Accuracy nudges are not effective against non-harmful deepfakes

TL;DR

The results replicated previous findings showing that prompting participants to assess the accuracy of a headline at the beginning of the survey significantly decreased their intention to share fake news, but this effect was not significant when applied to a non-harmful AI-generated video.

Abstract

I conducted a preregistered survey experiment (n=525) to assess the effectiveness of "accuracy nudges" against deepfakes (osf.io/69x17). The results, based on a sample of Colombian participants, replicated previous findings showing that prompting participants to assess the accuracy of a headline at the beginning of the survey significantly decreased their intention to share fake news. However, this effect was not significant when applied to a non-harmful AI-generated video.

Figures (2)

• Figure 1: Stimulus of the content evaluation task. Top-left: Deepfake video. Bottom-left: Real video. Top-right: Real article headlin. Bottom-right: False article headline.
• Figure 2: Likelihood to share content in different channels. Red bars represent the control group, and blue bars represent the treatment group.