Toxic Language Captures Attention: Evidence from 200 Million News Headline Impressions
Type
working paper
Date Issued
2025-09-20
Author(s)
Abstract
Online platforms extract revenue by capturing user attention and selling it to advertisers, leading algorithmic content curation to prioritize content that most effectively captures attention. This creates a potential pathway for toxic language to dominate public discourse if toxic content systematically outperforms civil alternatives in capturing clicks. Using data from 53,755 news headlines across 12,473 A/B tests that generated over 205.8 million impressions at Upworthy.com, I employ the Perspective API to measure headline toxicity and analyze click-through rates as the primary outcome variable to approximate attention capture. A fixed effects specification exploits within-experiment variation to isolate toxicity effects from article-level characteristics, comparing headlines that advertise identical content but vary in toxic language. Results demonstrate that toxic content increases click-through rates by 9.859% relative to civil alternatives, even when controlling for negativity and other linguistic features. This persistent toxicity bias suggests that toxic language triggers automatic attention-capture mechanisms that operate free from social signaling dynamics.
Refereed
No
Start page
1
End page
19
Pages
19
File(s)![Thumbnail Image]()
Name
Roggenkamp_2025_Toxicity.pdf
Size
2.4 MB
Format
Adobe PDF
Checksum (MD5)
edc25e51fb84c5fa73e0f378e9e1f1e3