A blank prompt is a tiny abdication.
It asks a machine to make the first move: name the problem, choose a frame, suggest the reasons, and arrange the furniture. The answer may be graceful. It may even contain a small brass telescope. It is still beginning from nowhere you have examined.
For strategy, give the machine a bad paragraph instead.
Not a polished manifesto. A provisional position with a claim that can be argued with. It is the rough pencil line under the paint. Once it exists, the machine has something useful to do: test the line, find the weak timber, and propose a different cut.
Writing is where the choice appears
Sönke Ahrens’s How to Take Smart Notes makes a practical claim that matters beyond note-taking: writing is part of developing an idea, not the box it goes into afterward. A sentence has edges. It can be questioned. It can disagree with the evidence in the next drawer.
That is why the first paragraph matters. A team cannot test “make the brand more relevant.” It can test: “Make the brand the dependable choice for small firms that need faster reporting without putting sensitive figures into a black box.” Now there is a customer, a change, a trade-off, and several opportunities for the facts to object.
Writing something yourself does not make it true. It makes the claim available for inspection.
BUREAU SYSTEM NOTE 18: The opinion has been removed from its protective fog. Please keep hands clear of the evidence.
The first draft is not clerical work
There is a modest body of research behind this otherwise untidy practice. In a classic set of experiments, Norman Slamecka and Peter Graf found that people remembered material better when they generated it than when they simply read it. Later, Michelene Chi and colleagues found that students prompted to explain material to themselves showed greater understanding than students who explained less.
These are learning studies, not instructions for a brand department. They do not prove that every handwritten slogan is wise. A 2026 preregistered study found no learning advantage for instructed self-explanation over rereading once time on task was controlled; correct inferences still predicted performance. That is a useful correction. The point is not that typing sentences makes a person clever. It is that a claim with reasons can be inspected, contradicted, and repaired.
Generative AI changes the conditions. In a 2025 CHI survey of 319 knowledge workers, Microsoft researchers found that higher confidence in AI was associated with less reported critical thinking, while confidence in one’s own ability was associated with more. This is self-reported evidence, not a verdict on every use of AI. But it supplies a sensible warning label for the blank prompt.
The better arrangement is neither solitary purity nor automatic deference. Another 2025 study compared a recommendation-style AI with one that built on users’ own reasons in an investment task. The latter fitted people’s decision process better and led to slightly better outcomes; the recommendation system produced more novel ideas with less effort (Reicherts and colleagues). That tension is valuable. Let the machine bring novelty. Make the human bring a position.
A paragraph with three joints
Before asking for routes, headlines, futures, or a comparative table with alarming confidence, write about 120 words containing:
- The change. What should become different for whom?
- The reason. Why might this work? Name the evidence, not just the mood.
- The price. What will the organization refuse, risk, or stop doing if it takes this route?
Then ask the machine for the strongest counterargument, the missing proof, and two genuinely different ways to meet the same aim. Check its sources. Mark every assumption. Revise the paragraph yourself.
The bad paragraph is not a ceremonial obstacle before the clever work begins. It is the first small piece of strategy. It gives the machine a wall to lean its ladder against.
Continue with The Brief Is Not a Vending Machine and The Assumption Is Doing the Strategy.