The two creators behind a small newsletter and podcast used Claude Code to analyse the thumbnails of 776 videos across 57 YouTube channels and pull out what the best-performing ones have in common.
- First reported
- smallcreatorbigworld.substack.com · 5 Apr 2026
- Added
- 13 Oct 2026
- Source checked
- 13 Oct 2026
What they did
They described the goal in plain English, had Claude Code gather the videos into a database, and iterated on the analysis by giving feedback on each result. They also built a channel analyser that finds slightly larger peer channels and suggests video ideas only when a pattern shows up in at least three examples.
What happened
The poster says the thumbnail study, which would have taken three days by hand, was done in an afternoon.
Keep in mind
The post does not share the findings in detail or explain how the analysis was checked, and it promotes the authors' own workshop.
Try it with your agent
Collect the last <number> videos from these channels: <list of peer channels>. For each, save the title, view count relative to the channel's average, and the thumbnail image into a local database. Then compare the top 20 percent against the rest and tell me which visual patterns appear in at least three top videos, with links to the examples.
Replace the <angle brackets> with your own details.
Original source
smallcreatorbigworld.substack.comUser storySomeone who uses the agent described their own setup. Figures are theirs.
Tools for this kind of work
The whole toolkit →
