Core Principles
OpenAI has designed the Sora feed with a clear vision: to foster learning about creative possibilities and spark inspiration for content creation. The platform operates on several foundational principles:
- Creativity-First Design: The ranking system prioritizes creative content and active user engagement over passive consumption, making the platform more enjoyable and interactive.
- User Empowerment: The feed includes customizable ranking options, allowing users to direct the algorithm based on their preferences. Parents can disable personalization features and manage continuous scrolling for teenage users via ChatGPT's parental control settings.
- Building Community: Sora encourages meaningful connections between users, particularly through innovative Cameo features. Content from connected users receives preferential treatment over unconnected global content.
- Safety-Freedom Balance: The platform maintains broad accessibility and safety standards through comprehensive guardrails that prevent harmful content generation, while preserving space for creative expression and community interaction.
Recommendation System Mechanics
The recommendation engine creates tailored experiences to support creative inspiration. The personalization system considers various factors:
Activity-Based Signals:
- User interactions on Sora: posts created, accounts followed, content liked and commented on, and remixed materials
- General device location data (city-level) derived from IP addresses
- ChatGPT usage history (optional, can be disabled in Data Controls)
Content Performance Metrics:
- Viewing statistics, engagement rates (likes, comments), remix activity
- User feedback such as "show less similar content" requests
Creator Information:
- Follower statistics, posting history, and historical engagement levels
Safety Assessments:
- Content appropriateness evaluations and policy compliance status
These combined signals help predict and surface content that aligns with individual user interests and creative goals.
Safety and Expression Framework
Sora implements a multi-layered approach to maintain platform safety while fostering creative freedom:
Content Moderation Standards
The platform may remove content that breaches OpenAI's Global Usage Policies or fails to meet Sora Distribution Guidelines, including:
- Explicit sexual material
- Graphic violence or content encouraging violent behavior
- Extremist messaging
- Discriminatory or hateful content
- Self-harm or eating disorder promotion
- Dangerous physical challenges targeting minors
- Depression glorification
- Promotion of controlled substances or age-restricted products
- Low-quality engagement bait
- Unauthorized reproductions of living persons or deceased public figures
- Intellectual property violations
Protection Mechanisms
Generation-Level Prevention: The primary defense occurs during content creation, with built-in guardrails preventing unsafe material from being produced.
Automated Screening: All feed content undergoes automated review for policy compliance and age-appropriateness, with special filters for teen accounts. These systems receive continuous updates to address emerging risks.
Human Oversight: A moderation team actively monitors user reports and conducts proactive content reviews to identify issues that automated systems might overlook. Users can submit reports for content they believe violates policies.
Adaptive Balance
OpenAI recognizes that excessive restrictions can limit creativity while insufficient moderation undermines platform trust. The approach combines proactive safeguards for high-risk areas with responsive report-and-removal procedures, allowing creative exploration within safe boundaries. This strategy builds upon successful implementations in ChatGPT's 4o image generation capabilities.
The recommendation and safety systems are designed to evolve based on real-world usage and community feedback, with OpenAI committed to ongoing refinement and improvement of these frameworks.