers and reverse the direction

Once an example is understood, try one with a different structure. A fall from forty percent to thirty percent is a decrease of ten percentage points and a relative decrease of twenty-five percent. The subtraction and the starting value still govern the calculation, but the direction has changed. That makes it harder to succeed by copying the previous sentence.

Next ask why a rise from twenty percent to thirty percent is not the exact opposite percentage change of a fall from thirty percent to twenty percent. Both moves span ten percentage points, but their starting values differ. The increase is fifty percent; the decrease is approximately thirty-three point three percent. Explaining the asymmetry tests the meaning of the denominator.

Another useful question removes the numbers: Can two changes have the same percentage-point difference but different relative percentage changes? Explain why and give an example. This requires generating evidence for the principle rather than choosing a familiar answer from a list. Keep a calculator available to check arithmetic after the attempt.

Do not ask for endless variations once the same reasoning is secure. Move to a short interpretation task, such as rewriting a fictional report sentence that confuses the two measures. The goal is to use the concept in a new setting. Producing fifty near-identical questions can create a busy session without revealing anything new.

If the new variation suddenly becomes much harder, identify what changed. Was it the direction, the language, the arithmetic, or the missing starting value? Ask for practice on that specific difference. Difficulty is easier to address when it has a name.

A short session needs an ending, not an endless chat

Here is one possible twenty-five-minute session, offered as a convenient time box rather than an optimal schedule. Spend the first four minutes reading the small lesson and defining the objective. Use the next six minutes for two independent attempts. Spend five minutes checking the reasoning and correcting an error. Use another six minutes for changed examples, then finish with four minutes away from the chat.

The exact timing can change. A demanding passage may require more reading; a familiar concept may need almost none. What matters is preserving time for an independent ending. If every minute goes into generating questions and discussing answers, the session never checks whether the learner can now work without the assistant.

For the final attempt, close the explanation and write a brief answer to a fresh question. In our example, a rate rises from twelve percent to fifteen percent. The change is three percentage points and the relative increase is twenty-five percent. Explain both quantities in ordinary language before checking the arithmetic.

Record the outcome honestly. An independent correct answer suggests that this particular task went well today. It does not prove that the whole topic is mastered or that the knowledge will remain available next week. A wrong answer identifies what to revisit. Either result is more informative than the number of messages exchanged.

End by choosing one next action: review the definition, attempt another denominator question later, or move to the next lesson. A specific next action makes the following session easier to start. There is no need to ask the chatbot for a motivational study plan every time a small exercise ends.

Keep an error log that makes the next session shorter

An error log should be small enough to use. I would keep four fields: the question, the mistaken step, the corrected rule, and the next independent check. Avoid copying the entire conversation. Long transcripts make it difficult to find the one decision that caused the problem.

For the worked example, an entry might say: Used the ending percentage as the base; relative change uses the starting value; next time explain why a rise and reversal have different relative percentages. That note points directly to the misconception. It is more useful than saving a paragraph of general praise or a complete AI-generated summary of the session.

Return to a fresh question on a later day without reading the explanation first. The interval can fit the course schedule; this article does not prescribe a universal spacing formula. The important practical distinction is between remembering an answer immediately after seeing it and retrieving the method after some time has passed.

If the later attempt fails, revise the plan rather than treating the earlier session as wasted. Perhaps the definition was memorized but not connected to the calculation. Perhaps the example was too similar to the demonstration. Add a question that targets that weakness, check it, and schedule another independent attempt.

Retire entries when they no longer deserve disproportionate attention, while retaining occasional mixed checks. An error log should guide learning, not become a permanent catalogue of every mistake. Its value is in selecting what to practice next with less friction and more precision.

Topic: Percentage change versus percentage points
Attempt: Hint-assisted
Error: Divided by the ending value
Correction: Relative change uses the starting value
Next check: Explain an increase and its reversal without notes
Verification: Compare with the lesson and check arithmetic separately

Keep practice separate from assessed work

A practice question you invent is not the same as a live exam question or an assignment with restrictions. Check the teacher's or institution's AI policy before using a chatbot with assessed work. Permission to study with AI does not automatically mean permission to submit AI-generated answers, upload course materials, or obtain assistance during an examination.

For an essay-based subject, ask for practice questions about a permitted reading and attempt your own explanation. Do not turn a restricted assignment into a supposed study exercise while keeping its exact instructions and expected submission. When the boundary is unclear, ask the person setting the work what help is allowed.

Protect other people's information too. A classroom discussion, a peer's essay, or a workplace training case can contain names and private details. Use an authorized excerpt, remove unnecessary personal data, or create a clearly fictional practice example. The percentage exercises here need no real survey respondents or private records.

The best reason to use AI for studying is not that it can always produce an answer. It is that it can help prepare focused practice and discuss a specific mistake. The learner still needs a trustworthy source, a real attempt, a check on feedback, and a moment without the assistant. I would judge the session by that final independent work, not by how impressive the conversation looked.