AI Rules? Characterizing Reddit Community Policies Towards AI-Generated Content

Research Background and Issues

• Problem or Challenge: With the rise of generative AI technologies (AIGC), online communities face challenges such as potential declines in content quality, authenticity, and trust. Reddit, characterized by its decentralized nature, allows different subreddits to establish their own rules to manage AI-generated content. However, the responses of these communities have not yet been comprehensively quantified and analyzed.
• Significance: The proliferation of generative AI content may lead to issues such as data pollution, questions about content authenticity, and a crisis of social trust for platforms. As an important space for learning and social interaction, studying how Reddit communities respond to generative AI can provide valuable insights for other online platforms in formulating policies and designing tools.
• Research Motivation: While some qualitative studies have explored the impact of AI on online communities, there is a lack of large-scale quantitative analysis to capture the overall trends in the evolution of rules on the Reddit platform.

Solution

• Method or Solution: The authors conducted two large-scale crawls and analyses (in 2023 and 2024) of Reddit community rules and their changes, collecting rules from 99,969 English-speaking communities and designing a novel classification system for AI-related rules.
• Innovations:
  • Proposed a new classification system to describe AI-related rules (including rule motivations, stances, and content types).
  • Combined quantitative analysis with perspectives from existing qualitative research to reveal platform-level patterns and trends.
  • Released the dataset to support future related research.
• Implementation Steps and Key Techniques:
  • Conducted two rounds of large-scale crawls of Reddit subcommunities and their rules, constructing and cleaning several sub-datasets.
  • Applied a hybrid approach of manual and automated labeling (supported by LLMs) to classify rule types, community types, and AI-related rules at multiple levels.
  • Performed statistical analyses on rule evolution, rule existence, and rule content types to address three specific research questions.

Research Findings

• Specific Findings:
  • Adoption of AI-related rules doubled within a year: from 0.6% in 2023 to 1.2% in 2024, with larger communities being more likely to adopt such rules (17% of the largest 10% of communities had AI-related rules in 2024).
  • AI-related rules were more common in communities focused on art or celebrity topics and less common in social support communities.
  • Most AI-related rules targeted AI-generated image content, emphasizing quality and authenticity, with unconditional bans on generative AI content being particularly prevalent.
  • Rules predominantly focused on restrictive requirements (96.8%) rather than prescriptive guidance (67.7%).
• Comparison with Existing Solutions:
  • Unlike prior qualitative studies focusing on community rules or AI impacts, this study provides a quantitative foundation for observing platform-level trends and differences between specific communities.
  • By discussing inter-community response mechanisms, this study offers detailed statistics and a novel classification perspective, enhancing the contextual sensitivity of policy design.
• Experimental or Evaluation Results:
  • Multi-level statistical analyses confirmed that larger and more complex communities were more likely to establish rules, and these rules were especially significant in communities centered on visual content generation.
• Limitations:
  • The dataset only includes public, English-speaking communities, potentially overlooking private or non-English community rules.
  • The analysis focuses solely on explicitly recorded rules, without addressing implicit norms or governance methods outside of formal rule texts.
• Future Directions:
  • Expand research to multilingual or private communities to understand governance differences across cultural or linguistic contexts.
  • Compare similar communities with differing rules to study the impact of varying policies on member behavior and engagement levels.
  • Investigate the long-term effects of generative AI on community interaction and content sharing (e.g., text and images) to guide platform tool design.

Conclusion

Through large-scale rule classification and analysis, this study quantifies the diverse responses of Reddit communities to generative AI content, revealing the complexities and trends of platform governance in the face of rapid technological changes. Additionally, its open dataset will support future HCI research in exploring richer governance mechanisms and support tools. These findings provide recommendations for designing online tools that encourage authenticity and quality while highlighting how platforms can balance contextual adaptability and technical support in the face of community autonomy.