RatioLogo
Back

The AI Misinformation Paradox

What if the most dangerous lies on the internet don’t look like news at all, but like entertainment? While the world has braced for a wave of deepfakes designed to topple democracies, new data suggests AI-generated misinformation is conquering the digital landscape by being "lighter," more visual, and strangely more likable than traditional falsehoods.

A Viral Advantage: The Data

A massive analysis of 91,452 flagged posts on X between January 2023 and January 2025 reveals AI-generated misinformation is significantly more viral than conventional misinformation. Synthetic lies are outperforming human-made ones by staggering margins, even when accounting for follower counts and account age.

The Engagement Gap

The numbers reveal a clear advantage for synthetic content compared to non-AI falsehoods:

  • 34.16% increase in likes (p < 0.01)
  • 10.81% increase in retweets (p < 0.01)
  • 10.32% boost in total impressions

How AI Lies Spread Differently

This discovery matters because it proves AI content doesn't need a "mega-influencer" to spread. Despite often originating from smaller accounts, it possesses a unique, inherent persuasive power.

The Virality Formula

Research indicates the secret to this virality lies in the medium and tone:

  • 1.33 times more likely to contain media like images or video.
  • Heavily skewed toward "Entertainment" (30.40% of AI posts vs. 16.60% non-AI).
  • Exhibits a more positive sentiment (24% positive vs. 16% in non-AI).

The Profiles Behind the Content

The authors of this synthetic content have distinct profiles compared to traditional misinformation spreaders.

Distinct Author Characteristics

  • Account Age: Older accounts, averaging 8.32 years vs. 7.31.
  • Follower Count: Smaller mean follower count (585,671 vs. 950,660).
  • Partisanship: Lean more conservative, with a mean partisanship score of 0.272 vs. 0.199 (p < 0.01).

The Unseen Threat

However, catching these synthetic lies remains a moving target. The study's reliance on user-flagged "Community Notes" means the data likely represents only the misinformation savvy enough users caught.

This suggests we may still be blind to stealthier, more sophisticated AI content that avoids detection entirely. While current AI lies are rated as less "harmful," their superior viral potential proves the digital armor of the average user is already being tested by the sheer scale of synthetic media.

Reference:
Drolsbach, C., & Pröllochs, N. (2025). Characterizing AI-Generated Misinformation on Social Media. arXiv:2505.10266v1 [cs.SI]. JLU Giessen, Germany.