Why does privacy reduce conformity? The key is removing social pressure.
When people answer privately, they don't have to worry about what others think of them. This removes the main driver of conformity: the fear of standing out or being judged. A massive field experiment on WeChat Moments with 37 million users tested this directly [3]. The researchers randomly showed some users ads that included social cues—friends' likes—while others saw the same ad without those cues. They found that seeing the first friend's like significantly increased the rate at which users publicly liked the ad (a visible, social action). However, showing additional likes did not further increase the private clickthrough rate (a hidden action). This means people were conforming publicly to fit in, but their private behavior was unaffected by the social pressure. The study concludes that when normative social influence (pressure to fit in) dominates, it can actually crowd out private responses, making private answers more authentic.
Private answers reveal the true diversity of opinion, reducing partisan bias.
Partisan bias often comes from people publicly aligning with their group's position, even if they privately disagree. Private answers bypass this. In studies of AI agents simulating cross-cultural collaboration, researchers compared agents' private responses (recorded before discussion) with their public statements during group chats [1][2]. They found that agents' private opinions were more diverse and stable, while public discussions pushed them toward conformity due to perceived peer pressure. The agents' public statements often shifted to match the group, introducing bias. The private answers, however, preserved the original, diverse perspectives. This suggests that in human contexts, private responses can capture what people really think, reducing the partisan bias that emerges when people feel compelled to echo their group's public stance.
Is there a catch? Yes—encouraging debate can increase instability.
One might think that encouraging people to argue for their views would reduce conformity, but the AI studies found the opposite. When agents were instructed to debate in support of their opinions (rather than collaborate), the rate of persona inconstancy—where agents changed their stated views—actually increased [1][2]. This counterintuitive finding suggests that pushing people to defend their positions can make them more likely to abandon them under social pressure, perhaps because the adversarial setting heightens the perceived cost of disagreeing. So while private answers reduce conformity, simply telling people to 'debate' may not help—it might even make things worse. The key is to create a private space for reflection before any public discussion.
About These Sources
This answer is built on 3 studies (1 peer-reviewed, 2 preprints) — published from 2021 to 2024, 2 from 2024 or later — selected as the most relevant from 3 studies that passed quality screening, drawn from 46 papers retrieved from a database of over 500 million.
Sources used in this answer
Conformity, Confabulation, and Impersonation: Persona Inconstancy in Multi-Agent LLM Collaboration
In simulations of multi-agent LLM collaboration, agents' private responses showed more diverse and stable opinions than their public statements, but instructions to debate increased the rate of persona inconstancy (conformity and confabulation) [1].
Persona Inconstancy in Multi-Agent LLM Collaboration: Conformity, Confabulation, and Impersonation
A similar study on multi-agent LLM systems found that group discussions led to collective decisions reflecting diverse perspectives, but this benefit was tempered by agents' conformity due to perceived peer pressure, and debate instructions increased instability [2].
The Effectiveness of Social Advertising: Evidence from a Large-Scale Field Experiment
In a large-scale field experiment with 37 million users on WeChat Moments, displaying social cues (friends' likes) increased public liking rates but did not further increase private clickthrough rates, showing that normative social influence drives public conformity but not private behavior [3].