The (Minimal) Persuasive Advantage of Political Video over Text

Chloe Wittenberg; Ben M. Tappin; Adam Berinsky; David Rand

doi:10.1073/pnas.2114388118

In brief

Political video is widely assumed to be more persuasive than text, but large-scale experiments show this advantage is minimal. While video makes information seem more believable (0.26–0.28 SD effect), its impact on attitudes and behavior is small (0.08–0.11 SD), comparable to or smaller than text-based persuasion. The disconnect between believability and persuasiveness challenges common narratives about video's unique power.

Contributions

Directly compares the persuasive efficacy of political video versus textual modalities in two large preregistered experiments (total n ≈ 7,600).
Proposes and operationalizes a two-dimensional framework: belief (perceived authenticity of depicted events) versus persuasion (attitude change), showing these dimensions respond differently to media modality.
Demonstrates video's advantage is primarily on the belief dimension, with minimal effects on attitudes, sharing intentions, or engagement.
Shows effects generalize across diverse messages (48 political ads and 24 COVID-19 clips) and demographic subgroups.

Method

Experimental design: Two high-powered randomized controlled trials with U.S. samples, employing a 3-condition within-subjects design. For each message, participants were randomly assigned to: (i) watch a short video clip, (ii) read a detailed transcript of the video with visual descriptions, or (iii) receive no new information (control).

Study 1 (Political ads): n = 4,266 respondents (7,982 consented; 4,629 passed attention checks). Presented 48 messages from the Peoria Project database of professionally produced political advertisements covering diverse hot-button issues (climate, healthcare, minimum wage).

Study 2 (COVID-19 pandemic): n = 3,343 respondents (5,713 consented; 3,698 passed attention checks). Presented 24 widely-viewed YouTube clips about COVID-19, spanning both overtly political (e.g., Trump clip on testing) and non-political (e.g., handwashing demonstrations) content.

Measures: All respondents rated (on 5-point scales) the believability of the message's core claim (did the depicted event actually occur?), their attitudes toward the message's persuasive claim, sharing intentions (likelihood of sharing the message), and engagement (interest in learning more, perceived importance of the topic). Study 1 additionally measured importance judgments.

Analysis: Bayesian multilevel linear regression models with random intercepts and slopes for messages and respondents, allowing treatment effects to vary across messages and participants.

Results

Believability dimension: Video significantly increased belief that depicted events occurred relative to text:

Study 1: 0.26 SD (95% HPDI = [0.23, 0.30])
Study 2: 0.28 SD (95% HPDI = [0.23, 0.33])

Persuasion dimension (attitudes): Video showed only minimal persuasive advantage:

Study 1: 0.08 SD (95% HPDI = [0.05, 0.11])
Study 2: 0.11 SD (95% HPDI = [0.07, 0.16])

The 95% credible intervals exclude effects larger than 0.16 SD, indicating high confidence in small effects.

Sharing intentions: Similarly small or negligible differences (0.08 SD in both studies).

Engagement: Minimal differences in interest in learning more or perceived message importance (0.01–0.04 SD).

Stability: Results replicate across individual messages (strongest attitude effect across all 72 clips remains weaker than weakest belief effect) and demographic subgroups (age, political affiliation, partisanship, digital literacy, cognitive reflection), with no significant heterogeneity by group.

Control comparison: Both video and text conditions showed measurable effects on attitudes relative to control (Study 1: 0.07 SD; Study 2: 0.14 SD), suggesting both modalities can persuade, but the video-text difference is comparable to or smaller than the text-control difference.

Notes

The paper makes a strong methodological contribution by separating belief from persuasion, two constructs often conflated in media effects research. The finding that enhanced credibility does not guarantee persuasion is important for understanding both misinformation threats and intervention design.

One limitation noted by the authors: the textual treatment was an annotated transcript with visual descriptions, potentially more detailed than typical written persuasive content (e.g., opinion pieces, news articles). Real-world video versus text comparisons might show even smaller differences in favor of video.

The results have direct implications for deepfake threat models — even if deepfakes enhance belief in false events (the mechanism they likely exploit), this may not automatically translate to large-scale attitude or behavior change.