The dark-mode-always-wins advice came from a real source: 2021-2023 A/B test data showing dark-background carousels outperforming light by 5-15% on tap-to-install. Mobile Action, StoreMaven, and a few ASO agencies published these numbers, indie devs repeated them, and the advice calcified.
By 2025 the gap had mostly closed. By 2026, when you look at category-level data instead of aggregate, the picture changes again: dark wins big in some categories, loses in others, and is a coin flip in the middle.
The 2022-2023 tests aggregated across categories. The test pool skewed toward categories where dark is conventional anyway — games, music, photo/video, social. In those categories dark carousels perform better not because dark is universally better but because dark matches user expectation. A bright-pink games carousel would test worse not because it's light but because it breaks category convention.
When researchers ran the same tests in categories where light is conventional (productivity, education, kids apps), light won. The aggregated number was the average of two opposite effects, and someone reading the aggregated number assumed it generalized.
Dark wins decisively in:
Light wins in:
Coin-flip categories:
The honest framework: pick the mode that fits your category convention plus your actual brand, and spend your A/B test budget on more impactful variables. Captions, device tilt, panel order, and palette specifics move the needle more than the binary dark/light choice.
Where dark/light still matters:
Showing dark mode in some panels and light mode in others (often paired with a "supports both modes!" caption on one panel) doesn't help. It looks like the designer couldn't pick. The carousel as a whole should commit to one palette family; the fact that the app supports both modes is communicated by the App Store metadata, not the screenshots.