How to use Super Intelligence when you do not know what question to ask begins with admitting that confusion often arrives before vocabulary. You may know that something is wrong, difficult, inefficient or uncertain without yet knowing how to frame it.
This is one of the strongest uses of SI. You do not need to arrive with a perfect prompt. You can begin with the situation, the symptoms, the constraints and what you are trying to change. The assistant can help discover the question before trying to answer it.
In the eduKateSG life series, Super Intelligence, or SI, is our editorial name for practical AI assistance. This guide follows How to Use Super Intelligence to See a Problem from Multiple Perspectives. That article expands viewpoint. This article expands the question space itself.
The core principle is: when you do not know the question, describe the situation honestly enough that the question can emerge.
The Problem Before the Question
Many real problems arrive as symptoms.
“I am overwhelmed.”
“My child is struggling.”
“This project keeps slipping.”
“I am spending too much.”
“I do not know whether this course is worth it.”
“Something about this workflow feels wrong.”
These statements are useful starting signals, but they are not yet questions.
The mistake is to force an answer-shaped prompt too early: “How do I fix everything?” “What should I buy?” “Which course is best?”
A stronger SI interaction starts one level earlier: help me identify what I need to understand before I can ask the right question.
The Question-Discovery Funnel
Use seven stages: Situation → Friction → Desired Change → Unknowns → Decision → Evidence → Next Question.
Situation
What is happening? Describe events without immediately deciding what they mean.
Friction
Where does the situation become difficult, slow, confusing, costly or risky?
Desired change
What would you like to be different?
Unknowns
What do you not yet know that could explain the gap?
Decision
What choice or action will eventually need to be made?
Evidence
What source, observation or test could reduce uncertainty?
Next question
Which question now has the highest value because its answer could change the next action?
This funnel turns “I do not know what to ask” into a sequence that produces a question from the situation itself.
The First Prompt When You Have No Question
Try this: “I do not yet know the right question. Here is the situation: __. Here is what feels difficult or important: __. Here is what I would like to improve: __. Do not solve it yet. Help me identify the main problem types this could contain, the information I am missing and the questions whose answers would most change what I do next.”
This prompt stops SI from rushing into solution mode. It also gives the assistant permission to say that the original framing may contain several separate problems.
Question Types: Know What Kind of Question You Need
Descriptive question
What is happening? Useful when the state itself is unclear.
Definitional question
What does this term mean in this context? Useful when vocabulary is blocking understanding.
Diagnostic question
Where is the failure or bottleneck?
Causal question
What may be causing the observed outcome?
Comparative question
How do the options differ on criteria that matter?
Predictive question
What may happen under these assumptions?
Decision question
What should I do given goals, evidence and constraints?
Verification question
How can I check whether this claim or action is correct?
Design question
How should the system or workflow be structured?
Constraint question
What limits the feasible options?
Learning question
What prerequisite, misconception or transfer gap blocks performance?
Reflection question
What did the previous attempt teach us?
SI can classify a vague concern into one or more of these types. The classification often reveals why the original question felt hard to form.
The Wrong-Question Test
A question can be grammatically clear and strategically wrong.
“Which laptop should I buy?” may be wrong if you have not established that a new laptop is necessary.
“How do I motivate my child?” may be wrong if the child is confused by one prerequisite rather than lacking motivation.
“How can I automate this?” may be wrong if the process should be removed or simplified first.
“How do I become more productive?” may be wrong if the actual issue is too many commitments.
Ask SI: “What assumptions are built into my current question? What alternative question would I ask if those assumptions were false?”
This does not mean every question needs philosophical deconstruction. Use the test when the proposed answers keep feeling unsatisfactory.
The Question Ladder
Questions can be arranged from broad to actionable.
Broad: Why am I struggling with Mathematics?
Narrower: Which topics produce the most errors?
Diagnostic: What is the first unstable step in algebra questions?
Mechanism: Is the error caused by negative-number control, bracket expansion or method selection?
Test: Which short question distinguishes these possibilities?
Action: What should the next practice block target?
Review: What result would show that the repair transferred?
When the broad question feels impossible, descend the ladder until the next question can be answered with available evidence.
Vocabulary Discovery: Sometimes You Cannot Ask Because You Do Not Know the Terms
A missing word can block a missing question.
You may know that a plan depends on one thing happening before another without knowing the term dependency. You may notice that an option keeps future choices open without knowing the term reversibility or option value. You may sense that one task constrains everything else without knowing bottleneck.
SI is highly useful for vocabulary discovery.
Prompt: “I can describe the phenomenon but I do not know the technical term. Here is what I mean: __. Give me the most relevant terms, explain the differences and show which term best fits this situation.”
The new vocabulary creates new search and reasoning possibilities. This is one reason domain language matters: better words create better questions.
The Concept-Neighbourhood Method
When you learn one useful term, ask what concepts sit around it.
If you discover bottleneck, neighbouring concepts may include throughput, capacity, queue, dependency and constraint.
If you discover opportunity cost, neighbouring ideas may include trade-off, sunk cost, marginal benefit and option value.
