Part V
Why ChatGPT is not a strategy process
General LLMs are extremely useful for strategic work. Give one a refined analytical task and decent information and you can get far more than “thanks, Captain Obvious.” Ask it to find patterns in a dataset, compare competitors, structure research or challenge a hypothesis and it can save a remarkable amount of time.
ChatGPT is a general reasoning environment. A strategy process is an operating sequence in which each decision changes the state of the next task.
There is no single task called strategy
Strategy is a combination of different things done in a useful order. A brief is interrogated. The problem is diagnosed. The audience is understood. Evidence is interpreted. A position is chosen. Campaign and creative decisions follow.
Each step changes the next one. The audience work may expose that the original problem was wrong. The positioning may rule out a message that looked promising earlier. A creative response may reveal that the strategy is correct and lifeless.
A good prompt can handle one part of this. The difficulty lies in moving through the whole process while keeping the relevant conclusions, qualifications and unresolved questions alive.
The blank box asks too much of the user
ChatGPT will usually answer the question it is given. That is helpful when the problem is clear and dangerous when it is not. Ask for a campaign strategy and the default interaction encourages an answer, even when the more useful response would be to challenge incompatible objectives or an empty audience definition first.
The user has to know which task should happen now, what information belongs in it and what a good answer looks like. They also need to spot when the model moves ahead too quickly, produces something plausible and quietly drops an important qualification.
An experienced strategist and prompt engineer can manage this. They can create separate conversations, carry outputs between them and intervene at the right moment. At that point, much of the process lives in the user. The software is providing intelligence on demand.
Chat does not impose commitment
Chat naturally encourages continuation. Ask another question, request another version, explore another direction. Strategy also needs stopping points.
A team has to decide, approve, reject or mark something unresolved. The audience definition is accepted for this assignment. One positioning route is chosen and the others are closed. A contradiction in the brief remains open and blocks the next step. These changes of state create commitments that later work can depend on.
A chat history can contain all of those moments without making their status clear. The conversation continues, while the process needs to know what became authoritative.
Plausibility is especially risky under pressure
The blank box is most dangerous when the user lacks experience in the problem or is short of time. A fluent result feels like progress. It can be structured, confident and full of strategically familiar language while remaining below the standard an experienced person would accept.
This is not a unique defect of ChatGPT. People produce weak work under the same conditions. The model makes it easier to produce a polished version quickly, which can hide the weakness long enough for it to enter the deck.
A more elaborate disclaimer will not protect the work. A process that asks uncomfortable questions early and makes evaluation part of the work has a better chance.
What a strategy process adds
A useful process gives sequence, state and criteria. It diagnoses before solving. It requires decisions at meaningful stopping points and allows work to move backward when a later response exposes a weakness. It gives users somewhere to challenge and correct the work.
ChatGPT can participate in all of this. It does not provide the operating model by itself. Calling the blank box a strategy process is a little like calling PowerPoint a consulting methodology because many consultants use it.
I will keep using general LLMs for strategic tasks. They are exceptionally flexible thinking tools. I simply would not make a shared agency process depend on every user knowing which prompt to write, what to paste next and when a good-looking answer should be ignored.