Study finds AI nudges people toward safer, more conventional choices

A Columbia Business School study found that AI agents made people’s choices less distinctive and less varied, even when the tools were tailored to individual preferences. The research, titled The Basic B*** Effect: The Use of LLM-based Agents Reduces the Distinctiveness and Diversity of People’s Choices, looked at 110,000 real-world decisions made by 1,000 U.S. social media users.

Sandra Matz, the Lulu Chow Wang Professor of Business at Columbia Business School, said the findings showed a trade-off between distinctiveness and psychological diversity. In the study’s comparison, researchers measured a human baseline against both a generic AI agent and a personalized AI agent. The school said generic AI made people’s choices more similar to one another, while personalized AI made individual users more predictable over time.

The result points to a subtle kind of narrowing that reaches beyond one app or one task. Large language model agents are already taking on everyday jobs, including writing emails, buying groceries and booking restaurants, and Matz’s research suggests those systems can steer people toward safer, more conventional outcomes inside those routines. The study found not just that AI can compress how unique one person’s choices look compared with others, but also that it can reduce how much a single person varies from one decision to the next.

Matz has long tied that risk to broader questions about psychological targeting, echo chambers, manipulation, privacy, creativity and cultural output. In a September 26, 2025 op-ed in Salon, Columbia Business School said she argued that LLM-based tools may reduce individuality and creativity even as they improve efficiency. Her 2025 book, Mindmasters, makes a similar case from a wider angle, warning that algorithms can shape purchases and political support while also serving useful ends, including better mental health, stronger financial decisions and help escaping echo chambers.

That balance is what gives the new findings their force. Personalized systems promise convenience, but the Columbia study suggests they can also flatten experimentation, making people easier to predict at precisely the moment the technology is marketed as a way to reflect individuality. In shopping, entertainment, study and work, the everyday danger is not that AI makes people reckless. It is that it makes safe choices feel default, and that can leave less room for curiosity, disagreement and independent judgment.