Longer Videos Are No Longer the Needle Mover: The Quick-Hit Carousel Results

Do longer videos still give you an edge in the Amazon carousel? Do custom thumbnails still matter? Does making more videos for a product actually do something for you? My carousel data analysis is done, I checked almost 900 products, and here’s the quick-hit version. Short answer: the thing we’ve all been treating as the primary lever for the last year, video length, is no longer the needle mover. Here’s what is, what isn’t, and what I’m doing about it.

The Carousels Stopped Rotating

First, the biggest change versus a month and a half or two months ago: consistency. Across those almost 900 products, 94% of the time I got the same set of carousel results over and over. Refresh the page, same lineup. That’s the opposite of the carousel chaos we all lived through over the summer, where every refresh was a brand-new set of videos.

About 301 products did rotate and give me fresh results on each refresh, and I refreshed those a hundred times apiece. So I graded both groups separately, rotating and non-rotating, to see whether longer videos, custom thumbnails, or video count behaved differently in either case. They didn’t, really. The patterns held.

Video Length: The Edge Is Mostly Gone

Longer videos do not have anywhere near the edge they had a couple of months ago. Videos over two minutes still have a slight advantage, and five-minute-plus videos a little more than that. But they’re now being pushed hard by videos under 30 seconds, which have closed the gap significantly. I don’t want to call it a coin flip, but it’s pretty dang close, and it’s not enough for me to keep length as a core part of my strategy.

I’m letting go of the notion that longer videos are one of the primary movers for the carousel. If anything, this data gives you a good reason to make a longer anchor video and a 30-to-45-second vertical short for the same product.

Custom Thumbnails: No Ranking Edge

Custom thumbnails were almost a coin flip too. The only position where a custom thumbnail had any edge was slot one, and it was minimal: based on how many creators use customs versus don’t, I expected about 62% visibility in slot one, and it came in at about 64%. Every other slot landed exactly where volume predicted it would. Slot five actually underperformed for custom thumbnails; there were more non-custom thumbnails sitting there than you’d expect.

And yes, I weighted this. There are now more creators using custom thumbnails than not, so I counted the totals on each side to make sure the raw numbers didn’t skew the result. Bottom line: no real ranking edge for a custom thumbnail. I still have a bunch of reasons they matter for user experience and winning the click once you’re in the carousel, but that’s a different conversation than ranking.

Tagging and Orientation

I checked how many products a creator tags per video. No effect. I did not break out vertical versus horizontal this time, because so many people are shooting vertical that the under-30-second bucket is already mostly vertical. If you want a rough proxy, read “under 30 seconds” as “vertical short” and you’ll be close.

The MABS2 Label on Every Carousel

Something I had never seen before popped up during this pass: every single product carousel I looked at is now labeled with an MABS2 scoring system. I believe that stands for multi-armed bandit system, a very common testing framework Amazon uses across a bunch of areas. The idea is that Amazon serves a set of results, watches how customers engage, and uses that engagement to decide what gets served next.

I don’t remember seeing MABS2 on anything before the two months of carousel chaos. When that chaos was happening, a lot of us said it was probably a test. Now every carousel carries the label and the results are stable. I think what we’re seeing now is intentional, and there’s a real possibility it’s person-to-person, tailored to what gets a given shopper to click.

Why Your Results Might Look Different

Go eyeball test your own carousels. You’re going to see the under-30-second videos, and you’re most likely going to see that the lineup doesn’t rotate much. When one does rotate, you can usually get a lot of rotations out of it. Some products never moved no matter how many times I hit refresh, then suddenly broke loose and rotated a hundred times in a row. Others never budged.

The specific videos in my carousel may not be the videos you see in yours. But I think the methodology behind which videos get selected is the same for everyone, even if the outputs differ. Amazon is a crazy place sometimes, and somebody else running this test might see different numbers. These are mine, across about 900 products, and I’ll do a fuller breakdown of the data in a longer video.

The New Strategy Conversation: Good Short Videos

Here’s the recap. Are longer videos getting the edge they were? No. Do custom thumbnails give any ranking superiority? Nope. Does making more than a few videos for a product give you an edge? For exposure, yes. For ranking and the number-one slot, no, and probably a slightly decreased shot at it.

I think there’s a viable strategy conversation here: start incorporating shorter videos intentionally, but not just because they’re short. We know what a good long Amazon influencer video looks like, in-depth and thorough. What does a good 30-second-and-under video look like? What would cause somebody on Amazon to actually buy in that window? And what can we also use on other platforms, because we know Amazon has something up its sleeve on the social side.

That’s where I’m pointing my energy next. In the meantime, the fundamentals haven’t moved: find products that convert, make unique videos for them, and don’t waste slots. If you want help with the product side of that equation, that’s what Oink is built for.