Can influencer be influenced on social media: evidence from a field experiment

How to cite

Liu, Zhiyu; Zhang, Tianyu; Shu, Susan; Cheng, Zihao. 2025. "Can influencer be influenced on social media: evidence from a field experiment." AEA RCT Registry. AEARCTR-0017495. https://doi.org/10.1257/rct.17495-1.2

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Zhiyu Liu* , Tianyu Zhang , Susan Shu , Zihao Cheng

* Primary investigator

Registry ID
AEARCTR-0017495
DOI
https://doi.org/10.1257/rct.17495-1.2
First registered
Last updated
Status
On going
Trial dates
2025-06-01 – 2026-03-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/17495

Abstract

In this study, we examine how dissenting opinions to that of key opinions leaders (KOL) influence the information environment on an investor-focused online platform. Prior research has highlighted the role of social media in shaping information environments. However, the communication online can easily to lead to an echoing chamber, where the opinions of KOL are exaggerated by his/her followers in the absence of critical or scientific check. We will investigate whether the dissenting comments from the audience can influence the behavior of influencers, the KOLs. To test this, we design a field experiment on an investor social platform, where standardized dissenting and supporting comments will be posted as replies to opinions of a sample of randomly selected KOLs. To avoid subjective bias by human, we will rely on AI to generate the comments with different position right after the KOL publish the post. We will perform several sets of analyses: 1) How the KOL responds to our comments; 2) How the posts of KOLs will change; 3) How the interaction between among the audience will change; 4) Whether the information quality of KOLs’ posts improve. We expect to find that dissent acts as a subtle but effective monitoring mechanism for information providers, improving effectiveness of communication and enhancing information quality in the social media.

Experimental design

In the experimental design, a sample of influential KOL accounts is randomly assigned to one of four groups, each of which receives instant, AI-generated comments following their new post. The first group receives dissenting comments that logically refute the viewpoint expressed in each new post. The second group receives supporting comments that reinforce the original opinion. The third group receives neutral comments, which are unrelated to the stance of the KOL but maintain a non-confrontational and context-relevant tone. And the fourth group serves as the control, receiving no comments, providing a baseline for comparison.

Design details

Randomization method
randomization done in office by a computer
Randomization unit
key opinion leader accounts
Clusters
The initial sample includes 1,292 opinion leader accounts, though the final sample may be smaller due to account deactivation during the experimental period. In addition, the study includes an auxiliary sample of 435 low-attention accounts that are exposed to the intervention as part of an additional analysis beginning on December 4, 2025.
Observations
The study is expected to include approximately 100,000 to 200,000 posts, though the exact number of observations cannot be predetermined.
Arms
Main sample: 322 accounts control, 324 accounts dissenting comment group, 324 accounts supporting comment group, 322 accounts neutral comment group Low-attention sample: 108 accounts control, 108 accounts dissenting comment group, 111 accounts supporting comment group, 108 accounts neutral comment group (the final sample may be smaller)
Primary outcomes
1) How the KOLs respond to our comments; 2) How the posting behavior of KOLs will change; 3) How the interaction among the audience will change; 4) Whether the information quality of KOLs’ posts improve

Cited by

No citation data available for this registration yet.

Keywords