How I built a YouTube performance classifier that adjusts tomorrow's video script bias

Short summary

The author built a Python script that analyzes YouTube video performance daily, classifies videos using median-based thresholds (not ML), and writes performance patterns back to a knowledge bank that their script generator reads. This creates a closed feedback loop: the script generator produces videos, performance data feeds back as context for the next generation. Key insights include using median for outlier resistance, a 72-hour grace period to avoid flagging young videos, and title-matching to reconnect archetype labels to performance data.

• •Built daily performance classifier using median-based bucketing, not ML
• •Feeds insights back to script generator via a knowledge bank document
• •Closed feedback loop enables continuous content optimization

Generated with AI, which can make mistakes.