Three public thumbnail tests that say more than the winner label

Top 3 tests

1. Brad Smith: the hook beat the decoration

• Who ran it: Brad Smith, who publishes YouTube growth tutorials for AutomationLinks.
• What changed: three versions of the same basic thumbnail, with the hook text rewritten so each variation pushed a slightly different promise.
• Result: thumbnail 2 won with 38% of the watch-time share, ahead of thumbnail 1 at 31% and thumbnail 3 at 30%. 1
• Why it likely won: the image stayed familiar, so the text had to do the real work. A sharper hook lowers the amount of reading a viewer has to do in the feed.
• How to apply it: test the hook before you test color, borders, or tiny layout changes. If the promise is weak, prettier packaging usually just makes the weakness easier to see.

2. Jeven Dovey: small differences produced a preferred result, bigger gaps produced a clearer winner

• Who ran it: Jeven Dovey, on videos walking through YouTube's Test & Compare feature.
• What changed: one two-thumbnail test and one three-thumbnail test, each using separate thumbnail concepts rather than tiny edits.
• Result: one two-thumbnail test ended with a preferred result at 52.9% versus 47.1%, while a separate three-thumbnail test ran from April 2 to April 5 and picked thumbnail 2 at 36%, ahead of 32.1% and 31.9%. 2
• Why it likely won: the tighter result shows what happens when the variants are close; the three-way run shows that a bigger concept gap gives the tool a cleaner read.
• How to apply it: start with noticeably different thumbnail ideas, then refine the winner. If the spread stays narrow, treat the result as directional and run a second pass.

3. Jamar Diggs: small channels can get a label without getting a verdict

• Who ran it: Jamar Diggs, on a client channel and on his own channel.
• What changed: three thumbnail options were tested on a client video, with a second round of testing on his own channel.
• Result: on the client video, thumbnail 2 reached about 40% of watch-time share, but the test still was not conclusive when the video had only around 200 views. On his own channel, the report was also inconclusive. 3
• Why it likely won: the sample was thin, so the report could point toward a preferred option without being confident enough to call a winner.
• How to apply it: if a channel is still early, the tool is useful as a directional read, not as a final authority. Manual swaps and longer observation windows can be more honest than waiting for a label that never becomes certain.

The practical rule