Consider two students.
The first can answer any question you put to her about the French Revolution. Causes, chronology, key figures, historiographic debates. Her essays are clean. She will score in the top band of any assessment you design.
The second stumbles over the same material, but halfway through the unit he asks: “If the food shortages were the trigger, why didn’t the earlier famines produce a revolution?”
Every teacher recognises both students. Every assessment system we have is built to measure the first one and structurally blind to the second.
This is not a small oversight. It is arguably the central design flaw in how we measure thinking — and it has become significantly more consequential in the last few years, for reasons worth being precise about.
The measurement we never take
The research on how often students ask questions is old, thin, and consistent in a way that should bother us more than it does.
Graesser and Person, working in the early 1990s, found student questions in whole-class settings running at roughly 1.3 to 4 per hour, with a median around 3. Not per student, but per classroom. Dillon’s earlier observational work expresses the same finding from the other end: in classes averaging 26.7 students, the rate per individual student was about 0.11 questions per hour. Multiply that out and you get the same three questions an hour, from an entire room of young people, most of whom will contribute none. Dillon also found that questions made up only 6% of student classroom talk, and that fewer than 1% of those were genuine information-seeking questions rather than procedural ones (“is this going to be on the test?”).
These studies are decades old and the samples were modest; anyone citing them as settled contemporary fact is overreaching. But nothing in the intervening thirty years suggests the picture has improved, and the mechanism producing it is not mysterious.
Students don’t ask questions because questions are not on the test.
Everything a school measures, it rehearses. We assess answer production constantly, from every angle, with enormous sophistication. We have rubrics, exemplars, item banks, moderation processes, and standardized instruments refined over a century. For question production we have essentially nothing — no rubric, no exemplar, no assessment, and therefore no rehearsal. A student can pass through thirteen years of formal education having never once been evaluated on the quality of a question they generated.
We then express surprise that graduates are better at answering than asking.
Why this matters more now than it did five years ago
For most of the history of formal education, the ability to produce a good answer was scarce and valuable. Retrieval, synthesis, and articulate expression were genuinely difficult, and a person who could do them reliably had a durable advantage.
That scarcity has substantially collapsed. Competent answer-generation, summarize this, explain that, draft the memo, structure the argument, is now available to anyone with a browser, instantly and at near-zero marginal cost. This is not a prediction; it is the current state of things.
What has not been commoditized is knowing what to ask.
Every person who uses these systems well and every person who uses them badly is separated by roughly one variable: the quality of their prompt, which is to say the quality of their question. The people getting extraordinary output are the ones who can specify a problem precisely, notice what the answer assumes, and follow up on the weak part. The people getting mediocre output are asking vague questions and accepting the first fluent response.
That is not a technology skill. It is the same skill that separates a good doctor from a mediocre one taking a patient history, a good journalist from a stenographer, a good manager from a status-report collector, and a good scientist from a competent technician. In every one of those domains, the binding constraint has always been question quality. The last few years have simply removed the camouflage — because answer-production, the thing we were measuring, no longer distinguishes anybody.
If you were designing an education system from scratch this year, question formulation would be a core assessed competency. Instead it’s an unmeasured by-product that some students happen to acquire.
What it would take to actually teach this
The good news is that this is teachable, and unlike most things labelled “21st century skills,” there is real methodology behind it rather than a poster.
Teach question production as a discrete, practised skill.
The Right Question Institute’s Question Formulation Technique is the most developed protocol available and it is deliberately simple: give students a focused prompt, have them produce as many questions as possible under four rules — ask as many as you can, don’t stop to discuss or answer any of them, write them down exactly as stated, and change any statement into a question. Then have them categorize their questions as open or closed, convert some in each direction, and prioritize. What makes it work is the separation of generation from evaluation. Students who have been taught to self-censor produce nothing when both happen at once.
Build a taxonomy students can actually name.
Questions do different jobs: clarifying, probing assumptions, requesting evidence, testing implications, challenging the frame, and asking what would change one’s mind. Students who can name these categories start choosing between them rather than defaulting to whichever occurs first.
