The problem with asking AI to think strategically isn't that AI can't think. It's that strategy isn't thinking the way the model thinks. Strategy is the act of choosing what to be, which means choosing what not to be. It requires the willingness to exclude. The model can't exclude. Excluding is the opposite of what it was built to do.
You ask ChatGPT for a positioning statement. It gives you something competent and forgettable. You sharpen the prompt. You add constraints. You tell it to be bold. The output gets slightly more polished. The underlying mediocrity stays the same. Because the model is producing the most likely answer, and the most likely answer is the consensus, and the consensus is exactly what your competitors will also arrive at when they ask the same model.
WHAT STRATEGY ACTUALLY IS
Strategy is the set of decisions you make about where you will and won't compete. The "won't" is the load-bearing part. A position that everyone could occupy is not a position. It's a description. A real strategic call says "we are this, not that, and we are willing to lose the customers who want that, because the customers who want this will pay more."
This is a kind of writing the model has very little of in its training data. Most strategy documents that exist online are post-hoc rationalizations of decisions made by humans in rooms. The actual decision (the moment of cutting off other paths) doesn't appear in the document. Just the polished narrative that came after. The model has read every "strategy" document and can produce more of them. It hasn't read the moment of choosing.
WHY THE OUTPUT FEELS HOLLOW
When you read AI-generated strategy work, something feels off and you can't always name it. The reason: every option is included. Every customer is welcomed. Every value is honored. The document is internally inconsistent because it refuses to commit to anything that would alienate anyone. The result is a strategy that is, structurally, not a strategy. It's a wishlist.
Real strategists know how to do the hard cut. They'll say "we are not for that customer." They'll say "we will not compete on price." They'll say "we are willing to look stupid to people who don't get it, because the people who get it are the only ones we want." The model will not say any of these things unless you force it to, and even then it will hedge.
WHAT AI CAN ACTUALLY DO IN STRATEGY
It can pressure-test. Take a position you've already committed to and tear it apart from the perspective of a sharp competitor. The model is great at this because criticism doesn't require commitment. It just requires identifying weaknesses, and that's a pattern-matching task.
It can map. Show you the landscape of competitor positions and identify the unowned territory. The model can synthesize what's already out there better than you can in a reasonable amount of time. The unowned territory becomes obvious once you can see the full map.
It can draft. Once you've made the strategic call, the model can produce competent versions of the messaging, the brief, the deck, the narrative. The hard work was in the call. The execution work is the kind of writing the model is actually good at.
WHAT IT CAN'T DO
It cannot make the call itself. The call requires accountability the model doesn't have. It requires a willingness to be wrong in public, which the model can't experience. It requires the kind of conviction that comes from having staked something personal on the answer. The model has staked nothing on anything.
This is why even the best AI-generated strategy work feels like it was written by a smart consultant rather than by the founder. The smart consultant produces the best version of the obvious answer. The founder produces the answer that nobody else would have given because nobody else has the same set of biases and history and obsessions.
HOW TO USE AI IN STRATEGY WITHOUT GIVING UP THE STRATEGY
Stop asking the model what your strategy should be. Start asking it to test the strategy you already have. The role isn't strategist. The role is sparring partner. A good sparring partner makes you better. It doesn't replace you.
Bring it your hypothesis. Ask it to find the holes. Ask it what a competitor would say. Ask it what a customer would dismiss. Ask it to generate the strongest version of the opposite argument. Then make the call yourself. The call is the part of the work the model can't help with. It's also the only part that actually matters.