Most PDP video statistics are not measurements

Product pages with video convert at 4.8%. Without video, 2.9%. A 65% lift.

You have probably seen some version of that number. It turns up in dozens of articles, usually without a link and rounded a little differently each time. We tried to trace where it comes from.

Three things that get called a statistic

Almost every conversion figure quoted for product page video is one of three things, and only one of them is a measurement.

A vendor benchmark. A company that sells video software publishes data showing that video works. The numbers may be real, drawn from real merchant stores, but the sample selects itself. It contains only brands that bought video software and cared enough to deploy it well. Nobody publishes the merchants who installed it and saw no change. This is marketing with a chart attached, and it is by far the most quoted category.

A survey. "82% of marketers say video delivered good ROI." That is a real finding about what marketers believe. It says nothing about conversion, yet it keeps getting quoted as if it measured performance. Think about what marketers would say if you asked them about any budget line they had personally approved.

A correlation. Pages with video convert better than pages without. That is almost certainly true and close to useless. Brands that pay for video also pay for better photography, better copy, faster pages and better merchandising, and they usually have more budget overall. Video on a PDP tells you the company is investing in its PDPs. Reading it as the cause of the lift is a guess presented as data.

What the citation chains actually look like

We followed one.

An article claims shoppable video converts at 9% to 17%, against 2.1% to 3.5% for static pages, and reports a 328x return on production spend. Every citation in it leads to a company that sells shoppable video, or to a blog citing a company that sells shoppable video. Its headline baseline, attributed to "50,000 plus listings", has no link at all. The site publishing it sells AI product imagery.

Nobody in that chain is committing fraud. They are doing normal content marketing. But four hops later a reader meets a hard number, sees it repeated in several places and reasonably assumes it is established. Repetition proves nothing when every copy traces back to the same few companies with something to sell.

We are not naming them. The pattern matters more than the names, and a careless takedown would only make us one more hop in someone else's chain.

The number that would actually mean something

Only one kind of evidence would tell you what video does for your business, and that is a randomized test on your own catalog.

Same products, same traffic sources, same prices, same period, with one variable changed and enough sessions to separate signal from noise.

That result would describe your business, which no published benchmark can do. Your category, your price point, the quality of your photography and how well your customers already know your fit will each move the result more than the presence of video.

So why does almost nobody have that number?

Because the test used to cost more than the answer was worth

Until recently, putting a clip on a PDP meant commissioning video. Studio time, talent, styling, an editor, a grade and a delivery spec. For a catalog of any size that is a real budget line.

So to find out whether video was worth the investment, you first had to make the investment. Nobody funds a test that costs as much as the rollout. Brands either committed on faith, usually on the strength of a vendor benchmark, or did nothing and kept quoting the statistic to each other in meetings.

The vendor statistic filled that gap. It existed because running your own test was unaffordable.

What actually changed

The conversion rate didn't change. The cost of the experiment did.

With AI-assisted production, an approved still can now be extended into a short clip at a cost much closer to a retouched image than to a shoot. The garment, lighting and color stay the same because the clip starts from your own approved frame rather than a new capture, so the campaign look carries through. A retoucher then checks the clip against that frame before it goes anywhere near a product page.

That does not prove video works, and we don't want to overstate it. What it does is move a real test out of the budget category of "commission a video production" and into the budget category of "add some assets to 30 product pages".

The claim we are comfortable making is narrow, and more useful than a benchmark. You can now afford to find out.

How to run it so the answer is real

If you run this test, run it properly. Most PDP video tests we hear about get called after two weeks on a few hundred sessions, which measures nothing.

• Choose the middle of your catalog. Skip your hero product, whose numbers are distorted by brand traffic and repeat buyers, and skip the tail, which will never reach significance. Pick 20 to 40 SKUs with steady mid-range traffic.
• Hold out a control in the same period. Comparing this month with last month won't work. Seasonality, promotions and shifts in your traffic mix will swamp the effect you are looking for.
• Decide your threshold before you look at the data. Write down what lift would justify a rollout. If you decide afterwards, the test will confirm whatever the person running it already believed.
• Measure returns as well as conversion. A clip that oversells drape or renders a color warm can lift conversion and returns at the same time. Net contribution is the number that matters, and nobody publishes it.
• Test placement, not only presence. A muted autoplay in the gallery, a tap-to-play thumbnail and a clip below the fold are three different things. A presence test treats them as one.
• Let it run. Give it weeks, not days, and long enough that one good Tuesday cannot swing the result.

Where the craft still matters

AI-generated motion fails in the details. Fabric swims, logos warp and skin turns plastic. A clip that misrepresents drape or color will lift your conversion rate and your return rate together. That is the worst outcome you can get, and a view-time metric will never show it to you.

This is where retouching experience applies directly. The AI model does the routine extension from the approved still. Human QC decides whether it ships. A retoucher checks frame by frame that the clip is true to the product, against the same standard a still has to meet.

What we will and will not tell you

We will not hand you a benchmark we have not earned. We have no randomized study of our own, and one drawn from our own clients would have exactly the sampling problem described above.

Motion on a PDP is worth testing. It is now cheap enough to test honestly, and the answer will depend on your catalog in ways no article can predict.

Want PDP clips made from your approved stills, with a retoucher checking every one? Talk to us.