The “Trendslop” Problem

In a recent article in the Harvard Business Review, researchers posed a deceptively simple question: Can large language models generate useful strategic advice?

The answer was sobering.

Instead of sharp, context-aware thinking, the models produced what the authors call “trendslop” — generic, buzzword-heavy recommendations that mirror fashionable management narratives rather than grounded strategic insight. It’s easy to dismiss this as a limitation of current AI systems. It would be a mistake to do so.

Because for anyone working in foresight, this isn’t just an AI problem.
It’s a mirror.

Foresight has always operated in a tension. On one hand, it aims to explore uncertainty, challenge assumptions, and surface discontinuities. On the other, it is constantly pulled toward what feels plausible, what is already visible, and what others are saying. That is precisely where AI excels.

Large language models are trained on the past and optimized for probability. They recombine dominant narratives, reinforce what is widely accepted, and produce outputs that feel convincing because they align with existing discourse.

AI is very good at producing futures that look like the present — just slightly upgraded.

That’s not foresight.
That’s extrapolation.

And when extrapolation becomes automated, scaled, and instant, we get trendslop.

This is not just a theoretical concern.

In “Using AI for developing foresight: Reflections on an experiment”, conducted at the Bavarian Foresight-Institute, we explored how AI changes the practice of foresight — not just its outputs.

The setup was simple: different teams worked through the same structured scenario process. Some used AI tools, others did not. What we observed was striking. AI did not just influence the results. It fundamentally reshaped how participants engaged with the process itself.

One of the most revealing findings was where attention went. Teams using AI became highly focused on interacting with the tool — refining prompts, improving outputs, iterating on responses. At the same time, they were slower to recognize AI itself as a key driver shaping the future they were supposed to explore. Teams without AI identified AI as a critical uncertainty much earlier.

This inversion is telling.

The more participants engaged with AI as a tool, the less they reflected on AI as a phenomenon.

This is the probability trap in practice. Attention shifts away from questioning assumptions and toward optimizing outputs. Exploration quietly gives way to refinement.

The experiment also revealed a deeper shift. Foresight traditionally draws on lived experience, domain expertise, and the interpretation that emerges through discussion. It is as much about meaning-making as it is about information. With AI, the source of insight changes. Participants relied more on generated content and less on their own perspectives. The outputs became more polished and coherent—but also more detached from context.

AI shifts foresight from a process of sensemaking toward a process of content generation.

And that shift matters. Because foresight is not just about producing scenarios. It is about engaging with uncertainty in a meaningful way.

A third effect connects directly to the idea of trendslop. AI-supported teams tended to converge more quickly. Shared framings emerged earlier, and alternative interpretations were explored less extensively. In contrast, teams without AI spent more time in disagreement, negotiation, and reinterpretation. This friction is often seen as inefficiency. In foresight, it is not. It is where insight emerges.

If AI reduces friction, it may also reduce the diversity of futures we are able to imagine.

At scale, this becomes a systemic issue. If many organizations rely on similar models trained on similar data, we risk producing not just trendslop—but standardized futures.

None of this makes AI irrelevant for foresight. But it clarifies its role. AI is extremely effective at mapping what is already being said. It can synthesize large bodies of knowledge, identify dominant narratives, and make the “obvious future” visible with remarkable speed. And that is precisely where its value lies.

AI helps us see what is already thinkable.

Which, in foresight, is only the starting point.

The real danger is not that AI produces trendslop. The danger is that it changes how we think. The experiment shows that AI does not simply support foresight—it reshapes attention, compresses exploration, and redirects cognitive effort toward the tool itself.

A subtle shift occurs: from thinking about the future to thinking with the machine. If left unchecked, this leads to less reflection, earlier convergence, and fewer genuinely different perspectives.

Not just weaker outputs—
but weaker thinking.

All of this points to a shift that is already underway. As AI makes analysis, synthesis, and even narrative generation increasingly accessible, the value of foresight moves elsewhere.

Toward maintaining diversity of thought. Toward creating productive friction. Toward questioning what appears obvious.

And perhaps most importantly, toward designing processes that resist the pull toward trendslop.

AI will not replace foresight. But it will expose what foresight actually is.

Not the aggregation of trends.
Not the generation of plausible narratives.
Not the automation of strategy.

But the disciplined practice of thinking beyond the obvious—
especially when the obvious can be generated instantly, at scale, by machines.

If AI gives us the most likely future,
foresight becomes the practice of refusing to stop there.