thumbnail Factor

Thumbnail Visual Style

The overall visual composition approach of the thumbnail. Classified by AI analysis of the thumbnail image to identify which design pattern was used.

Thumbnails compete with dozens of other videos in a viewer's feed. Face closeups with emotional expressions consistently outperform other styles because human faces are processed faster by the brain than text or graphics. But the best style depends on your content — educational channels may benefit from graphic designs that signal 'explainer.'

Each value below shows two numbers when available: Regular/Shorts is the effect on a typical video, and Top 10% is the effect on videos in the top 10% of views (from a quantile regression). Use the Top 10% column to see factors that disproportionately help videos that go big.

Other

+9% views Top 10% +90% not significant p = 0.791 · vs. average

Face And Text

Regular +15% Top 10% +2% not significant p = 0.335 · vs. average

A face (usually the politician) paired with bold text reinforcing the title's hook. Combines human connection with explicit messaging.

Real examples from the dataset

Face Closeup

Regular +4% Shorts +8% Top 10% +2% not significant p = 0.608 · vs. average

A tight close-up of a single face filling most of the frame. The expression is the primary hook — emotion is visible even at small thumbnail sizes.

Real examples from the dataset

Text Heavy

Regular -3% Shorts +13% Top 10% +5% not significant p = 0.737 · vs. average

Real examples from the dataset

Collage

+0% views not significant p = 1.000 · vs. average

Multiple images composited together — often the politician alongside their opponent, a news screenshot, or a relevant image. Tells a mini-story in one frame.

Example title

Split image: the politician on one side, their opponent on the other, with a 'VS' graphic between them

Text Only

+0% views not significant p = 1.000 · vs. average

The thumbnail is primarily text on a colored background with no face or photograph. Common on government channels but performs poorly in most contexts.

Example title

White text on a blue background reading 'COMMITTEE HEARING RECAP'

Graphic Design

Regular -5% Top 10% -12% not significant p = 1.000 · vs. average

A designed graphic — charts, infographics, icons, or illustrated elements rather than photographs. Signals educational or explainer content.

Real examples from the dataset

News Screenshot

Shorts -7% not significant p = 1.000 · vs. average

A screenshot from a cable news broadcast showing the politician on TV. Signals that this is a media appearance reposted to YouTube.

Real examples from the dataset

Frame Grab

Regular -4% Top 10% -17% not significant p = 0.569 · vs. average

Real examples from the dataset

Composite Collage

Regular +2% Shorts -5% Top 10% -5% not significant p = 0.628 · vs. average

Real examples from the dataset

Candid Photo

Regular -8% Shorts -10% Top 10% -16% not significant p = 0.264 · vs. average

Real examples from the dataset