Ask SI to build a small concept neighbourhood rather than an encyclopedia. The goal is to reveal adjacent questions you did not know existed.
Information-Gain Questions
The best next question is often the one whose answer changes the most.
Suppose three course options are close, but the decision depends heavily on whether one includes meaningful project feedback. Instead of researching another ten general features, ask the provider how project feedback actually works.
Suppose a student may have either a concept gap or time-pressure problem. One untimed diagnostic set can distinguish them better than more broad revision.
Ask SI: “Which question has the highest information value because its answer would eliminate options, change the diagnosis or alter the next action?”
This prevents research from expanding evenly across every unknown.
The Decisive-Question Test
A decisive question is not necessarily the most interesting question. It is the question most likely to change what you do.
Career example: not “What are the trends in this industry?” but “Which capability is repeatedly required in the roles I am genuinely considering?”
Purchase example: not “Which product has more features?” but “Does the current device already meet the required workload?”
Learning example: not “Why is algebra difficult?” but “Can the learner expand brackets accurately without hints?”
The decisive-question test compresses uncertainty toward action.
Worked Example: My Child Is Struggling in School
Starting statement: “My child is struggling and I do not know what to ask.”
SI first separates possible domains: academic understanding, execution, workload, motivation, health, social context and school transition.
The parent adds observation: Mathematics score fell, English is stable, homework takes longer and errors cluster in algebra.
The question space narrows. A weak next question is “How do I improve all grades?” A stronger diagnostic question is “What is the first repeated algebra error, and which prerequisite does it depend on?”
Evidence can come from recent working, short diagnostic questions and teacher or tutor observation. The next action is to target the demonstrated gap and retest.
The parent did not need the perfect question at the beginning. The question emerged from progressively better description.
Worked Example: I Hate My Job but Do Not Know Why
Start with situation rather than a resignation decision.
SI asks categories: task content, manager, workload, autonomy, commute, values, growth, compensation and instability.
The user notes that pay is acceptable and colleagues are fine, but work is constantly interrupted and decisions are tightly controlled.
The better question becomes: “Is the main source of dissatisfaction lack of autonomy, fragmented work or the occupation itself?”
Next evidence: compare days with protected deep work versus highly interrupted days; ask what decisions the role could realistically own; explore adjacent roles with greater control.
The original “Should I quit?” question was too early.
Worked Example: Our Family Is Always Rushed
SI maps possible causes: overcommitment, travel, poor preparation, task ownership, transition time, unpredictable school notices and unrealistic scheduling.
The family observes that delays concentrate on mornings when bags and documents are prepared at the last minute.
The decisive question becomes: “Which items can be moved to a previous-evening preparation checklist, and who owns each one?”
The intervention is smaller than a complete family productivity system. Question discovery prevents overengineering.
Worked Example: Should I Buy This?
The user begins with a product link and uncertainty.
SI should not begin with reviews. It should ask what problem the purchase solves.
The user says the current device is slow during one specific application. Now the question sequence becomes: Does the current device meet the application’s requirements? Can the bottleneck be repaired? Will the new device materially improve that workload? How often does the workload occur?
The product comparison comes last. A good question-discovery process may end with “do not buy yet”.
Worked Example: Preparing for a Doctor’s Appointment
The person has symptoms but does not know what medical question to ask.
SI can help organise factual observations: onset, frequency, duration, triggers, medication, relevant history and impact. It should not decide the diagnosis.
Useful questions for the clinician may include: What are the plausible categories that need evaluation? Which warning signs matter? What information would help distinguish them? What tests or observations are appropriate?
The output is better preparation for a qualified healthcare professional, not a replacement for one.
The Question Tree
When several explanations remain plausible, build a question tree.
Start with one branching question that separates the problem space.
Example: “Can the student solve the same algebra skill accurately when untimed?” If no, investigate understanding or prerequisite. If yes, investigate time pressure, mixed-topic recognition or execution under assessment conditions.
Each answer selects the next branch. Question trees are efficient because later questions depend on earlier evidence rather than being asked all at once.
The Question Queue
Complex projects can contain many unresolved questions. Keep a queue instead of trying to answer all of them immediately.
Each question should include why it matters, who or what can answer it, whether it blocks progress and what happens if it remains unknown.
Sort by decision impact, not curiosity. SI can help maintain the queue, but the queue should shrink as evidence arrives.
Unknown Unknowns: Questions You Do Not Know Exist
You cannot directly list an unknown unknown. You can create conditions that expose one.
Ask for adjacent concepts, common failure modes, hidden dependencies, exception cases, another perspective and surprising observations that would indicate the current model is incomplete.
These prompts expand the edge of the known problem without pretending to enumerate everything unknown.
The Scope Question
Some questions are hard because the scope is wrong.
“How does education work?” is too broad for a practical decision. “Why did this student fail this one algebra question?” may be too narrow if the same error appears across several chapters.
Ask SI to propose three scope levels: narrow, working and system level. Choose the level whose answer can change the next action.
