Does AI make strategy generic?

The published work says it does. Researchers at Esade, Sydney and NYU Stern reported in March 2026 that leading models recommend strategies that match current managerial trends and vocabulary rather than with the logic of the situation, and their own recommendation is to use these systems to expand options rather than to make choices.

The finding#

Angelo Romasanta, Llewellyn Thomas and Natalia Levina published the result in Harvard Business Review on 16 March 2026 under the name trendslop: leading models, asked for strategic advice, converge on what currently sounds right in management rather than on what follows from the situation described. The article page discloses no sample, no measures and no figures, so nothing here rests on a rate, and the underlying study has not been examined for this entry.

What the article does carry is a recommendation from inside the finding, which is the useful part: expand options with these systems, and keep the choosing human.

Why convergence matters more than correctness#

A strategy is a bet that differs from what others will do. An answer can be entirely reasonable and worth nothing competitively if every firm in the sector receives a version of it in the same week, because the position stops being available at the moment it becomes obvious.

That is a different failure from hallucination and it is harder to see. Wrong output gets caught by review. Generic output survives review comfortably, because everybody in the room recognises it as sensible, and its plausibility is what makes it worthless.

The wider pattern#

Convergence in strategic advice sits inside a broader finding about homogenisation of reasoning and perspective, treated at does AI make everyone think alike. The strategic case is the sharpest version, because the output is a competitive position rather than a piece of writing, and because the effect is felt by the organisation rather than by the individual who produced it.

It also compounds with altitude lock. If a leader can reach every level instantly, and what arrives at each level is the median view, then the abundance that was supposed to widen perspective is delivering the same perspective at three resolutions.

What to do about it#

• Use the model for the option set, not the choice. The authors' own recommendation, and the one the evidence supports best.
• Record what was rejected and why. The distinctive part of a strategy is whatever a person overrode, and it exists as evidence only where somebody wrote it down. This is the Head entry in Hand, Head, Hours, Heart.
• Test for genericness directly. Ask whether a competitor running the same exercise would arrive here. If the truthful answer is yes, the document is a summary of the current consensus rather than a strategy.
• Put the context in, then check whether it changed anything. If a detailed brief and a thin one produce the same recommendation, the context was not being used.

What this does not establish#

Any rate. No figures were available on the page read, so this page supports no quantitative claim about how often or how strongly models converge. It does not establish that AI-assisted strategy performs worse, which nobody has measured, nor that the convergence persists with better prompting or with newer systems. The claim that generic strategy destroys advantage is an inference from what strategy is for rather than a measured effect.

Key sources

• Romasanta, A., Thomas, L. D. W. and Levina, N. (2026). Researchers asked LLMs for strategic advice. They got trendslop in return. Harvard Business Review, 16 March 2026.

Related SuperSkills research#

• Big picture thinking in the age of AI
• What is altitude lock
• If everyone has AI, where is the advantage