Large language models (LLMs) consistently generate "trendy, buzzword-laden strategic advice," a phenomenon dubbed "Trendslop," according to a recent Harvard Business Review study. This issue, which researchers identified after testing seven leading LLMs across more than 15,000 simulations, is attributed to the nature of their training data. Brian Roemmele, an independent researcher and AI theorist, has further asserted that this problem is a "massive issue that will get worse," stemming from what he terms "post-1970 Internet Sewage."
The HBR article, titled "Researchers Asked LLMs for Strategic Advice. They Got 'Trendslop' in Return," was published on March 16, 2026. Authored by Angelo Romasanta of Esade Business School, Llewellyn D.W. Thomas of the University of Sydney Business School, and Natalia Levina of NYU Stern School of Business, the study found that LLMs consistently defaulted to popular, aspirational recommendations such as "differentiation" and "augmentation," often overlooking context-specific trade-offs. The models showed a strong preference for culturally fashionable choices, even when presented with binary strategic dilemmas.
According to the HBR authors, this bias is baked into the models because their training corpora are heavily influenced by business books, LinkedIn posts, and TED Talks that celebrate low-trade-off language. Brian Roemmele echoed this sentiment in a recent tweet, stating, > "It’s The Training Data Stupid." He further elaborated on his Read Multiplex platform, arguing that LLMs trained on post-1970 internet content, characterized by "low-quality, echo-chamber-like" information from Reddit, Wikipedia, and SEO-optimized blogs, are inherently prone to producing "trendslop."
Roemmele proposes a solution: training AI models on "high-protein data" from the 1870-1970 era. He suggests that this historical corpus, comprising books, peer-reviewed journals, patents, and court records, offers more rigorous and less trend-amplified content due to the economic and legal stakes associated with publishing during that period. He claims to be actively curating and experimenting with such undigitized materials, asserting that models trained on this data demonstrate superior originality and resistance to groupthink.
The emergence of "trendslop" highlights a critical limitation in current AI applications for strategic decision-making. While LLMs can efficiently summarize information, their tendency to parrot popular but generic advice poses a risk for organizations seeking nuanced, context-specific guidance. Experts suggest treating AI output as a starting point for idea generation rather than a definitive solution, emphasizing the continued necessity of human judgment in strategic choices.