The Unit-of-Analysis Question
Ask what object you are actually analysing: one student, one class, one family, one transaction, one project, one month, one city or one product.
Confusion often comes from mixing levels. Evidence about national trends does not automatically describe one child. A single customer complaint does not automatically describe the market. One difficult week does not automatically describe a career.
SI can help identify when the unit of analysis changes inside the conversation.
The Time-Frame Question
A question can change depending on time horizon. “Is this expensive?” today may differ from total cost over five years. “Is this study method working?” after one day differs from retention after a month.
Ask: “What time horizon matches the outcome I care about?” The correct question often appears only after the horizon is specified.
Question Quality Tests
- Does the question have a clear object?
- Can an answer change what I do?
- Is the scope appropriate?
- Are key terms defined?
- Does the question hide an assumption?
- Does it confuse fact, value or prediction?
- Can evidence answer it?
- Is another question logically prior?
- Is it asking a model to know another person’s mind?
- Would a smaller test answer it faster?
The Minimum Viable Question
When a topic is overwhelming, ask the smallest question that can unlock the next step.
Instead of “How do I plan my career?” ask “Which capability appears most often in the roles I am genuinely considering?”
Instead of “How do I fix our family schedule?” ask “Which repeated morning delay happens most often, and what event immediately precedes it?”
Instead of “How do I master Mathematics?” ask “Which prerequisite is causing the current algebra error?”
The minimum viable question creates movement without pretending to solve the whole system at once.
The Next-Best-Question Rule
Do not search for the perfect final question. Search for the next question that increases information enough to improve the next action.
Question → evidence → updated model → better question.
SI is valuable because it can help generate the next question quickly after each update.
When to Stop Asking Questions
Questioning can become a form of avoidance.
Stop when the remaining unknowns have low decision impact, the next action is reversible, further evidence costs more than it is worth or a real-world test will teach more than discussion.
For high-consequence irreversible decisions, the stopping threshold should be higher.
The purpose of question discovery is action under clearer understanding, not endless inquiry.
A Copyable Question-Discovery Prompt
“I do not yet know the right question. Here is the situation, what feels wrong, what I want to change and what I already know. Do not solve it immediately. Classify the possible problem types. Identify key terms I may be missing. Separate facts, assumptions and unknowns. Show me the decision that eventually has to be made. Generate a question tree and identify the next one or two questions with the highest information value. Tell me what evidence could answer them.”
The Anatomy of a High-Value Question
A high-value question usually contains five things: an object, a boundary, a reason, a source of evidence and a consequence for action.
Object: what exactly are we asking about?
Boundary: what is inside and outside the question?
Reason: why does the answer matter?
Evidence: what could answer it reliably?
Action: what would change depending on the answer?
Compare “Why is my child bad at algebra?” with “Which algebra step fails first on unfamiliar questions, and what prerequisite would explain that failure?” The second question has a clearer object, avoids a fixed identity label, points toward evidence and leads to an intervention.
Ask SI to inspect a proposed question against these five fields. If one is missing, the question may still be useful, but you know where ambiguity remains.
Symptom, Problem, Cause, Decision and Test Are Different
One of the fastest ways to discover the right question is to separate five layers.
Symptom: what is visibly wrong?
Problem: what system or capability is failing?
Cause: what may explain the failure?
Decision: what choice must be made?
Test: what evidence can distinguish the possibilities?
Example: symptom—student score falls. Problem—performance is unreliable on mixed algebra questions. Possible cause—method selection or bracket expansion. Decision—which skill should the next week target? Test—short untimed diagnostic questions followed by a mixed transfer set.
When the user says “I do not know what to ask”, SI can classify the situation into these layers before proposing questions.
Exploration Questions and Decision Questions
Early in a problem, exploration is valuable. Later, exploration can become delay.
Exploration questions
What categories could this problem belong to? What concepts am I missing? What alternatives exist? What would an expert check? What does this pattern resemble?
Decision questions
Which option meets the must-haves? Which unknown could change the choice? What is the smallest reversible action? What should be verified before committing?
Use exploration until the decision structure becomes visible. Then switch to decision questions. A conversation that keeps expanding after the decisive unknown is known is no longer helping.
The Explore–Narrow–Test Cycle
Explore: generate plausible problem categories and missing concepts.
Narrow: identify which explanation or decision matters most.
Test: ask the question whose answer separates the remaining possibilities.
After the test, repeat. This cycle is stronger than trying to design the perfect question tree in advance because each answer changes the next branch.
Question Discovery Through Contrast
When you cannot define something directly, compare it with a nearby alternative.
“Is the student unable to understand the method, or able to understand it but unable to select it independently?”
“Is the job problem the occupation itself, or the current manager and working conditions?”
“Is the workflow slow because generation is slow, or because verification and source retrieval dominate?”
Contrast questions are powerful because they turn vague dissatisfaction into competing hypotheses.
Ask SI to generate contrasts that would lead to different actions. Avoid contrasts where both answers lead to the same next step.
