Testing thumbnails properly: what YouTube's A/B test measures

How YouTube's test works

In YouTube Studio on a computer, you can test up to three titles, three thumbnails, or combinations of both on one video. YouTube shows the versions to different viewers and, when the test ends, uses the one with the highest share of watch time. Results are labelled Winner, Performed same, or Inconclusive. Tests usually finish within two weeks, depending on how many impressions the video gets.

Not every video qualifies. Shorts, scheduled live streams and premieres can't be tested, nor can videos made for kids, marked for mature audiences, or set to private.

Why watch time, not clicks

A click-through rate rewards whatever gets the tap, including a thumbnail that promises something the video doesn't deliver. Viewers who feel misled leave quickly. Scoring by watch time share rewards the version that attracts people who then keep watching, which is what a channel actually needs. That's also why a 'winning' thumbnail can have a lower click rate than the loser.

Designing versions that teach you something

If version B changes the face, the colours, the text and the crop all at once, a win tells you that B is better but not why. Change one main idea per version, and write down what you expect before the test starts.

• Text vs no text, with the same picture.
• Close-up subject vs wider scene.
• A different promise: the problem vs the result.
• Warm vs cool background, checked in both light and dark mode first.

Limits of any test

"Inconclusive" is a real answer: the versions were too similar to matter for this audience. A small channel may also wait a long time for enough impressions. And a result applies to that video and that audience, so treat it as evidence for your next design, not a law. Before testing at all, make sure every version is readable at small sizes; a test can't rescue a thumbnail nobody can read.