What makes a good question?

A good question changes what becomes visible. Three properties do most of the work: it is specific enough to be investigated, open enough that it does not smuggle in its own answer, and consequential enough that knowing would change what you do next. A question failing the third test can be interesting without being useful.

This is the most important open item on this research map and the one with the least written about it, which is itself the point. Answers have been studied exhaustively. The thing that produces them has not.

Question-asking is a method, not a temperament

The default assumption in most organisations is that curiosity is a disposition: some people have it and the rest can be encouraged. Rothstein and Santana's work is the strongest argument that this is wrong.

Their Question Formulation Technique gives students a structured protocol for producing their own questions, then improving and prioritising them: generate as many as possible without stopping to judge, decline to answer them while producing them, convert between open and closed forms to see what each reveals, then choose. They report more engagement and better metacognition than when the teacher supplies the questions. Make Just One Change (2011).

The specific move worth stealing is converting a question between open and closed form and watching what changes. "Did the pilot get it wrong?" and "What made that action look correct at the time?" are about the same event and produce entirely different investigations. One is answerable by assigning blame. The other is answerable only by reconstructing the situation.

Warren Berger's sequence, why then what if then how, is the popular articulation of a similar shape. A More Beautiful Question (2014). It is useful for the arc and thin as evidence; the teachable-method claim rests on Rothstein and Santana.

Three tests

Investigable. "Is AI good for society?" cannot be pursued. "Did unassisted detection rates change after this tool was introduced?" can, and did. The second is not a smaller question; it is the version of the first that admits of an answer.

Not self-answering. "How do we stop AI destroying junior careers?" has decided the destruction before asking anything. The estate's own version, whether the graduate market is changing because of AI, keeps the causal claim open and consequently has to report that the exposure evidence is real and the causal evidence is not.

Consequential. The test is whether any available answer changes a decision. If both answers lead to the same action, the question is decoration. This is the test that removes most questions in most meetings.

A question is not a prompt

These get conflated constantly and the difference is why prompt advice ages so badly.

A prompt is an instruction to a particular system. It depends on that system's behaviour and depreciates as the behaviour changes, which is the argument at why learn to prompt is weak career advice. A question is a claim about what is worth finding out. It holds regardless of what answers it, and it was worth asking before the system existed.

The practical consequence is that a well-formed question survives a model release and a well-formed prompt frequently does not.

What changed, and what did not

The properties above are old. What has changed is the price of the other half of the pair.

When answers were expensive, the binding constraint on thinking was getting one, and most institutional design went into answer production: libraries, research functions, analysts. Now that a fluent answer to almost any question is close to free, the constraint moves to which question gets asked. Models answer extremely well and have no questions of their own. Nothing in a system that responds to prompts will tell you that you are asking about the wrong thing.

That is a claim about where the scarcity sits rather than a measured finding about human curiosity, and it should be read as such. Whether people ask fewer or worse questions when answers are cheap is open, and this site lists it as open at does AI change the questions people ask.

What this page does not establish

Rothstein and Santana's evidence is classroom practice with reported outcomes rather than controlled trial, and it comes from school settings rather than professional ones. The technique's transfer to a boardroom or a research team is plausible and untested.

The three tests offered here are editorial rather than empirical. They are the ones this research uses to decide what goes on the question map, and no study establishes that questions meeting them produce better outcomes than questions that do not.

Key sources

• Rothstein, D. and Santana, L. (2011). Make Just One Change: Teach Students to Ask Their Own Questions. Harvard Education Press. In the essential works.
• Berger, W. (2014). A More Beautiful Question: The Power of Inquiry to Spark Breakthrough Ideas. Bloomsbury USA. In the essential works.
• Galef, J. (2021). The Scout Mindset. Portfolio. In the essential works.

Related SuperSkills research

On the capability this depends on, curiosity and critical thinking. On using a model against your own framing, how do I get AI to challenge me. On why prompting is the wrong thing to invest in, why learn to prompt is weak career advice. The whole public map of what this research is asking is at the questions.

About this research

Rahim Hirji is the author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company. Curiosity is one of the seven SuperSkills, and the question-formulation method described here is Rothstein and Santana's rather than this research's.

How this research works  ·  Reviewed quarterly  ·  Found an error? Tell me and it is corrected on the page.