Question Discovery Through Exceptions
Exceptions reveal structure.
If a student usually struggles but succeeds on one type of question, ask what is different. If a workflow is slow except on certain days, ask what changes. If family mornings are chaotic except on Fridays, identify the Friday condition.
Prompt: “Identify the cases where the problem does not occur. What variables differ, and what question would test whether one of those variables is causal?”
This method is often more informative than examining failure alone.
Question Discovery Through Sequence
For process problems, ask where the first unstable transition occurs.
A school task may move through read → interpret → choose method → calculate → check. A project may move through request → clarify → approve → execute → verify. A household renewal may move through notice → extract requirement → assign owner → submit → confirm.
Ask SI to map the sequence and identify the earliest step where reality diverges from the intended process.
The best next question often belongs at that earliest unstable step, not at the final visible failure.
Question Discovery Through Constraints
Sometimes the correct question is not about the ideal solution. It is about the binding constraint.
“What can I learn fastest?” becomes less useful when only two evenings are available. “Which learning route fits four sustainable hours a week?” is stronger.
“What is the best family schedule?” becomes less useful when school, work and sleep already consume most hours. “Which commitments can move or be removed?” is more decisive.
Ask SI: “Which constraint makes most of the option space irrelevant?”
Question Discovery Through Stakeholders
A missing perspective can imply a missing question.
If you are planning a household system, what question would the person maintaining it ask? If you are choosing school support, what would the student ask? If you are automating work, what would the operator handling exceptions ask?
Use multiple-perspective analysis to generate stakeholder questions, then verify them with real people when the answer matters.
The purpose is not to invent perspectives. It is to reveal questions your own role may not naturally produce.
Question Discovery Through Evidence Gaps
Look at your current conclusion and ask which claim has the weakest support.
Suppose you think a new job will provide more autonomy. Salary and commute are confirmed, but autonomy is inferred from the title. The evidence gap produces the next question: “Which decisions does this role actually own?”
Suppose you think extra tuition will help. The evidence gap is whether the student needs more teaching or a narrower repair. The next question is diagnostic.
Question discovery becomes much easier when you ask where the evidence is weakest and decision impact is highest.
Question Discovery Through Assumptions
Take the current plan and ask what must be true for it to work.
Plan: automate a weekly report. Assumptions: source data is stable, verification remains cheap, recipients still want the report, permissions are appropriate.
Now ask which assumption is load-bearing and least verified. That assumption generates the next question.
This links question discovery with assumption challenge.
The Question-Value Matrix
When many questions compete, compare them on four dimensions.
Decision impact: could the answer change what you do?
Answerability: can reliable evidence be obtained?
Cost: how much time, money or effort does the answer require?
Dependency: does another question need to be answered first?
High-impact, answerable, low-cost questions usually come first.
SI can rank the question queue against this framework. The ranking is advisory; the user decides which cost or consequence matters most.
The Question Portfolio
For a large project, keep different question classes rather than one undifferentiated backlog.
Blocking questions: must be resolved before progress.
Risk questions: could reveal serious downside.
Learning questions: improve the model over time.
Optimisation questions: make a working solution better.
Curiosity questions: interesting but non-essential.
This classification protects the project from spending most of its attention on interesting but non-blocking questions.
A Full Case Study: Choosing a New Career Direction
Starting state: “I feel stuck and do not know what to ask about my career.”
Situation: current work is stable, pay is adequate, but growth feels limited.
Friction: boredom, low autonomy and uncertainty about alternatives.
Desired change: work that produces stronger learning and more control without a large immediate income drop.
Question discovery begins by separating occupation, employer, manager, role design and skill trajectory.
First high-value question: “Would the same occupation feel acceptable with more autonomy and a stronger learning environment?”
Evidence: compare current tasks with adjacent roles, speak with people in those roles and identify which responsibilities feel attractive.
Second question: “Which missing capability prevents me from testing one of those adjacent roles through a small project?”
The process turns an existential career question into a sequence of evidence-producing experiments.
A Full Case Study: A Student Says Science Is Too Hard
Starting state: “Science is too hard.”
SI asks for examples rather than explanations.
The student can recall definitions but loses marks on open-ended explanation questions.
The next question becomes: “Is the main gap scientific understanding, vocabulary precision or answer structure?”
A short diagnostic asks the student to explain one familiar process orally, then write it, then compare required key terms.
If oral explanation is accurate but written response lacks scientific vocabulary, the next question changes again: “Which terms are recognised but not actively retrieved during writing?”
The original complaint produced a vocabulary-transfer question only after evidence narrowed the field.
A Full Case Study: A Workflow Feels Too Complicated
Starting state: a personal SI system contains many prompts, notes and connected tools. The user says, “I do not know what is wrong, but this is becoming harder to manage.”
Possible question types include maintenance, duplication, source-of-truth, permissions, active-project load and retrieval.
The first decisive question is not “Which tool should I replace?” It is “Where do I spend the most avoidable maintenance time each week?”
Observation shows that the user repeatedly updates the same project status in three places.
