Designing a Creative-First Video Platform

OpenAI has outlined the core philosophy behind Sora's content feed, emphasizing a fundamental shift away from passive scrolling toward active creativity and meaningful connections. The design principles reflect OpenAI's approach to building AI-powered social platforms that balance creative freedom with robust safety measures.

Four Core Principles

The Sora feed is built on four foundational principles:

1. Optimize for Creativity

Ranking algorithms prioritize creativity and active participation over passive consumption. The goal is to inspire users to create and remix content rather than endlessly scroll.

2. Put Users in Control

Sora ships with steerable ranking, allowing users to tell the algorithm exactly what they want to see. Parents can disable feed personalization and control continuous scroll for teens through ChatGPT parental controls.

3. Prioritize Connection

The platform favors content that strengthens and forms new connections, especially through "Cameo" flows. Connected content from people users know ranks higher than generic global content.

4. Balance Safety and Freedom

Robust guardrails prevent unsafe content while leaving room for creative expression and community building. The system is designed as a "living, breathing" platform that adapts based on real-world use.

How Personalization Works

Sora's recommendation system uses multiple signals to personalize each user's feed:

User Activity on Sora: - Posts, followed accounts, liked and commented content
- Remixed content
- General location (city-level) based on IP address

ChatGPT Data Integration: - Users' ChatGPT history (can be disabled in Data Controls)

Content Engagement Signals: - Views, likes, comments
- "See less content like this" instructions
- Remix activity

Author Signals: - Follower count
- Past post performance
- Historical engagement

Safety Signals: - Content violation status
- Age-appropriateness ratings

Parents can turn off feed personalization entirely and manage continuous scroll for teen accounts using parental controls.

Safety Architecture: Prevention and Enforcement

OpenAI employs a multi-layered safety approach:

Generation-Level Guardrails

Because all Sora content is AI-generated, safety filters operate at the point of creation—preventing unsafe content before it exists. This "upstream" approach is more effective than post-publication moderation.

Content Distribution Guidelines

Content violating OpenAI's Usage Policies may be removed from the Feed, user galleries, and other sharing surfaces. Prohibited content includes:

- Graphic sexual or violent content
- Extremist propaganda and hateful content
- Self-harm promotion or disordered eating content
- Bullying and appearance-based critiques
- Dangerous challenges targeting minors
- Engagement bait and low-quality content
- Unauthorized likeness recreation
- Intellectual property infringement

Automated + Human Review

Automated scanning systems continuously monitor all feed content for policy compliance. Human reviewers complement automation by:

- Monitoring user reports
- Proactively checking feed activity
- Catching edge cases automation may miss

User Reporting

Users can report content that violates policies, triggering both automated and human review.

Balancing Restriction and Expression

OpenAI acknowledges the tension between safety and creativity: "Too many restrictions can stifle creativity, while too much freedom can undermine trust."

The approach combines:
- Proactive guardrails where risks are highest (e.g., generation-level filters)
- Reactive systems that give users creative freedom while enabling rapid response to problems

This philosophy builds on lessons from ChatGPT's 4o image generation model, where similar balancing principles have proven effective.

An Evolving System

OpenAI emphasizes that Sora's recommendation and safety systems are "living, evolving systems." The team expects to continuously refine the balance based on user feedback and real-world learning.

"We know we won't get this balance perfect from day one," the announcement states. "Your feedback will be essential in helping us refine them."

TL;DR

- Sora's feed prioritizes active creativity over passive scrolling, with steerable ranking and parental controls
- Four core principles: optimize creativity, user control, connection, and balanced safety
- Multi-layered safety: generation-level guardrails, automated scanning, human review, and user reports
- Platform evolves based on user feedback and real-world use patterns


Source: OpenAI Announcements - Sora Feed Philosophy