Score the question that changes a condition, not the one that repeats the example

You asked a chatbot to explain an idea, read the example, and then answered a question that was almost the same example. That feeling is comfortable. It can also be the feeling of remembering the wording. A more useful check starts on the next question: if the situation changes, can you still use the same principle?

You do not need another study app. Turn the self-check into three steps: say it in your own words, apply it nearby, then change the context. The point is to find where the knowledge is stuck before you carry “I already know this” into an exam or a real task.

A constructed example: supply and demand, without repeating the slogan

Use an example that is entirely constructed. Suppose the lesson says: if other factors stay the same, and demand for a product rises while supply does not, pressure on the price can rise. You just saw an example about concert tickets. More people want to buy, and the number of seats stays fixed.

If the chatbot asks again, “When demand rises and supply stays the same, what tends to happen to the price?”, you can answer at once. That question helps recall. It is still very close to the sentence you just read.

Add two more steps.

Step	Constructed question	What you are checking
1. Say it yourself	Explain the relationship in your own words, without the phrase “supply and demand.”	Can you state the idea, or are you copying the keyword?
2. Apply it nearby	A bakery makes only 100 cakes a day, and suddenly more people want to buy them. If other factors have not changed, what can happen to the price, and why?	Can you connect the principle to a new example with a similar structure?
3. Change the context	Demand for the cakes rises, and the bakery also raises output from 100 to 180 cakes a day. From this information alone, can you say the price will certainly rise?	Do you notice that the original condition changed, and avoid applying the slogan automatically?

The third question is the important part of the routine. An answer like “demand rose, so the price rises” sounds familiar and contains the right keyword. It ignores that supply changed too. In this example, a good answer has to say that there is not enough information to claim the price will certainly rise, and explain why.

This is not a standard test of a chatbot, and it is not a measure of learning ability. It is a boundary case built so you can see the difference between remembering a sentence and using a principle.

Score 0–2 yourself. Do not ask the chatbot whether you understood

After you answer each question, and before you look at a solution, score on three levels:

• 0: you cannot explain it, or the explanation conflicts with the source principle.
• 1: you can state the main idea, but you lean on keywords, miss a condition, or cannot say why.
• 2: you say it in your own words, name the important condition, and use that condition to explain the conclusion.

The total is not an academic grade. The useful part is the pattern of the miss. If you score 2 on the recall question and 0 or 1 on the context-change question, do not generate ten more questions that look like the first one. Go back to the conditions of the idea, explain them once yourself, then try a new situation.

A well-known study by Roediger and Karpicke found that retrieving knowledge through testing can support long-term memory better than only rereading, under the conditions they studied. That does not mean every kind of quiz produces the ability to carry knowledge into every new situation. This article uses the context-change question as a check, not as a complete measure of understanding.

Let the tool write the questions, and keep the answers behind you

Any suitable chatbot can do this. If you use a study mode that can ask one question at a time, the routine is the same. Put the passage you are actually studying into the context and ask:

Write 3 check questions from this material.
Question 1 asks me to explain it in my own words.
Question 2 applies it to a nearby situation.
Question 3 changes at least one important condition, to see whether I am applying the idea automatically.
Ask one question at a time.
Do not give the answer or a hint until I reply.
After each question, ask me to point to evidence in the source material for my explanation.

The point to keep is that the answer does not appear before your first attempt. If the chatbot shows the solution immediately, put that solution off the screen, make a new question, and try again. If it marks you “correct” and the solution does not match the course material, the source material is the reference. The chatbot is not the final examiner.

For graded work, follow the rules of the school or the instructor. This routine is for studying and review on your own. It is not a way around a rule about using AI.

A five-minute pass after one section

After a section of a lesson, you can use this short loop:

1. Close the explanation or the sample answer.
2. Answer the “say it yourself” question in 60–90 seconds.
3. Do one nearby application.
4. Do one context-change question in which at least one condition has changed.
5. Open the source material, score yourself 0–2, and write down exactly one thing to study again.

If the context-change question is so hard that you do not know where to start, that is not a failure. Drop one step: go back to the nearby application, add a hint about the condition to watch, and try again. If all three questions are too easy, move the setting farther away, or ask for an explanation of an exception.

When you can stop

Do not stop because the chatbot says “Correct!” A more practical stopping point is this: you can explain the idea without looking at the answer, you can handle a new example, and you can see when a changed condition means the old slogan is no longer enough to conclude.

The next day, try a new question without opening the old conversation. The gap in time makes the check less dependent on a very short memory of the exact wording you just saw. If you can still explain the principle and the condition, that is a more useful signal than finishing another round of summaries.

Further reading

1. Retrieval Practice — The Learning Scientists — an introduction and practice resources on actively recalling knowledge instead of only rereading. Use it to understand the basis of the self-check step.

Primary sources

2. Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention — Roediger & Karpicke, Psychological Science (2006) — the experimental study behind the testing effect: under the conditions they studied, testing supported longer memory than restudying. This article does not turn that result into a promise that every context-change question improves learning.

3. Using study mode in ChatGPT — OpenAI Help Center — product documentation that study mode can ask one question at a time, create practice, and check understanding. OpenAI also notes that the mode can make mistakes, and that learners need to follow class and school rules.