Spent the past few months building CreatorPilot, a tool that connects to
YouTube Analytics API and uses Claude to diagnose why specific videos
underperformed.
The core problem I was solving for myself: YouTube Studio gives you tons
of data but zero interpretation. You can see your CTR dropped, but not
why, or what to actually do about it.
Tech stack:
Next.js for the frontend
YouTube Analytics API + YouTube Data API for the data layer
Claude API for the analysis and recommendations
Stripe for billing
The interesting technical challenge was getting the AI analysis to be
useful instead of generic. Early versions just said things like "consider
improving your thumbnail" for every low-CTR video, which is technically
true but useless. Had to build in actual comparison logic against the
channel's own historical baseline before passing data to the model, so
the output references real numbers instead of generic advice.
Currently supports 5 languages (EN/KO/JP/DE/ZH) since YouTube creators are
obviously a global audience and most analytics tools are English-only.
Free plan available if anyone wants to poke around. Would love feedback
from other devs, especially on how you've handled the "AI sounds confident
but might be wrong" problem in your own projects.
creatorpilothq.com
