In March 2026, researchers from Barcelona, Sydney and New York confronted seven leading AI models 15,000 times with the same strategic decisions. ChatGPT, Claude, Gemini, Grok and three others. Over and over, each time freshly phrased. The result was unambiguous — and uncomfortable.
The models almost always picked the same side.
01 — What is Trendslop?
Trendslop — the tendency of large language models, when faced with strategic decisions, to reflexively favour the answer that management books, HBR articles and TED talks treat as modern: differentiate, collaborate, think long-term, augment human and machine rather than replace.
The term comes from a Harvard Business Review study by Angelo Romasanta (Esade), Llewellyn Thomas (University of Sydney) and Natalia Levina (NYU Stern). They did not measure whether AI is clever. They measured whether AI has any opinion of its own at all.
The answer: mostly not.
02 — The study
The researchers framed classic strategic trade-offs — decisions where there is no “both are right”:
- Short-term growth vs. long-term stability
- Automation vs. augmentation of the workforce
- Cost leadership vs. differentiation
- Centralisation vs. decentralised structures
15,000 simulations. Seven models. Different industries, different company sizes, different phrasings. And the result held steady: the models consistently clustered on the same side of the debate — the side that management literature treats as progressive, modern and future-proof.
Even when the researchers reworded the questions or asked for the pros and cons of both sides, the direction of preference remained visible. The AI did sometimes praise the other option — but as its clear favourite it always picked the same one anyway.
03 — Why does this happen?
That is the genuinely interesting question, and the answer is structural, not accidental.
LLMs are trained on whatever dominates the internet. And what dominates the internet on strategy and management is exactly the literature that produces Trendslop: HBR articles celebrating differentiation. LinkedIn posts extolling collaboration. TED talks selling augmentation as a human duty. Business books declaring long-term thinking a virtue.
The model never learned what works better in context A than in context B. It learned what counts as good within this corpus — and this corpus is not representative of reality. It is representative of how consulting firms, keynote speakers and business publishers present themselves.
Use AI for strategy and you get distilled mainstream thinking — with an authority boost.
04 — What this means in practice
The researchers offer concrete recommendations. I consider three of them essential:
- AI opens up options, it does not make decisions. Use models to widen the space of possible strategies — not to outsource the decision. The point at which a human has to choose comes sooner than you think.
- Multiple models, used deliberately. If all seven models give the same answer, that is not consensus — it is an echo. Cross-model validation only helps if you know what you are looking for.
- Local before global. AI does not know your company’s context from its own experience. Industry specifics, company culture, market position — that is what the human has to bring in. The model fills the gaps with the average.
05 — The blind spot
What stays with me about this study is not that AI is sometimes wrong. That was already known.
What stays with me is the direction of the errors. Trendslop is not a random error — it is systematically skewed in a particular cultural direction. The models do not favour just any side. They favour the side that Western, anglophone, academically shaped management texts treat as progressive.
Anyone who does not know this and treats AI as a neutral strategy advisor does not get analysis. They get a distilled version of what McKinsey, TED and Harvard have held to be right over the past twenty years — fed back as supposedly objective advice.
This is not an argument against AI in strategy. It is an argument for knowing what it is you are actually consulting.
The original study: Researchers Asked LLMs for Strategic Advice. They Got “Trendslop” in Return. — Harvard Business Review, March 2026. Authors: Angelo Romasanta, Llewellyn D.W. Thomas, Natalia Levina.
Leave a Reply