youtube-rl-tracker

The youtube-rl-tracker is a specialized tool for creators and AI agents operating in the content production space, designed to implement a 'poor man's reinforcement learning' loop for YouTube channel growth. By treating every video upload as a controlled experiment with specific hypotheses regarding visual hooks and title copy, the agent enables systematic performance improvement through iterative testing. It bridges the gap between raw data from YouTube Studio and actionable content strategy, allowing users to move beyond intuition toward evidence-based optimization.

This skill is intended for technical content creators, growth hackers, and autonomous agents tasked with managing a brand's digital presence. It helps users analyze why certain videos gain traction while others stagnate by tracking key performance indicators like Click-Through Rate (CTR), average view duration (retention), and normalized views-per-day. By centralizing the feedback loop, it transforms the unpredictable nature of viral content into a structured data science workflow.

• Systematic Logging: Uses Notion as a database schema to store outcome metrics alongside input features like thumbnail styles (Face+UI, Meme, Talking Head) and title hooks.

• Performance Attribution: Directly compares winners and losers based on specific variables such as the presence of text overlays, product UI visibility, or brand name mentions in titles.

• Hypothesis Testing: Facilitates A/B testing frameworks, allowing you to test specific thumbnail patterns against plain talking head approaches with a goal of achieving 10x performance improvements.

• Integration Ready: Designed to work alongside YouTube Studio workflows, enabling the agent to update video metadata, refresh thumbnails, and perform periodic reviews.

• Analytical Sorting: Includes a weekly review process that sorts by views-per-day, groups content by category, and validates hypotheses through binary outcome tracking.

• Inputs: Requires manual or automated input of Video Title, CTR, Retention, Thumbnail style, and categorical tags (e.g., AI Finance, Automation, Tutorials).

• Output: Data-driven insights identifying high-performing content patterns to apply to future uploads.

• Constraints: Performance data is most reliable after a 48-72 hour 'wait' window to allow for organic discovery and reach.

• Best Practices: Normalize all data by days live to ensure accurate comparisons between older and newer content.