The Shift from Visual to Textual Risk

Policymakers and the public have focused heavily on visual deepfakes, but evidence suggests this is a misallocation of concern. A meta-analysis by the International Panel on the Information Environment (IPIE), covering 60 effect estimates from 24 randomized controlled trials, indicates that AI-generated text is currently more persuasive than truthful information. While public skepticism toward visual AI has grown—partly due to the visibility of 'AI slop'—users remain highly susceptible to machine-generated text, which is increasingly embedded into search engines and conversational interfaces. Because text is cheap, scalable, and highly customizable, it represents a more immediate and pervasive threat to the information ecosystem than visual media.

Effective Interventions: Preventive vs. Reactive

The IPIE research identifies 'preventive corrective information' as the most consistent and effective intervention. Rather than attempting to debunk misinformation after it spreads, platforms should provide users with context and knowledge before they encounter potentially misleading content. This includes:

  • Explaining how AI systems generate content.
  • Providing reminders that AI models are prone to errors.
  • Using short prompts that encourage users to critically evaluate the credibility of the information they are consuming.

While content labeling is a common policy tool, the authors warn that it is not a 'plug-and-play' solution. The efficacy of labels varies wildly based on design, wording, and timing; poorly implemented labels can be ineffective or even trigger a backfire effect. Consequently, labeling requires rigorous, platform-specific testing rather than broad, uniform mandates.

Addressing the Research Gap

Current policy efforts are hampered by a significant evidence gap. Most existing research is concentrated in the United States and Western Europe, leaving the global information environment under-studied. Furthermore, there is a mismatch between the speed of AI development and the pace of academic research; by the time studies are published, the underlying technology has often evolved. To close these gaps, the authors argue that policymakers must:

  • Mandate timely access to platform data for independent researchers.
  • Fund cross-regional research to ensure policies are globally applicable.
  • Move away from reactive, static policy frameworks toward adaptive, evidence-based interventions that account for the conversational nature of modern AI.