How to A/B Test Your Cover Art
Published · updated · by Loopin
Your artwork is almost never seen at the size you designed it. It is seen at about 60 pixels, in a list, next to other artwork — and that is the only condition worth testing under.
Test at 60 pixels, in context
Shrink both options to 60 pixels and place them in a screenshot of an actual playlist, surrounded by real covers. Almost every design decision that felt important at full size becomes invisible, and one or two become the whole image.
What survives at that size is contrast, one shape, and colour. Text almost never survives, including your own name — which is why so many covers are effectively blank in the only place they are seen.
The free test: post both as a story poll
Two options, one poll, no context about which you prefer. It takes five minutes and gives you a rough answer, and the comments are usually more useful than the vote — people volunteer what they actually saw.
Treat this as directional rather than decisive. Your existing audience already knows you, and the artwork's hardest job is with people who do not.
The better test: run both as the same ad
Five dollars a day for three days, one ad set, two creatives differing only in the image. The system will spend more on whichever performs, and cost per landing page view gives you a real answer from strangers.
This is the only method that tests the artwork on the audience it actually has to work on. Fifteen dollars for a decision you keep for the life of the release is cheap.
The one that costs nothing: the squint test
Shrink it, blur it, and look from across the room. If you cannot tell what it is, neither can somebody scrolling. Then show it to someone for two seconds and ask what they saw — if they describe a mood rather than a thing, it is too diffuse.
Three seconds of this rules out most options and it needs no audience at all.
What to actually decide on
Recognisability at thumbnail size, and nothing else. Not which is more beautiful, not which better represents the song, not which you prefer — those are the criteria that produce covers nobody can pick out of a list.
Then keep the visual language across the release. Two singles and an EP that look related do more for recognition than any individual cover does, and that decision is free.
Frequently asked questions
How do I test music cover art?
Shrink both options to about 60 pixels and place them in a screenshot of a real playlist. Then either run a story poll for a directional answer, or run both as one ad set for three days for a real one.
Should album art have text on it?
Usually not. Text rarely survives at the 60-pixel size artwork is actually seen at, including your own name — which is why so many covers are effectively blank where it counts.
What should I judge cover art on?
Recognisability at thumbnail size. Not beauty, not how well it represents the song, not personal preference — those criteria produce covers nobody can pick out of a list.