Your best prompt is still a better question

Every few months the internet discovers a new “perfect prompt.” Magic phrases. Role-play costumes. Elaborate scaffolding that promises better answers if you speak to the machine in the right dialect. Some of it is mildly useful. Most of it is theatre for people who would rather collect techniques than clarify what they are trying to learn.

The best prompt is still a good question. Curiosity will outlast prompt engineering. Models change. Interfaces change. The skill of noticing what you do not know-and asking with precision-does not.

This piece is about treating questioning as craft, building a research mindset that tools can amplify, and why founders who chase prompt tricks often skip the harder work of framing problems. Related: your competitive advantage was never typing faster.

A good question is also a management tool. It aligns a room faster than a slide. It exposes disagreement early. It makes AI useful because the tool finally has a job that is not “sound smart.” If your meetings produce prompts instead of questions, you have already lost the plot.

Prompt tricks age; questions compound

A trick is brittle. It depends on a model version, a UI, a meme. A good question is portable. You can ask it of a customer, a colleague, a dataset, or a model. When the tool improves, the question gets a better answer. When the tool disappoints, the question still teaches you what to investigate next.

I am not anti-technique. Structure helps. Constraints help. Examples help. What I resist is the idea that the bottleneck is vocabulary for talking to machines rather than clarity about the problem. If your question is mush, no incantation will make the answer sharp.

• Tricks optimise for today’s interface; questions optimise for understanding.
• A precise question reveals assumptions you did not know you held.
• Curiosity produces follow-ups; tricks produce one-shot outputs.
• Teams that fetishise prompts often underinvest in interview skill and problem framing.

What a good question does

A good question narrows the search space without strangling discovery. It names the decision it serves. It admits what would falsify a preferred answer. It is specific enough that a vague reply looks obviously useless. In strategy work, questions are how you avoid decorating the wrong problem-see what founders actually need isn’t more marketing.

If you need the model to pretend to be a genius before it can help you, the problem is probably your question-not its résumé.

A framework: wonder → frame → probe → prove

1. Wonder: protect unstructured curiosity

Not every inquiry starts neat. Wonder is the messy noticing: “Why do demos stall after pricing?” “Why do people compliment the brand and not convert?” Capture wonders without forcing them into a prompt yet. Perspective begins with noticing-see AI can write your content; it can’t replace your perspective.

2. Frame: turn wonder into a decision-shaped question

Framing adds stakes. Who needs the answer? By when? What action will change based on it? What would count as good enough evidence? This is product strategy work more than content work-close to a product strategy framework before designing a screen.

3. Probe: ask in layers, not in one theatrical prompt

Good research is iterative. Ask for categories of answers. Challenge the first reply. Request counter-arguments. Compare against customer quotes. Use the model as a sparring partner, then verify against reality. Probing is a habit; a single mega-prompt is a wish.

1. State the decision the answer must serve.
2. Ask for 3-5 competing explanations, not one tidy narrative.
3. Demand what evidence would support or kill each explanation.
4. Take the strongest explanations to customers, data, or experiments.
5. Update the question based on what you learned-then repeat.

4. Prove: close the loop with reality

An unanswered beautiful question is still incomplete. Prove means interviews, analytics, prototypes, shipping a stake and watching what happens. AI can accelerate synthesis. It cannot replace contact. Organic trust still has to be earned in the world: organic growth isn’t free-it’s earned.

A fictional contrast: two discovery processes

Team A spends an afternoon collecting viral prompts for “go-to-market.” They generate a plan that sounds like every Series A blog post. Team B spends the afternoon framing one question: “Why do activated users fail to invite a teammate?” They probe support logs, ask the model for competing hypotheses, then interview ten users. The plan that emerges is narrower and true. Team A has a document. Team B has a direction.

Questioning as craft in an AI workplace

Craft means standards. Teach teams to write questions that name the decision, the audience, the constraint, and the falsification test. Review prompts the way you review briefs. Reward people who kill a bad question early. Do not reward the longest prompt.

• Decision: what will we do differently if we learn X?
• Scope: what is in and out of the inquiry?
• Evidence: what sources count?
• Bias check: what answer are we hoping for-and how might we be wrong?
• Next step: what is the smallest test after the first reply?

Curiosity beats cleverness when the market is noisy

Noisy markets punish cleverness that is not attached to learning. Clever prompts can produce clever copy that still answers the wrong brief. Curiosity keeps you near the customer’s actual confusion. It is slower at the keyboard and faster at finding the sentence that unlocks a sale.

I would rather a founder ask one awkward, specific question in a customer call than generate fifty generic “thought leadership” outlines. The awkward question creates perspective. The outlines create calendar filler. Consistency of inquiry beats brilliance of phrasing-see again the internet rewards consistency more than brilliance.

Curiosity also protects you from false certainty. Models speak confidently. Confidence is not evidence. A research mindset treats the first fluent answer as a hypothesis generator, not a verdict. That habit alone separates teams who use AI to think from teams who use AI to stop thinking. The second group looks efficient until the market disagrees in public.

1. Replace one weekly brainstorm with one structured customer conversation.
2. Write three questions before the call; forbid pitching until two are answered.
3. After the call, ask the model only to help cluster themes-not to invent insights.
4. Turn one theme into a public stake within seven days.
5. Track whether the stake produces replies that sound like real buyers.

What research is useful for

For the discipline of inquiry, Harvard Business Review often covers question-driven leadership without the prompt-hype cycle. Nielsen Norman Group is excellent on research questions in UX. The Right Question Institute publishes practical material on question formulation-oddly relevant to teams who think their bottleneck is ChatGPT syntax.

How this shows up in my work at nau

I start projects with questions, not tools-sometimes not even Figma: why I start every project without opening Figma. AI enters when the question is sharp enough that generation or synthesis saves time. If you need help drawing the line between repetitive synthesis and judgements you must keep, read AI should handle the repetition; you should keep the judgement.

Teach juniors to bring better questions than better prompts. The first skill transfers to customer calls, hiring, product reviews, and strategy. The second skill expires with the next model release. That is not nostalgia. That is capital allocation for attention.

When in doubt, ask what would change if you learned the answer by Friday. If nothing would change, you are collecting trivia. If something would change, you have a real question-and a reason to use the tools.

Outlast the trick cycle

Prompt engineering will keep evolving. Some of it will be real craft. Much of it will be fashion. Curiosity is older and harder to obsolete. Ask better questions. Frame them around decisions. Probe until reality answers. The best prompt was never a secret sentence. It was a mind that knew what it was trying to find out.