In this video I’m building what I wish YouTube Studio had: a YouTube AI Analytics tool that helps me understand what’s actually working (and what the algorithm is ignoring). My goals are pretty simple: make better videos, design thumbnails that lift click-through without being clickbait, and generate titles that are interesting but still true to the content. I’ve already played with captions + ChatGPT before, but here I push the MVP further by wiring up a full stack app that can pull real channel and video analytics. I build this on Next.js 14 (first time using it properly), with NextAuth for Google login so I can lock it to YouTube accounts. I hit the classic token expiry problem, so I add MongoDB to store refresh/access tokens and cache channel/video data to avoid burning YouTube API quota. Once the dashboard is working, I bring in the fun stuff: Cloudinary’s AI for video highlights (content-aware “interesting parts” detection) and thumbnail analysis. I upload thumbnails, generate tags/captions, and even do on-the-fly edits like removing objects or swapping backgrounds—without opening Photoshop. Finally, I use OpenAI chat completions + the transcript to generate title ideas that are contextual, useful, and not overly clickbaity.