Deepfakes in the 2025 Canadian Election: Prevalence, Partisanship, and Platform Dynamics

View on arXiv ← Back to list

Authors: Victor Livernoche, Andreea Musulan, Zachary Yang, Jean-François Godbout, Reihaneh Rabbany

Published: 2025-12-15 21:49:40+00:00

AI Summary

This study analyzes 187,778 posts containing images across X, Bluesky, and Reddit during the 2025 Canadian federal election to determine the prevalence and dynamics of AI-generated content (deepfakes). Findings indicate that 5.86% of shared images were deepfakes, with right-leaning accounts sharing them more frequently and often with conspiratorial or defamatory intent. Although deepfakes were present, their overall reach was modest, yet highly realistic fabricated images demonstrated higher engagement potential.

Abstract

Concerns about AI-generated political content are growing, yet there is limited empirical evidence on how deepfakes actually appear and circulate across social platforms during major events in democratic countries. In this study, we present one of the first in-depth analyses of how these realistic synthetic media shape the political landscape online, focusing specifically on the 2025 Canadian federal election. By analyzing 187,778 posts from X, Bluesky, and Reddit with a high-accuracy detection framework trained on a diverse set of modern generative models, we find that 5.86% of election-related images were deepfakes. Right-leaning accounts shared them more frequently, with 8.66% of their posted images flagged compared to 4.42% for left-leaning users, often with defamatory or conspiratorial intent. Yet, most detected deepfakes were benign or non-political, and harmful ones drew little attention, accounting for only 0.12% of all views on X. Overall, deepfakes were present in the election conversation, but their reach was modest, and realistic fabricated images, although less common, drew higher engagement, highlighting growing concerns about their potential misuse.

Key findings

The prevalence of deepfakes was 5.86% across all platforms, peaking at 9.70% among posts from right-leaning users on X, compared to 6.45% for left-leaning users. Most deepfakes were benign or non-political; politically harmful content received minimal amplification, accounting for only 0.12% of total views on X. However, realistic fabricated images, while less common, achieved higher engagement metrics than other political deepfake categories.

Approach

Researchers collected image-containing posts related to the 2025 Canadian election from social media platforms and applied a high-accuracy deepfake detector (ConvNeXt-V2) trained on diverse synthetic models. A Vision-Language Model (VLM) was used to infer the communicative intent of the deepfakes, and an LLM was used to estimate the political leaning of the authors, enabling a multi-layered analysis of circulation and impact.

Datasets

X (Twitter), Bluesky, Reddit (for analysis); OpenFake, DF40, GenImage (for detector training).

Model(s)

ConvNeXt-V2-Base (Deepfake Detector), Qwen3-VL-32B-Instruct (Intent Classifier), Llama 3.3 70B Instruct (Leaning Classifier).

Author countries

Canada

← Previous