Voice Analytics of Online Influencers

In the rapidly growing influencer marketing industry, positive consumer sentiment toward sponsored content is critical for the success of sponsoring brands and influencers alike. Unfortunately, consumers tend to react more negatively to sponsored videos than to non-sponsored videos. This paper studies how influencers may attempt to curb the negative effect of sponsorship on consumer sentiment by modulating their vocal characteristics in sponsored videos relative to non-sponsored videos. We use deep learning methods to extract multi-modal 3V features (voice, visual, and verbal), and we use three causal identification strategies: instrumental variables, fuzzy regression discontinuity, and two-way fixed effects. We find that influencers decrease the average loudness of their voices in sponsored videos relative to equivalent non-sponsored videos, and the difference in average loudness partly mitigates the negative effect of sponsorship on consumer sentiment. Influencers are heterogeneous in the extent to which they decrease loudness in sponsored videos; for example, the difference in average loudness between sponsored and non-sponsored videos is smaller among influencers with more followers but larger among influencers with a steeper short-term increase in the number of Instagram followers. We discuss implications for four stakeholders: influencers, brands, consumers, and the Federal Trade Commission