Questions we could not answer, so we measured.
Every study here starts with an argument we kept having with a client and could not settle from opinion. We run the numbers on the accounts we manage, publish the method, and publish the parts where the answer was no.
Since September, Sponsored Products in Europe also run on other websites and apps. In every European account with enough clicks, Off Amazon returned less than the same ads on Amazon, a median 0.63 times. One campaign setting took it to zero.

We followed 106 Best Deals from five weeks before the event to three weeks after, against the same brands' products without a deal. The typical deal sold 3.5 times its expected units, we found no give-back in the weeks around it, and deeper discounts went with bigger lifts.

We looked for every time the ads stopped, were cut or ran out of budget in the accounts we manage. When they stayed off for days, total sales fell by about as much as Amazon credited to the ads. When a budget ran out mid-day, a third to a half was lost that evening.

We split 32 weeks of ad orders by keyword and target with Amazon Marketing Cloud. Among search keywords, the brand's own name brings the lowest share of new buyers, and still the cheapest new customers by far. Other brands' names bring the most new buyers, at about the price of generic keywords.

We matched every prompt click with the same product's ad clicks in the same week, across 16 brands. “Why choose [brand]?” prompts cost three cents a click and sold at least as well as the ads. Other questions brought half the orders of an ad click, statements a fifth.
![Prompt clicks by price per click: a spike of Why choose [brand] clicks at exactly 3 cents, the rest spread between 1 and 4 dollars](/research/amazon-sponsored-prompts/card.webp)
We ran both rank trackers on the same keywords for two weeks, plus DataDive on one brand, and re-ran it in November on six more weeks. They agree on who ranks, on the top 10 and on trends. Further down the exact positions differ, even when they check at the same moment, and the differences go both ways.

Daily bids, placement modifiers and realized CPC per placement on exact-match branded campaigns. Top of Search costs half the product-page click because brands pay 11% of their bid there and 88% on product pages. The modifier buys share, not price.

We matched daily organic rank against Search Query Performance across six US consumer brands, week by week. Impression share identifies the rank band reliably. It cannot see week-to-week movement.

Including the parts that did not work.
A study that only reports what confirmed the hypothesis is marketing. The SQP paper's most useful finding is a failure: subtracting ad impressions to clean the signal made the proxy measurably worse. The branded CPC paper spends a full section on the days where its own effective-bid math breaks.
Got a question worth measuring?
Most of these studies started as a client asking something we could not answer honestly. If you have one, send it over. It might become the next paper.