Why and How People Check Generative AI Output for Mistakes

Public awareness of AI mistakes is already high: The survey of 1,503 representative U.S. respondents found widespread recognition that generative AI can produce errors like hallucinations and inaccurate content, suggesting users understand these systems are not infallible.

Many people are already taking verification steps: A significant portion of respondents reported actively checking generative AI output by comparing results with other online resources, indicating that users are developing practical strategies to validate AI-generated information.

User behavior shows demand for transparency: The fact that people are independently verifying AI output suggests there is public appetite for clearer information about when and why AI systems make mistakes, creating an opportunity for better user education.

In-product disclosures could improve responsible use: The researchers recommend that AI companies provide explanations and built-in disclosures about potential mistakes, implying that current systems may lack sufficient warnings about limitations and error types.

Responsible AI requires meeting user expectations: The gap between what users are already doing (checking outputs) and what systems currently communicate about their limitations suggests that responsible AI development should include better mechanisms to help people understand and navigate AI mistakes.