Most prompting advice is about volume. Write longer prompts. Add context. Be more specific. These things help. They are also not the thing. The real variable isn't length. It's form.
Questions invite answers. Answers don't require reasoning. They require selection. The most effective way to use an AI thinking partner isn't to ask it things. It's to give it a situation.
WHY QUESTIONS PRODUCE WEAK OUTPUT
When you ask "what should our positioning be?" the model is doing something specific. It's predicting the most likely answer to a positioning question given the context you've provided. The output is shaped like a positioning answer. It includes positioning answer ingredients. It hedges where positioning answers usually hedge. It commits where positioning answers usually commit.
What it's not doing is reasoning about your specific situation. It's pattern-matching to "positioning answer." The result is competent and forgettable. It sounds like the kind of thing that would come out of a brand consultancy. It sounds that way because the training data has read every brand consultancy's work.
WHAT A SITUATION DOES INSTEAD
A situation forces the model to model. You're not asking for an answer. You're describing a state of the world and asking the model to operate inside it. "Here's our positioning. Here's our top three competitors' positioning. Here's what our last three campaigns said. Here's what our churned customers tell us. Now: where's the contradiction?"
The output is different. The model has to reason about the specific configuration you've given it, not pull from a generic answer template. The pattern matching becomes harder because the pattern is less common. You get something closer to actual analysis instead of closer to the average answer.
EXAMPLES
Bad: "Write me a tagline for [brand]."
Better: "Here's our brand. Here's our customer. Here are the five taglines competitors have used in the last three years. Write me a tagline that none of them could have written."
Bad: "Help me with my content strategy."
Better: "Here's what we've published in the last six months. Here's the engagement data. Here's what our best-performing competitors are publishing. What story is the data telling that we're not seeing?"
Bad: "What should we name this product?"
Better: "Here's the product. Here's the feeling we want it to evoke. Here's what we've already considered and rejected, and why. Generate names that fit the rejection criteria but go in a direction we haven't considered."
THE PRINCIPLE
The model is good at operating inside a situation. The model is bad at deciding what the situation is. Your job, as the human, is to define the situation as completely as possible. Then ask the model to do something inside it.
The richer the situation, the better the output. Most "good prompts" are good because they describe the situation richly, not because they ask the question well. The question is almost an afterthought. The setup is the work.
WHY THIS MATTERS FOR STRATEGIC WORK
For execution work (write this, summarize that), questions are fine. The model is good enough at the most likely interpretation of a basic ask.
For strategic work, questions actively hurt you. They invite the average answer. The average answer is the enemy of strategic work. You need to push the model into a specific, non-average space. Situations do that. Questions don't.
The shift in how you prompt is small. The shift in what you get back is large. Try it on the next strategic question you're tempted to ask. Don't ask. Describe. Then point.