James Ding
Mar 17, 2026 14:27
OpenAI’s Sora video platform launches with recommendation system designed to inspire creation over passive scrolling, featuring steerable ranking and parental controls.
OpenAI is betting its AI video platform can avoid the attention-hijacking pitfalls that plague TikTok and Instagram. The company published its Sora Feed philosophy on February 3, 2026, revealing a recommendation system built around an unusual premise: rewarding creativity over engagement.
The approach directly challenges conventional social media wisdom. Where Meta and ByteDance optimize for time-on-platform, Sora’s algorithm explicitly favors content likely to inspire users to create their own videos. Passive scrolling, the dopamine-loop that powers most feed-based platforms, isn’t the goal here.
How the Algorithm Actually Works
Sora’s personalization pulls from several signal sources: your posts, followed accounts, likes, comments, and remixed content. Location data from IP addresses factors in. Perhaps more controversially, the system can incorporate your ChatGPT conversation history—though OpenAI says users can disable this in Data Controls.
The “steerable ranking” feature stands out. Users can tell the algorithm what they’re in the mood for using natural language, rather than relying on endless thumbs-up/thumbs-down training. Connected content—videos from people you follow or interact with—gets weighted above viral global content from strangers.
Parents running ChatGPT parental controls can disable feed personalization entirely for teen accounts and manage continuous scroll settings.
Content Guardrails Built at Generation
Because every piece of content originates from Sora’s AI generation, OpenAI claims a structural advantage on moderation. Guardrails kick in before content exists, not after it’s already spreading. The company blocks graphic sexual content, violence promotion, extremist material, self-harm content, and what they call “engagement bait.”
Automated scanning checks all feed content against OpenAI’s Global Usage Policies. Human reviewers monitor reports and proactively audit feed activity. Teen accounts face additional filtering for age-inappropriate material.
The policy explicitly prohibits recreating living individuals’ likenesses without consent—a direct response to deepfake concerns that have dogged AI video tools.
The Tension OpenAI Acknowledges
The company isn’t pretending it has content moderation figured out. “Too many restrictions can stifle creativity, while too much freedom can undermine trust,” the policy states. OpenAI describes its approach as proactive guardrails for high-risk content combined with reactive report-and-takedown for edge cases.
This mirrors the strategy used for ChatGPT’s 4o image generation model. Whether it scales to video—where context and intent prove far harder to parse—remains the open question.
OpenAI explicitly frames Sora’s recommendation system as “living and evolving,” signaling adjustments will come as real-world usage exposes gaps in the current approach. For creators and advertisers watching AI video platforms mature, the next several months will reveal whether creativity-first ranking can sustain user growth without the engagement tricks that made social media both addictive and toxic.
Image source: Shutterstock
