Wait, what? One of the hardest questions in a lesson may be the one the student has to invent.
Answering a question begins with a structure someone else has already built. Generating a question requires the learner to build part of that structure: identify something worth asking, define the target, decide what information belongs, and imagine what a defensible answer would need.
Quick Answer
Owned learner job: generate a question that accurately targets a relationship, uncertainty, distinction or application in the material being learned.
This is not the same as ordinary retrieval practice, where someone else supplies the prompt. It is also not the broad “generation effect” of attempting an answer before seeing one. Here the learning operation is the construction of the question itself.
Why question generation is cognitively interesting
To write a useful question, the learner has to inspect the knowledge structure. What is central? What is merely a detail? Which two ideas are easy to confuse? What condition changes the answer? What evidence would settle the issue?
A weak question can therefore be informative too. If every generated question asks only for a copied definition, the learner may not yet be seeing relationships, conditions or applications.
Observable learner states
- The learner can answer teacher questions but cannot invent one beyond a definition.
- The question is answerable directly by copying one sentence from the source.
- The learner creates a question whose answer is ambiguous because the target is unclear.
- The learner generates a strong question but cannot answer it independently.
- The learner can generate verification, comparison, explanation and application questions across the same topic.
None of these observations is a clinical diagnosis. They may reflect knowledge depth, language demands, familiarity with question forms, or the quality of the source material.
A discrimination test
Ask for three different questions about the same small section: one that checks a fact, one that tests a relationship, and one that changes a condition. If the learner can answer but not generate the latter two, the weak link may be in organising and interrogating the knowledge rather than simple availability of facts.
The Question Generation Protocol
1. Start with a small knowledge object. One paragraph, worked example, diagram or concept is enough.
2. Generate one question with a clear target. The learner should be able to say what the question is testing.
3. Write the answer separately. This exposes questions that sound sophisticated but have no stable answer.
4. Add a distractor or alternative. Ask what a plausible wrong answer would be and why it is wrong. This can expose the distinction the question is actually testing.
5. Change the question type. Move from recall to comparison, explanation, prediction or application where the subject allows it.
6. Remove the source. Later, generate a fresh question from memory and answer it. That separates source-dependent question writing from independent knowledge use.
How do we know?
Research on student-generated questions suggests potential benefits, but the evidence is not a licence to claim that question generation always improves achievement. Studies have found gains in some cognitive and metacognitive processes, while other work has reported no overall examination improvement even when higher-order question performance improved. For example, a study of 133 higher-education students found no statistically significant overall exam gain after question-generation activities, but did find improvement on higher-order thinking questions. Read the study.
Research with chemistry students also found that the kinds of questions students generated were meaningfully associated with comprehension monitoring and performance: verification and disambiguation questions were positively associated with exam performance. Read the study.
The safe educational conclusion is narrower: generating questions can be a useful operation for making learners inspect what they know, what they do not know, and how knowledge is structured. Its effect depends on the task, question quality, support and context.
Common misconception: difficult-looking questions are better
No. A long question with elaborate wording can test less than a short precise one. Quality comes from what distinction or relationship the question exposes, not from how intimidating it looks.
Parent and tutor guide
After a learner studies a page, resist asking all the questions yourself. Say: “Write one question that would tell me whether someone really understood this page.” Then ask the learner to explain why that question is diagnostic.
If the learner produces only factual prompts, model one relational question, then fade the model. The goal is not to turn the child into an examiner. It is to make the structure of knowledge visible.
Transfer and return test
On a new topic, can the learner independently generate a question that targets a meaningful relationship rather than copying surface wording? Can they answer it and explain what a plausible wrong answer would miss? That is stronger evidence that the operation transferred.
MindOS direction
Observe: learner answers supplied prompts but does not interrogate the material → Discriminate: compare answering with question construction → Operate: generate, answer and justify a question → Fade: remove question stems and source → Transfer: generate on a new topic → Return: does the learner now expose meaningful distinctions independently?
Neighbouring owners include Generation State, Elaboration State, and Metacognitive Monitoring State. This page owns the narrower operation of constructing the learning question itself.
eduKate Sengkang 