Pre-print: The Promise of AI-Generated Video Designs for Survey Experiments in International Relations: An Experimental Test

In a newly released preprint titled "The Promise of AI-Generated Video Designs for Survey Experiments in International Relations: An Experimental Test", we tested whether AI-generated video vignettes offer genuine methodological advantages over traditional text-based designs in IR survey experiments, using a hypothetical scenario of Russian aggression against Latvia.

Study design
The study employed a preregistered 2×4 factorial online survey experiment with a quota-representative sample of US adults (N = 2666). Participants were randomly assigned to one of two crisis scenarios (a conventional invasion or a "hybrid" operation) and one of four communication modalities: an AI-generated video styled as a breaking-news broadcast, long text accompanied by AI-generated images, plain long text, or a short text-based vignette of the kind typically used in IR research. Outcomes were measured across four preregistered dimensions: affective response, support for military engagement, attentiveness to the treatment, and survey satisfaction.

Key findings

AI-generated videos did not outperform text-based formats

Contrary to expectations, the AI-generated video vignette failed to produce stronger emotional responses or greater support for military action than the text-only or text-with-images conditions. Interestingly, the text-with-images format elicited marginally higher negative affect and slightly greater support for missile strikes than the video did.

AI-generated video's one advantage was survey satisfaction

Participants rated the video as more satisfying to engage with than plain text, but this advantage disappeared when video was compared against text accompanied by images. The two AI-enhanced formats performed comparably on satisfaction.

Long-form formats beat the traditional short vignette, except for attention

The AI-generated video, text-with-images, and plain long text all produced significantly stronger affective responses, greater policy support, and higher satisfaction than the conventional short-text vignette. The short vignette's one advantage was attentiveness: participants were more likely to correctly pass the manipulation check under the brief format, suggesting that longer, more immersive content may come at a cost to message clarity.

Practical recommendation favors text-with-images over full video production

Given that text paired with AI-generated images matched or exceeded video's performance across nearly every measure at a fraction of the production cost and effort, the authors' tentative recommendation is that this combination represents the more cost-effective path to greater ecological validity in IR survey experiments, rather than full AI-video production.