Assess questions directly.
This is the hard one, and it’s where most attempts fail. But a workable rubric exists: does the question target an assumption rather than a detail? Is it answerable in principle? Would answering it change something? A question log — every student maintaining a running record of questions generated across a unit, with a reflection on which turned out to be productive — is a defensible assessment artifact.
Make “what would change my mind?” routine.
Applied to their own position, it is the most demanding question a student can be asked, and the one that most reliably reveals whether a belief is held for reasons or for reasons-shaped feelings.
Give the quiet students a channel.
This is the unglamorous, high-return intervention. A large fraction of the missing questions are not missing because the student didn’t have one. They’re missing because asking a question in front of thirty peers carries a specific social risk — the risk of appearing not to know — and adolescents price that risk very highly. A written or anonymous submission channel makes the cost approximately zero, and the effect on volume is usually immediate and startling.
Where the tools come in, and their limits
If you want to work on question quality seriously, you need to be able to see the questions. That’s harder than it sounds, because in a normal classroom or lecture the questions are ephemeral: three get asked aloud, the rest evaporate, and nobody has a record of what the room was actually confused about.
Several audience-questioning platforms solve the collection problem well enough to be useful here — Slido, Mentimeter, Pigeonhole Live and VersyNow among them. The common mechanism is the same: attendees submit from their phones without an account, submissions can be anonymous, the room can upvote what it most wants answered, and — the part that matters pedagogically — you end the session with a full written corpus of every question that was asked rather than the three that reached the microphone.
Some of these, VersyNow included, will cluster that corpus into themes afterwards, which turns a session’s questions into something you can look at analytically: what did they actually not understand, and where.
The honest limit: none of this teaches question quality. It gives you the raw material. A hundred collected questions are worth exactly as much as the time you spend reading them with students and asking which ones were good and why. Software solves visibility. The instruction is still yours to do.
There’s a second caution worth naming. Anything assessed will be gamed, and question quality is unusually gameable, a student can learn the shape of an impressive-sounding question without the underlying curiosity. That risk is real. It’s also the risk we accept for every other assessed skill, and it’s not a reason to keep measuring nothing.
The reframe
The most useful sentence in this whole discussion is a reversal of how we usually think about assessment:
A student’s answers tell you what they’ve been taught. Their questions tell you what they’re doing with it.
We have spent a century building extraordinarily sophisticated instruments for the first thing.
That seems worth changing before another cohort passes through.
Frequently asked questions
Research from Graesser and Person found roughly 1.3 to 4 student questions per hour across an entire classroom, with a median near 3. Dillon’s observational work put the per-student rate at about 0.11 questions per hour, and found that questions made up only 6% of student talk, with fewer than 1% being genuine information-seeking questions. These studies are decades old, but no subsequent evidence suggests substantial improvement.
Two reasons compound. First, question quality is almost never assessed, so it is almost never rehearsed — schools rehearse whatever they measure. Second, asking a question publicly carries the social risk of appearing not to know, which adolescents price very highly. Removing either constraint raises question volume noticeably.
A protocol developed by the Right Question Institute in which students produce as many questions as possible about a focused prompt under four rules: ask as many as you can, don’t stop to discuss or answer them, write them exactly as stated, and turn any statement into a question. Students then classify questions as open or closed, convert examples in both directions, and prioritize. Its key design feature is separating question generation from question evaluation.
Competent answer-production is now available instantly and at near-zero cost, which means it no longer distinguishes anyone. What separates strong from weak use of these systems is the quality of the question asked — the ability to specify a problem precisely, notice what an answer assumes, and follow up on the weak part. This the skill that has always separated good practitioners from adequate ones in medicine, law, journalism and research.
Workable criteria: does the question target an assumption rather than a surface detail, is it answerable in principle, and would answering it change anything? A question log kept across a unit, with student reflection on which questions proved productive, is a defensible assessment artifact. Collecting questions in writing, including anonymously, gives you the corpus needed to do this at all.
