Working Paper: CEPR ID: DP18554
Authors: Daniel Ershov; Yanting He; Stephan Seiler
Abstract: We quantify the prevalence of undisclosed influencer posts on Twitter across a large set of brands based on a unique data set of over 100 million posts. We develop a novel method to detect undisclosed influencer posts and find that 96% of influencer posts are not disclosed as such. Despite stronger enforcement of disclosure regulations, the share of undisclosed posts decreases only slightly over time. Compared to disclosed posts, undisclosed posts tend to be associated with younger brands with a large Twitter following and are posted from smaller accounts that generate higher engagement per follower.
Keywords: Social Media; Influencer Marketing; Advertising; Disclosure; Consumer Protection
JEL Codes: C55; M31; M37; M38
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| true organic posts (Y60) | machine learning algorithm (C45) |
| disclosed sponsored posts (M37) | machine learning algorithm (C45) |
| classification error (C38) | prevalence of undisclosed posts (Z13) |
| stronger enforcement of disclosure regulations (G38) | share of undisclosed posts (Y90) |
| younger brands with large Twitter following (D26) | share of undisclosed posts (Y90) |
| smaller accounts generating higher engagement per follower (L25) | share of undisclosed posts (Y90) |