AI at the Front Lines of Platform Governance: Using LLMs to Support Illegal Content Reporting under the Digital Services Act

AI assistants can help users report illegal content more accurately: The study found that an "evaluative" AI assistant presenting balanced arguments across different legal categories improved users' accuracy in classifying illegal content under EU law, particularly when the AI made errors, compared to both unaided reporting and conventional AI that suggested only one category.

Different AI designs serve different purposes in content moderation: A simpler AI assistant that suggested a single legal category helped users make faster decisions when the AI was correct, but a more balanced approach that presented multiple perspectives led to more accurate classifications overall—suggesting platforms must choose designs based on whether speed or accuracy is prioritized.

AI assistance alone doesn't guarantee better explanations: Despite improving categorization accuracy, neither type of AI assistant reliably helped users write better substantiated explanations for their reports—the core requirement under the EU's Digital Services Act—indicating that AI support for content reporting requires more than just classification help.

AI errors have predictable impacts on user decision-making: The study systematically tested how different types of AI mistakes affected users, finding that the balanced AI approach was more resilient to certain error types, which has implications for how platforms should design safeguards when deploying AI-assisted reporting tools.

Compliance with EU regulations requires rethinking reporting interfaces: The research highlights that meeting the Digital Services Act's requirement for "sufficiently substantiated" reports in "good faith" is challenging for average users, and AI tools alone cannot solve this without also improving how platforms structure the reporting process itself.