The next question becomes: “Which one record should own project state, and which other systems can reference rather than duplicate it?”
Again, the question emerges from the failure pattern rather than from tool comparison.
Question Discovery for Research
Research often begins with a topic rather than a question.
Topic: AI and education.
Possible question families: learning outcomes, teacher workload, assessment integrity, student agency, privacy, access, curriculum change, labour-market preparation.
SI can help turn the topic into a question by specifying population, intervention, outcome, time period and geography.
Example: “How does guided generative-AI use affect independent writing performance among secondary students over one school term?”
That question is still only useful if appropriate evidence exists. Question quality includes answerability.
Question Discovery for Personal Decisions
Personal decisions often mix factual and value questions.
“Should I move?” contains factual questions about cost, commute and schools; uncertainty about future work; and value questions about community, space and lifestyle.
SI should separate those layers before building a recommendation.
The factual questions can be researched. The value questions belong to the people affected. The uncertainty can be modelled through scenarios.
The right question is often a bundle, not one sentence.
Metacognitive Questions: Ask About Your Own Thinking
When the topic remains stuck, ask a question about the reasoning process.
“What am I assuming?”
“Which answer am I hoping to hear?”
“What evidence am I avoiding because it would complicate the decision?”
“Am I asking a factual question when the real issue is a value choice?”
“Am I trying to predict something that would be better tested through a reversible action?”
SI can act as a mirror here, but treat its interpretations as prompts for reflection rather than psychological diagnosis.
The Question Reset
If a long conversation has become messy, reset the question space.
Prompt: “Ignore the questions we have already asked. Based on the current facts and goal, what are the three questions that now have the highest decision value? Explain why each matters and what evidence would answer it.”
This prevents old questions from surviving after the model of the problem has changed.
A Question-Discovery Ledger
For complicated work, record the evolution of the question.
Starting concern: what felt wrong.
First question: initial framing.
Evidence gained: what was learned.
Question changed to: better framing.
Decision affected: what action became clearer.
Open question: what remains unresolved.
The ledger is useful in learning, research and long projects because it records how understanding changed rather than only storing answers.
The Question Review After Action
After the outcome is known, ask whether the decisive question was the right one.
Did it identify the real bottleneck?
Which question turned out to be irrelevant?
Which missing question created the biggest surprise?
Which term or concept would have helped earlier?
Which evidence source answered the question most reliably?
This review improves the question-discovery process itself.
A 30-Day Question-Discovery Practice
Week 1 — Describe before asking: practise turning complaints into observations, friction and desired change.
Week 2 — Classify: label questions as descriptive, diagnostic, causal, comparative, predictive, decision, verification or design.
Week 3 — Prioritise: use information value, question dependencies and decisive-question tests.
Week 4 — Transfer: use the method across one learning problem, one work problem, one household problem and one personal decision. Review which question actually changed action.
At the end of the month, keep the question templates that improved diagnosis and retire prompts that merely produced longer conversations.
The Question-Discovery Exit Test
Before leaving question discovery, confirm that the current question has a clear object, appropriate scope, identifiable evidence and a connection to action.
You should know which unknown matters most and what would answer it.
If the question is factual, you know where to verify it. If it is diagnostic, you know what test distinguishes the main possibilities. If it is a value question, you know that evidence alone will not decide it. If it is a decision question, you know the relevant constraints and trade-offs.
The goal is not to discover the most impressive question. It is to discover the question that makes the next useful action possible.
Question Debugging: Why a Good-Looking Question Can Still Fail
When SI keeps producing weak answers, debug the question before blaming the answer. A question can sound specific while still hiding the wrong object, the wrong scope or the wrong success condition.
The object is unclear
“How do I improve this?” does not identify what “this” is. Name the object: a study routine, a paragraph, a household workflow, a budget or a decision.
The success condition is missing
“How do I revise algebra?” becomes stronger when the target is “solve mixed Secondary 1 algebra questions independently under timed conditions”.
The question contains the answer
“Why is this new app the best way to organise my life?” assumes the conclusion. Replace it with “What problem does this app solve, what alternatives exist and what evidence would justify switching?”
The evidence standard is missing
“Is this claim true?” becomes more useful when you specify what source would verify it and which date, population or context matters.
The decision link is missing
A question can produce interesting information without affecting action. Ask what you would do differently depending on the answer.
Use SI to debug by asking: “Why might this question produce vague, biased or non-actionable answers? Rewrite it three ways: diagnostic, decision-oriented and evidence-oriented.”
Question Inversion
Sometimes the most revealing question is the inverse of the one you started with. “How can I fit more into my week?” becomes “What should I remove so the important work fits?” “How can SI answer this faster?” becomes “Which part of the workflow should not require SI at all?”
“How can my child study more?” becomes “Which study activity currently produces the least learning value?” “How can I choose the best option?” becomes “What would disqualify an option before comparison?”
Inversion is useful when the original question assumes that more, faster or bigger is automatically better.
