Answers are grounded in this paper and cite their sources.
The 2024 Indian General Election was defined by pervasive warnings that generative AI would unleash a tsunami of deepfakes and fundamentally destabilize democratic processes. To empirically test these predictions, we analyzed over 5.5 million messages from more than 6,000 WhatsApp groups, obtained through a large-scale data donation project in India. Our findings reveal a profound disconnect between pre-election hype and ground truth: generative AI constituted only about 1.1 percent of the image dataset once corrected for the measured recall of our detection pipeline, a figure we treat as a lower bound, with no evidence of a viral “deepfake” epidemic. Furthermore, a qualitative taxonomy of this content reveals that the technology was rarely deployed for deception. Instead, it was primarily utilized for “phatic” labor—automating the production of hyper-stylized “Good Morning” greetings, religious iconography, and patriotic art. We argue that AI adoption was constrained by technological friction and by economic incentives, as traditional “cheapfakes” remained comparatively more cost-effective alternatives for political actors. These results challenge the prevailing focus on the threat of AI-driven messaging, suggesting that the technology may not always lead to epistemic disruption, but rather to the efficient reproduction of existing social rituals.
No takes yet. Share an insight, caveat, or question.
Chauchard et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: