Why your AI always recommends the same thing on strategy questions

You are facing a decision about direction and ask an AI for a second opinion. Differentiate or compete on price. Centralise or decentralise. Build or buy. The answer comes instantly, sounds considered and tailored to your situation. That is precisely where the trap lies.

The same advice every time

A study published in Harvard Business Review tested several leading models across thousands of strategic decisions. The result was strikingly monotonous. The models almost always recommended the same thing. Differentiation over price. Collaboration over going it alone. Long term over short term. Augment rather than replace.

The researchers tried to break the pattern. They changed the industry, the context and the wording, and gave the models more background. The reflex largely persisted. Regardless of the situation, the models produced the obvious, well-behaved standard answer. The researchers called it trendslop.

Presentation instead of thinking

The reason is simple. At its core, a language model tells you what appears most frequently on the internet, only more eloquently phrased. It is not a thinking product, it is a presentation product. It makes the average sound like an insight.

That is more dangerous than it sounds. A weak proposal that sounds weak does no harm. A weak proposal that is brilliantly worded ends up in the minutes unchallenged. AI does not turn everyone into a strategist. It pulls the sharpest minds in the room towards mediocrity, because everyone receives the same polished template.

How to use it well

This does not mean keeping your hands off it. It means holding the tool correctly. Four points.

First, ask for options, not for the decision. Not “What should I do”, but “What three fundamentally different paths exist, and what argues against each”. That way you use the breadth and keep the judgement with yourself.

Second, feed it your own knowledge. The more genuine context from your company the AI receives, the less it falls back on the internet average.

Third, make it argue against itself. “Where does this plan break first” yields more than any confirmation. A model that only agrees is worthless.

Fourth, remain the arbiter. The AI delivers drafts, you make the decision. Conviction, taste and judgement come from you, not from the model.

The difference

That is exactly where I come in. The value does not lie in a faster standard answer, but in bringing in your own knowledge and your process logic, so that the decision remains yours and stays traceable, instead of adopting the internet's template.

A simple everyday test. When the AI gives you confident advice, ask yourself whether it would give a competitor in your position exactly the same advice. If so, it is not advice. It is the average.

This is the second of three articles. The first was about AI doing tasks, but not jobs. The third is about what digital sovereignty concretely means for an SME.