Counterfactual Questions
A counterfactual asks what would be different if one assumption changed. “If the student understood the concept but still scored poorly, what would I expect to observe?” “If the new course credential had no employer value, would I still want the course?” “If this automation saved no time but reduced errors, would it still be worthwhile?”
Counterfactuals generate sharper diagnostic questions because they force the reasoning to predict different evidence under different explanations.
Socratic Questioning with SI
Socratic questioning is useful when the goal is to examine reasoning rather than receive a conclusion. Ask SI to use a bounded sequence: What do you currently believe? What evidence supports it? What assumption connects the evidence to the conclusion? What alternative explanation fits the same facts? What evidence would distinguish them? What next action would create that evidence?
Do not turn every conversation into endless Socratic interrogation. Sometimes a direct explanation is appropriate. The method is most useful when the user needs to inspect their own model.
Question Discovery for Learning
In learning, the wrong question is often “Can you explain this again?” A better sequence is: What can I already do? Where does my working first become unstable? Is the problem recall, concept, method selection, execution, notation or transfer? What changed question would test the same skill without copying the example?
SI can help identify the question from a learner’s attempt rather than from the topic title. A student may say “I do not understand simultaneous equations” while their work shows that substitution is accurate but method selection is weak. The useful question then becomes “What features of the equations indicate which elimination method is efficient?”
That question teaches selection, not only procedure.
Question Discovery for Writing
Writers often ask “Can you improve this?” before deciding what the paragraph is supposed to do. Ask first: What is the claim? What should the reader understand or do next? Which evidence supports the claim? What objection must be answered? Which sentence contains information that does not serve the purpose?
Once those questions are clear, editing becomes easier because “better” has an operational meaning.
Question Discovery for Workflows
When a workflow feels inefficient, avoid asking only “How do I automate it?” Ask: What outcome is the workflow supposed to produce? Which step consumes the most effort? Which step creates the most errors? Which information is repeatedly reconstructed? Which handoff fails? Which part could be removed?
SI can map the workflow and convert each failure point into a question. The highest-value question may reveal that the workflow should be simplified before automation begins.
Question Discovery for Decisions Under Uncertainty
When the future is uncertain, do not ask for a single confident prediction if a scenario question would be more honest. Instead of “Will this career be good in five years?” ask “Which capabilities remain useful across several plausible five-year scenarios?”
Instead of “Will this AI workflow keep saving time?” ask “Which source changes, permissions or maintenance costs would erase the benefit?” Scenario questions replace false certainty with robust decision criteria.
The Question Hierarchy for High-Stakes Decisions
For consequential decisions, use a stricter order: What exactly is being decided? Which facts are authoritative? Which assumptions are load-bearing? Which unknowns can change the choice? Which qualified expertise is required? Which options are reversible? What downside is difficult to recover from? What evidence is sufficient before action?
In medical, legal or major financial matters, SI can help organise these questions, but the final individual judgement may require appropriate qualified professionals.
The Question Handoff to a Human Expert
One of the most useful outcomes of SI can be a better conversation with a teacher, doctor, lawyer, financial professional, manager, tutor or service provider.
A strong handoff includes the situation, relevant facts, what has already been tried, the uncertainty and the exact question requiring their expertise. Do not use SI to simulate expert certainty when the better move is to prepare for real expert input.
The Question Handoff to Action
Once the question is answered enough, convert it into a next action. Diagnostic answer → targeted practice. Comparison answer → verify one decisive fact or choose. Workflow answer → remove, simplify or redesign one step. Stakeholder answer → update the plan or negotiate the trade-off.
Question discovery should terminate in evidence, decision or action—not another decorative prompt.
The Question-Value Matrix
When many questions compete, compare them on four dimensions: decision impact, answerability, cost and dependency.
Decision impact: could the answer change what you do?
Answerability: can reliable evidence be obtained?
Cost: how much time, money or effort does the answer require?
Dependency: does another question need to be answered first?
High-impact, answerable, low-cost questions usually come first. SI can rank the question queue against this framework, while the user decides which consequence and cost matter most.
The Question Portfolio
For a large project, keep different question classes rather than one undifferentiated backlog: blocking questions, risk questions, learning questions, optimisation questions and curiosity questions.
This classification protects the project from spending most of its attention on interesting but non-blocking questions.
A Full Case Study: Choosing a New Career Direction
Starting state: “I feel stuck and do not know what to ask about my career.” Current work is stable, pay is adequate, but growth feels limited. The friction is boredom, low autonomy and uncertainty about alternatives. The desired change is stronger learning and more control without a large immediate income drop.
Question discovery begins by separating occupation, employer, manager, role design and skill trajectory.
First high-value question: “Would the same occupation feel acceptable with more autonomy and a stronger learning environment?” Evidence can come from adjacent roles, practitioner conversations and a small project.
Second question: “Which missing capability prevents me from testing one of those adjacent roles through a reversible experiment?”
The process turns an existential career question into a sequence of evidence-producing steps.
A Full Case Study: A Student Says Science Is Too Hard
Starting state: “Science is too hard.” SI asks for examples rather than explanations. The student can recall definitions but loses marks on open-ended explanation questions.
