Sora Feed Philosophy

The goal of the Sora feed is to help users learn what is possible and get inspired to create. To achieve this, OpenAI designed a ranking system that optimizes for creativity and active participation rather than simple scrolling.

Users are given steerable ranking to directly control the algorithm, and parents can manage teens' feed personalization and continuous scrolling through ChatGPT's protection features. Additionally, connected content is prioritized to strengthen connections between users.

The recommendation algorithm uses the following signals to provide a personalized experience:

• Sora activity: posts, follows, likes, comments, remixes, and location information
• ChatGPT data: users' conversation history (can be turned off in settings)
• Content engagement: views, likes, comments, "show me less of this" requests, etc.
• Author signals: follower count and past post engagement
• Safety signals: whether content violates policies and its appropriateness

To balance safety and freedom of expression, OpenAI has built guardrails at the point of generation as the first line of defense. Harmful outputs are blocked at the content generation stage, and after generation, automated tools scan all feed content to check for policy violations. This is paired with user reports and human review to maintain a safe environment.