eBay Promoted Listings Standard. Does it still work?

David Brackin, contributing editor to ChannelX and founder of StuffUSell takes a look at eBay Promoted Listings Standard seven years after he first reported on experiments to test their efficacy:

When Promoted Listings first launched widely, I ran an experiment to test whether it made a difference. It was randomised and blind, it covered 15,000 listings across several categories, and it found a 27% uplift in sell-through rate in eight weeks out of nine.

That number has been quoted back at me ever since. Seven years on, and many changes to the programme later, it seemed only fair to find out whether it still holds.

Why this is hard to measure

Three things make Promoted Listings unusually difficult to assess, and they are the reason most sellers have never done it properly. And yet we spend thousands of pounds on advertising.

The first is the attribution trap. eBay reports attributed sales and a return on ad spend, and both numbers come from the same system that sends you the bill. Since last year’s attribution change, a single click on a promoted item activates it for 30 days, and any sale of that item in the window is charged for, to any buyer, whether or not that buyer ever saw an ad. A listing that was going to sell three times this month sells three times this month, one person clicks an ad, and all three sales attract a fee. Almost every sale on a promoted listing gets counted. Reported ROAS therefore tells you what eBay charged for, not what the advertising added. You pay for what you would have sold anyway.

The second is data. eBay gives you impressions and clicks per listing, which is useful, but no holdout, no experiment tools, and a set of exports that take real work to join up. If you want to know what would have happened without the ads, you have to build the comparison yourself.

The third is the house message. eBay’s line, in every seller update and every account manager call, is that Promoted Listings works. The dashboard is designed to agree with it. That is not dishonest, but it does mean the burden of proof sits with the seller, and the only way to carry it is to hold some listings back and not advertise them.

So the design matters more than the result. Here is what we worried about and what we did about each worry.

The design

The market moves from week to week. Weather, payday, a competitor’s sale and eBay’s own promotions all shift sales, so a before-and-after comparison confounds the ads with everything else that changed. We ran promoted and unpromoted listings side by side over the same three weeks instead.

The listings you promote may be the ones that were going to sell anyway. If you pick the popular items, or the algorithm already favours some, you will see an uplift that has nothing to do with the ads. Every listing was sorted into a price band (under £25, £25 to £50, £50 to £100, over £100) crossed with whether it sat in the top or bottom half of the previous month’s impressions. Impressions are eBay’s own count of how often a listing surfaces in search, so this is matching directly on how desirable a listing already is. Within each of the eight cells, listings were allocated at random between the arms.

Random allocation can still produce unequal groups by chance, so we checked before starting. Baseline click-through rates across the three arms were within 0.03 percentage points of one another.

Is it the ads, or the rate? Two promoted arms, one at 2% (the campaign minimum) and one at 5% (the account’s existing rate), plus an unpromoted control. The arms held 263, 259 and 125 listings.

Relisting muddies the data. We relist stale listings, and campaign membership follows the SKU rather than the item number, so everything was tracked by SKU.

The account we ran this on belongs to a business I manage the eBay store for, which sells branded jewellery on consignment, around 650 listings, with a solid gross margin but not one that can be given away without thinking.

The effect

Three weeks, 24 August to 13 September 2026, 100 units sold.

Arm	Listings	Units per 100 listings	Revenue per listing, control = 100	Fee per listing, 2% arm = 100
2% ad rate	263	16.7	98	100
5% ad rate	259	13.9	82	253
Not promoted	125	16.0	100	0

item revenue per listing by arm, indexed to the unpromoted arm

The arm that spent nothing earned the most per listing. The arm that spent the most earned the least. The two promoted arms were attributed almost identical sales, for fees that differed by a factor of two and a half.

Measured against the control, the 2% arm sold 4.4% more units per listing and the 5% arm sold 15.1% fewer.

None of this reaches statistical significance. The tests give p values of 0.40 for 2% against 5%, 1.00 for 2% against control and 0.64 for 5% against control. In plain terms, the trial cannot show that the revenue generated in any arm was different from any other. This was a smaller trial than the first one. The 2019 study ran for 22 weeks over 15,000 listings. This one ran for three weeks over 650.

The cost and the benefit

Statistical significance is not the test a commercial decision has to pass. The question is whether the best estimate of the effect is above or below the level at which the ads pay for themselves, and that breakeven can be worked out in advance from your own margin with no statistics at all.

The rule is that breakeven incrementality is roughly the ad rate divided by your net margin per pound of revenue. A business keeping 70p in the pound needs about 7% of attributed sales to be sales that would not otherwise have happened before a 5% campaign washes its face. At 30p in the pound it needs 17%. At 20p it needs 25%.

required incrementality against margin, for 2% and 5% ad rates, with the measured effect shown dashed

Our measured effect at 2% sits below the breakeven line at any margin under about 45p in the pound. At 5% it is below zero, so there is no margin at which it pays. A seller with a fat margin on own-brand stock might clear the 2% bar. A consignment or thin-margin operation will not. The same ad rate is a bargain for one business and a slow leak for another, and you can tell which you are before you spend anything.

What changed since 2019

The trial cannot separate the candidates, so here is my weighting.

Everyone is doing it now. This is the one I believe most. A premium slot is only premium while your competitors are not in it. Once most of a category is promoted, the advantage is competed away and the rent goes to the platform. You are paying to stand still. The 2019 result may have been a true finding about a market that no longer exists.

Category. My 2019 data showed a weak link between rate and uplift (an R² of 43%) and I said at the time it would vary by category. That trial spanned several categories. This one is a single deal in one of them.

Placement. The top slot in search is now reserved for Advanced, the pay-per-click product, so Standard campaigns bid for what is left. That would explain why bidding 5% bought so little over bidding 2%.

What we did, and what you should do

We ended both campaigns once I had these results, as I could not justify spending more of the store’s money to refine the numbers.

If you have been running Promoted Listings since 2019 on the strength of results like mine, this is your nudge to re-run the test on your own stock. Three weeks. A held-back control, matched on last month’s impressions. Judge on units per listing against the control, not on reported ROAS. Then put the effect you measure against your own breakeven. It took a few days of work to set up and run properly, and if you would rather someone did it for you, you know where I am.

I stand by the 27% figure for that inventory in that year at that level of adoption. I no longer think it holds here. It might be that your products in your category with your margin mean that Promoted Listings makes money for you. Unless you have tested it properly, it is very hard to know for sure. If you run it and get the opposite answer, I would be glad to hear it. That would be the most interesting result of all.