The next question becomes: “Is the main gap scientific understanding, vocabulary precision or answer structure?”
A short diagnostic asks the student to explain one familiar process orally, then write it, then compare required key terms.
If oral explanation is accurate but written response lacks scientific vocabulary, the next question changes again: “Which terms are recognised but not actively retrieved during writing?”
The original complaint produced a vocabulary-transfer question only after evidence narrowed the field.
A Full Case Study: A Workflow Feels Too Complicated
A personal SI system contains many prompts, notes and connected tools. The user says, “I do not know what is wrong, but this is becoming harder to manage.”
Possible question types include maintenance, duplication, source-of-truth, permissions, active-project load and retrieval.
The first decisive question is not “Which tool should I replace?” It is “Where do I spend the most avoidable maintenance time each week?”
Observation shows that the user repeatedly updates the same project status in three places. The next question becomes: “Which one record should own project state, and which other systems can reference rather than duplicate it?”
The question emerges from the failure pattern rather than from tool comparison.
The Question Reset
If a long conversation has become messy, reset the question space.
Prompt: “Ignore the questions we have already asked. Based on the current facts and goal, what are the three questions that now have the highest decision value? Explain why each matters and what evidence would answer it.”
This prevents old questions from surviving after the model of the problem has changed.
A Question-Discovery Ledger
For complicated work, record the evolution of the question: starting concern, first question, evidence gained, question changed to, decision affected and open question.
The ledger is useful in learning, research and long projects because it records how understanding changed rather than only storing answers.
The Question Review After Action
After the outcome is known, ask whether the decisive question was the right one. Did it identify the real bottleneck? Which question turned out irrelevant? Which missing question created the biggest surprise? Which term or concept would have helped earlier?
This review improves the question-discovery process itself.
A 30-Day Question-Discovery Practice
Week 1 — Describe before asking: practise turning complaints into observations, friction and desired change.
Week 2 — Classify: label questions as descriptive, diagnostic, causal, comparative, predictive, decision, verification or design.
Week 3 — Prioritise: use information value, question dependencies and decisive-question tests.
Week 4 — Transfer: use the method across one learning problem, one work problem, one household problem and one personal decision. Review which question actually changed action.
At the end of the month, keep the question templates that improved diagnosis and retire prompts that merely produced longer conversations.
The Question-Discovery Exit Test
Before leaving question discovery, confirm that the current question has a clear object, appropriate scope, identifiable evidence and a connection to action.
You should know which unknown matters most and what would answer it. If the question is factual, you know where to verify it. If it is diagnostic, you know what test distinguishes the main possibilities. If it is a value question, you know that evidence alone will not decide it. If it is a decision question, you know the relevant constraints and trade-offs.
The goal is not to discover the most impressive question. It is to discover the question that makes the next useful action possible.
The Diagnostic Question Bank
When you cannot find the right question, borrow a diagnostic family rather than inventing from zero.
State: What is happening now?
Difference: What is different between successful and unsuccessful cases?
Sequence: Where does the process first deviate?
Dependency: What must be true before this step can work?
Constraint: What limits the available options?
Evidence: What observation would distinguish the main explanations?
Ownership: Who can change the failing part?
Threshold: What condition flips the decision?
Transfer: Does the capability survive when the surface changes?
Review: What result would show that the intervention worked?
These question families transfer across education, work, family, personal finance, travel and AI workflows. The wording changes; the diagnostic structure remains useful.
The Question Dependency Map
Questions themselves can have prerequisites.
“Which option should I choose?” may depend on “What outcome am I trying to produce?” and “Which criteria are non-negotiable?”
“How much should I automate?” may depend on “What happens if the task is wrong?” and “Can I verify completion?”
“How much more should the student practise?” may depend on “Which skill is failing?” and “Is the current practice targeting it?”
Ask SI to draw the dependency order among questions. Answer the upstream questions first. This prevents late discovery that the entire comparison was built on an undefined goal.
The Question Escalation Rule
Some questions should move out of a general SI conversation and into a real source or professional channel.
Escalate when the question concerns individual medical diagnosis or treatment, consequential legal interpretation, regulated financial advice, another person’s consent, an official requirement whose current wording matters, or any matter where the user cannot reasonably verify the answer alone.
SI can still help formulate the question, organise facts and prepare the handoff. The escalation itself is part of good question discovery.
The Evidence-Producing Question
A strong question does not only request information. It can define an experiment.
“Does the student understand bracket expansion?” becomes “Can the student solve three changed bracket-expansion questions without hints and explain the sign changes?”
“Is this productivity app useful?” becomes “Does the current workflow still lose time after I remove duplicate lists and use one source of truth for two weeks?”
“Would I enjoy management?” becomes “Can I lead one bounded cross-team project and compare the experience with my specialist work?”
Evidence-producing questions are powerful because the answer arrives through reality rather than speculation.
The Question Compression Rule
After a long inquiry, compress the current state into three questions: What do I know now? What remains load-bearing and unknown? What next question would change the action?
