Why AI strategy becomes generic
Ask an LLM for a brand positioning and you will often get something perfectly sensible. That is the problem. It takes the information you supplied, reframes it fluently and produces a line that sounds like a positioning. The words may be polished. The strategic journey that should have created them never happened.
Positioning is a useful example
A strategist rarely goes straight from a pile of client information to the final positioning line. They work through the brand, the customer and the competitive context. They decide what matters. They develop a fuller value proposition. They make choices. Then they compress the result into language people can actually remember.
Sometimes they refine it again so the client can buy it without the articulation becoming empty theatre. Each stage reduces possibilities and creates constraints for the next one. Ask an LLM to jump directly to the final sentence and those intermediate choices happen invisibly, if they happen at all. The output tends to contain a little bit of everything. That creates mush.
Doshi and Hauser’s 2024 experiment found that AI-generated ideas could improve individual short stories while reducing diversity across the set. This supports testing whether assisted work converges; it does not prove that brand positionings generated by every current model are generic.
More data does not automatically fix the problem
You can paste in more research. You can upload the brand book, competitor decks and customer interviews. The model now has more material to average into the answer. That may improve factual grounding. It does not create a strategic process by itself. The important change is to structure the reasoning.
What should be understood first? Which questions need an answer before the next decision makes sense? What is the strategic philosophy guiding the work? Which trade-offs define a good result? What does bad look like?
The philosophy comes first
Two agencies can inspect the same brand, customer and competition and still produce different strategies. One agency may believe everything should begin with a deep understanding of the consumer. Another may put much more weight on breaking category conventions. They can use the same frameworks and reach different conclusions because they value different things.
That is strategic philosophy. A system that does not know what good looks like for the people using it will tend toward plausible generality.
Make that philosophy explicit when deciding which strategic decisions AI should support.
Good structure makes LLMs much better
Mirel is built on general-purpose models. The reason it can produce useful strategic work is not that we discovered a model nobody else can access. The models work inside structure, process and frameworks with instructions about what good and bad work look like. They are not asked to perform the whole strategic journey in one leap.
That is an important distinction. The problem with generic AI strategy is not that LLMs are incapable of strategic work. The problem is that the blank box encourages us to ask for the answer before we have designed the thinking that should produce it.