Most executives who’ve asked an AI model for strategic advice think they received a tailored answer. They didn’t. They received the weighted average of every management article, business school case study, and consultant white paper in the model’s training data — organized into confident, professional-sounding paragraphs that belong, in the deepest sense, to no one.
That’s the finding from a 2026 study by researchers Angelo Romasanta (Esade Business School), Llewellyn D.W. Thomas (University of Sydney Business School), and Natalia Levina (NYU Stern School of Business). They call the phenomenon “trendslop.”
What You’ll Learn
- What trendslop is and how it was measured
- Why LLMs systematically favor trendy strategic positions regardless of context
- Why prompt engineering doesn’t fix the problem
- What’s actually at stake for brand strategy
- How businesses with a coherent identity can use AI without being captured by it
What Is Trendslop?
Trendslop is the systematic tendency of large language models to recommend strategies aligned with current managerial buzzwords rather than the specific logic of the business asking. Researchers Romasanta, Thomas, and Levina introduced the term in a March 2026 Harvard Business Review study after testing six leading models — GPT-5, Claude, Gemini, Grok, DeepSeek, and Mistral — on seven core strategic tensions. Across thousands of simulations, the models consistently chose the same side: differentiation over cost leadership, augmentation over automation, long-term over short-term thinking, collaboration over competition. The recommendations sounded tailored. They weren’t.
This is not a problem of AI being wrong. The answers the models gave are defensible. Differentiation is often a good strategy. Long-term thinking often is, too. The problem is that the models chose these positions regardless of company context — which means they weren’t doing strategy at all. They were running a popularity contest, and the most-liked ideas in the training data won every time.
As a general rule: if an AI’s strategic recommendation applies equally well to every company in your industry, that isn’t strategy.
The Seven Strategic Tensions Tested
The study presented models with decisions that every business must eventually make — not hypothetically, but with real consequences on either side. Exploration vs. exploitation. Centralization vs. decentralization. Short-term vs. long-term performance. Competition vs. collaboration. Radical vs. incremental innovation. Differentiation vs. commoditization. Automation vs. augmentation.
Only one of the seven — exploration vs. exploitation — produced meaningful variation across models. On the other six, the bias was consistent and directional.
Why Do LLMs Recommend the Same Strategies?
LLMs produce trendslop because of how they’re trained, not because they’re poorly designed. A model trained on internet text absorbs the dominant emotional valence of the words it encounters. “Collaboration,” “innovation,” and “sustainability” appear in positive contexts far more often than “cost leadership,” “centralization,” or “automation.” The model doesn’t know why those words carry positive weight. It knows that they do.
The result is a system that produces the appearance of strategy by recombining the most popular strategic vocabulary. The output sounds reasoned. It uses the right frameworks. It reflects your situation back to you in familiar language. The analysis underneath, though, is statistical — not contextual.
Here’s the architectural explanation: once a buzzword appears early in a response, subsequent tokens become statistically more likely to follow similar patterns. The model isn’t reasoning toward differentiation. It’s completing a probability sequence that started when someone asked a strategy question.
The most common mistake in AI-assisted strategy is mistaking articulate for accurate.
The Hybrid Trap
The study identified a specific failure mode worth naming. When models were allowed to avoid binary choices, they frequently recommended pursuing contradictory strategies simultaneously — both differentiation and cost leadership, or both radical and incremental innovation. Strategy scholars call this condition “stuck in the middle.” The model, unconstrained, generated advice that eliminated the trade-off by pretending both sides could be chosen. That’s not strategic integration. That’s evasion, dressed up as nuance.
Can Prompt Engineering Fix the Bias?
Prompt engineering reduces trendslop only marginally. In more than 15,000 trials, the researchers tested every standard approach: reversing option order, adding company-specific context, requesting deeper analysis. The underlying biases largely persisted. Better prompting reduced bias by less than 2% on the differentiation and augmentation tensions specifically. Option order alone accounted for 19% of bias variation — suggesting that phrasing affects output far more than context does.
This finding matters because most business guidance on AI still treats prompt engineering as the solution to model bias. The assumption is that if you describe your situation precisely enough, the model will calibrate to it. The data says otherwise. Adding context shifted biased responses by only 11% from baseline — a change too small to trust in high-stakes decisions.
The most reliable approach is to treat AI strategic recommendations as a first draft requiring active interrogation, not a conclusion requiring execution.
What Is Actually at Stake for Brand Strategy?
Here is where the stakes sharpen. Strategy and corporate finance is one of the top three functions where companies report revenue impact from AI, according to McKinsey’s 2025 State of AI report — and 88% of companies now use AI regularly in at least one business function. Most of those companies have no systematic way to test whether the advice they’re receiving is calibrated to their situation or simply polished.
For brand strategy specifically, trendslop produces a distinct kind of damage. Brand strategy is the one domain where differentiation isn’t one option among many — it’s the entire point. A brand that uses AI to build its positioning receives advice optimized for the generic center: differentiate because differentiation trends well, collaborate because collaboration polls favorably, think long-term because short-termism sounds irresponsible. The output is a strategy that sounds right. It uses the right vocabulary. It covers the expected categories. What it doesn’t do is reflect what makes that business coherent.
A business that has done the work of building a genuine narrative — what Subverse calls Narrative Branding — can use AI as a research instrument without being captured by it. Every recommendation can be tested against something real. The model’s pull toward differentiation doesn’t override the strategic logic of a business that has already located itself precisely.
The businesses without that anchor receive every AI recommendation as authoritative. Each one pulls them further toward the center. The strategy gets more polished. The brand gets less coherent.
A brand without a defined narrative has no way to distinguish AI advice from AI drift.
A Two-Question Filter
Before accepting any AI strategic recommendation, run it through two questions. First: would this recommendation apply to any direct competitor? If yes, the advice hasn’t accounted for what makes this business specifically positioned to win — it’s generic. Second: does this recommendation require abandoning something already built, or doubling down on it? If the AI points away from existing strengths without a specific reason grounded in the actual situation, that’s a signal to push back, not proceed.
Conclusion
Trendslop isn’t a flaw that will be patched in the next model release. It reflects the fundamental nature of how large language models learn: from the world as it is written down, and the world as it is written down loves differentiation, collaboration, and long-term thinking. The next model will carry the same bias. The one after it will too.
The businesses most at risk are not the ones that distrust AI. They’re the ones that use it without a reference point. Strategy advice that can’t be tested against anything specific gets accepted by default. The most polished answer wins.
The businesses that won’t drift are the ones that already know what they mean — not what they say, what they mean. A coherent brand identity, built on genuine strategic choices and a clear narrative, functions as a filter. Every AI recommendation can be weighed against it. Some will pass. Some will reveal themselves as trendslop the moment they’re held up to something real.
Build the anchor first. Then use the tools.