If the conversation cannot be compressed this way, the inquiry may have expanded beyond the decision it was meant to support.
The Final Transfer Test
Take the question-discovery method into a new domain where you have less familiarity. Do not begin by asking SI for the answer. Begin by describing the situation, identifying the friction and asking which concepts, problem types and evidence sources you are missing.
Then compare the resulting question with the one you would have asked initially. If the second question is narrower, more testable and more connected to action, the method has transferred.
The long-term goal is not to depend on SI to invent every question. It is to internalise the sequence that turns confusion into an answerable, useful inquiry.
The Question Relevance Test
A question can be answerable and still be irrelevant. Before spending time on it, ask how the answer connects to the current outcome.
Suppose you are choosing a course. “How old is the institution?” is answerable. But unless institutional age changes credibility, support or your eligibility, the answer may not affect the choice. “How is project feedback delivered?” may be more relevant if practical skill is the goal.
Suppose a student is struggling with algebra. “What is the historical origin of algebra?” may be interesting but irrelevant to the repair. “Can the student expand brackets accurately without a model?” is directly relevant.
Use SI to ask: “If I knew the answer to this question, what decision, diagnosis or action would change?” If the answer is “nothing”, move the question to curiosity rather than the active queue.
The Question Ownership Rule
Some questions are yours to answer. Some belong to another person. Some belong to an authoritative source. Some require qualified expertise.
“What do I value more?” belongs to you. “Are you willing to take this responsibility?” belongs to the other person. “What is the current official requirement?” belongs to the issuing source. “What does this medical result mean for me?” may require a healthcare professional.
SI can help route the question to the correct owner. This prevents a general assistant from answering questions that should instead be asked of reality, another person or an appropriate expert.
The Final Question-Discovery Rule
When you do not know what to ask, do not force certainty. Describe the state, identify the friction, find the unknown that matters most, and ask the smallest question capable of changing the next action.
Then let the answer improve the next question.
Question discovery is a loop, not a one-time prompt: situation → question → evidence → updated model → better question → action.
The Question Confidence Check
Before acting on an answer, ask how confident you are that you asked the right question. Confidence should rise when the question matches the observed failure, the evidence source is appropriate and alternative framings lead to similar next actions.
Confidence should remain lower when the question depends on an unverified assumption, another person’s motive, an unclear unit of analysis or a future prediction with little evidence.
If changing the wording of the question changes the recommended action dramatically, the problem may still be under-framed. Compare the competing questions before committing.
A good question is not merely specific. It is stable enough that nearby reasonable framings do not completely change the action without new evidence.
The Question Traceability Rule
For important decisions, record which evidence answered which question. This makes it possible to revisit the reasoning later without reconstructing the entire conversation.
A simple note is enough: question → source or test → answer → decision affected. If new evidence changes the answer, you can see which part of the decision model needs updating.
Traceability also prevents one answer from being reused outside the question it originally addressed. A fact that answers “What is the current deadline?” does not automatically answer “Is this project feasible?”
Good questions become more useful when their answers remain connected to the evidence and decision they were meant to inform.
The Question Version Rule
When the problem changes, version the question. “Why is this project late?” may become “Which approval now determines whether the revised launch date is feasible?” after new evidence arrives. Do not keep answering the old question merely because the conversation began there.
A current question should reflect the current state, not the historical starting point. This makes long SI conversations easier to steer and prevents obsolete assumptions from surviving through repetition.
The best next question belongs to the problem as it exists now.
Frequently Asked Questions
What should I tell SI if I do not know what to ask?
Describe the situation, what feels difficult, what you want to change and what you already know. Ask SI to identify problem types and high-value questions before solving.
Can SI discover the right question for me?
It can propose useful questions, but “right” depends on your goal, evidence and decision. Treat question suggestions as candidates to evaluate.
What is an information-gain question?
A question whose answer would substantially reduce uncertainty, eliminate options, change the diagnosis or alter the next action.
How many questions should I ask at once?
Usually one or a small sequence. A long list can create research overload. Prioritise the questions that unlock later questions.
What if I do not know the vocabulary?
Describe the phenomenon in ordinary language and ask SI for the relevant terms and distinctions. Better vocabulary often reveals better questions.
When should I stop asking questions?
When the remaining uncertainty is low relative to the consequence or a reversible experiment will create more information than further discussion.
What comes next?
The next guide shows how to separate facts, assumptions and opinions so answers do not collapse different kinds of statements into one confident narrative.
Helpful Reading
- How to Leverage Your Life with Super Intelligence
- How to Think More Clearly with Super Intelligence
- How to Use Super Intelligence to See a Problem from Multiple Perspectives
- How to Use Super Intelligence to Break Down Complicated Problems
You Do Not Need the Perfect Question to Begin
The first question can be imperfect. Start with the situation. Name the friction. State what you want to change. Let the unknowns become visible. Then ask the question whose answer creates the most useful next move.
Good questioning is not a prerequisite for using Super Intelligence. It is one of the capabilities you can build by using it well.