Amazon Ads will stop counting display impressions on its own store at the moment an ad starts downloading to a shopper's device, and expects the change to leave impression counts on those placements about 63% lower on average. The new rule takes effect on January 1, 2027, according to a notice to advertisers that Saul Jacobs, chief marketing officer at Elevate33, posted on LinkedIn today.

The notice, addressed "Dear Advertiser" and carrying no date, covers display ads sold through Display, Video & Audio, or DVA, the name that replaced Amazon DSP at the end of September. It applies to placements on the Amazon Store, from the homepage to order pages. After the switch, an impression is recorded only once the ad appears on the shopper's screen.

Few measurement changes in digital advertising carry a number this large. Google, which is moving its own buying tools to a stricter count in February 2027, expects only a slight decrease in billed impressions for Display & Video 360 and Campaign Manager 360. Amazon's estimate implies that, on its own figures, most of the display impressions currently counted on these placements never reach a screen.

In Short

Right now Amazon counts a display ad as delivered as soon as it starts loading on a shopper's phone or computer, even if the shopper never scrolls far enough to see it. From January 1, 2027, display ads on Amazon's own store will count only once they actually show up on screen, and Amazon says that will cut the number of counted impressions by about 63% on average. Reports for these ads will show far fewer impressions from that date, and the notice does not say whether what advertisers pay changes with them.

What the notice says

Amazon's notice opens with the date and the mechanism. "On January 1, 2027, we are changing how we count impressions for Display, Video & Audio (DVA) display ads on the Amazon Store," it reads. "Today, we count an impression when the ad starts to download to the shopper's device. This is known as count-on-download (COD). Starting January 1, 2027, we will count an impression once the ad appears on the shopper's screen."

The scope is defined by page type. "This change applies to DVA display ads placements on the Amazon Store, including the homepage, product detail pages, search results pages, and order pages," according to the notice.

Under a heading that reads "What changes in your reports from January 1, 2027," Amazon lists two consequences. The first is the volume effect: "Impression counts on these placements will be about 63% lower on average compared to the counting methodology used prior to January 1, 2027, but the impact on your specific campaigns will vary based on the placements you target: placements lower on the page will see a larger decrease in impression count."

The second is a performance claim. "In testing, advertisers saw improved performance with return on ad spend (ROAS) increasing and cost-per-action (CPA) decreasing, though the result will depend on your specific campaigns," the notice states.

The text Jacobs shared ends with those two points. It gives no figure for the size of the test, the number of advertisers involved or the period it covered. Nor does it define what "appears" means in pixels or seconds, say whether invoices are calculated on the new count, or mention DVA video and audio formats, DVA campaigns running on third-party sites, or sponsored ads.

From download to screen

Three different moments can trigger an impression in display advertising, and the gap between them is where Amazon's 63% sits.

The earliest is the one Amazon is leaving behind. Under count-on-download, an impression is recorded when the creative starts arriving on the device, triggered by a client-side request. The Interactive Advertising Bureau (IAB) and the Media Rating Council required that kind of client-initiated counting in guidelines issued in September 2004, ruling out counts made purely on the server's decision to send an ad.

The industry minimum then moved. In December 2016, the IAB, the MRC and the Mobile Marketing Association proposed shifting mobile web and in-app measurement from a count-on-download minimum to begin-to-render, and desktop display from a count-on-ad-insertion minimum to the same model. Public comment closed on February 1, 2017, and final desktop, mobile web and in-app guidelines followed on October 31, 2017. After a one-year grace period, compliance became a condition of MRC accreditation. Under that rule, the creative must have been received and must have started to draw on the page or in the app. It does not require any part of the ad to be visible.

Viewability is the third and strictest bar. The MRC's 2014 standard for display requires 50% of an ad's pixels to be on screen for one continuous second.

Amazon's wording - "once the ad appears on the shopper's screen" - goes past begin-to-render on its face, because begin-to-render is satisfied by an ad that renders below the fold and is never scrolled into view. It resembles the approach Google's Campaign Manager 360 and Display & Video 360 take in apps where Open Measurement is present, counting an impression when at least one pixel is on screen. The notice does not say whether one pixel is enough, or whether a share of the ad or a minimum duration is required.

The detail that "placements lower on the page will see a larger decrease" fits that reading. An ad slot on a product detail page that loads with the page, but sits several screens below the buy box, is counted today even if the shopper adds to cart and leaves. Web publishers use lazy loading to defer fetching ad slots until they approach the viewport, which narrows the gap between a download and a display. The size of Amazon's estimate suggests that much of the creative on its own store pages has been fetched well before a shopper could see it.

Jacobs put it bluntly. "I'm betting most advertisers weren't aware of the current impression methodology onsite, it's definitely a departure from IAB standards," the post reads.

Amazon is not alone in still counting display on download, however. Google's AdSense and Ad Manager will keep counting banner display impressions on download until February 17, 2027, when they move to begin-to-render for web, mobile web and connected TV banners.

The arithmetic of 63%

A 63% fall leaves 37 counted impressions for every 100 counted today. On PPC Land's calculation, a campaign whose spend and clicks stayed exactly the same would report a cost per thousand impressions 2.7 times higher than before: a $5.00 CPM would become $13.51. Click-through rate would rise by the same factor, from 0.10% to 0.27% in the same example. Neither change would reflect any difference in what shoppers saw or did.

Jacobs expects exactly that pattern. "I suppose the reduction in impression volume will make every onsite metric look even better. I'd expect ROAS and CPA to stay steady and CPMs to skyrocket," the post states.

Amazon's notice claims more than steady ROAS. Its testing, it says, produced higher ROAS and lower CPA. That is harder to explain as a reporting change alone. ROAS divides attributed sales by spend, and relabeling the impression column moves neither number. If fewer impressions count, there are also fewer impressions available to receive view-through conversion credit, which on its own would push attributed sales down, not up.

So for ROAS to rise and CPA to fall in Amazon's tests, spend must have dropped relative to the sales attributed to it. Two mechanisms could produce that: billing on the new, smaller count at bid levels that did not rise to compensate, or campaign optimization that moved budget toward placements shoppers actually see. Both could be at work. The notice does not say which applied, or whether the test conditions match what advertisers will face from January.

The billing question is the one that matters most for budgets. The automated DVA+ tier that Amazon introduced in September is billed on a cost-per-thousand basis with no minimum spend. If the thousand in that CPM becomes a thousand on-screen impressions from January 1, each billed unit is a different product than it was in December. If billing stays on the old count while reports move to the new one, invoices and reports will no longer reconcile. The notice frames the change under "your reports" and does not address invoices at all.

Year-over-year comparisons will break either way. January 2027 reports for Store display placements will show roughly 63% fewer impressions than January 2026 even if delivery were identical.

What the comment thread raised

The post drew replies from agency and ad sales executives, several of them focused on what the notice leaves out.

Mansour Norouzi, partner and director of advertising at Incrementum Digital, pointed to the narrow scope. "I wish they changed it for all ad types. Not just DSP," Norouzi wrote. Jacobs replied: "great call out!" The notice names only DVA display ads. Sponsored Products and Sponsored Brands, which make up Amazon's search advertising, are not mentioned.

David Buckley, a sales executive whose LinkedIn headline lists ad tech, CTV and commerce media, raised attribution. "Theoretically this would mean that any view-through conversion was essentially inflated by a larger margin," Buckley wrote. Jacobs answered that Amazon would likely protect its auction: "you have to imagine they have a plan to protect the auction dynamics but even if this is just a reporting change eCPM is going up."

Buckley's point connects to a change Amazon made twelve months earlier. On January 1, 2026, Amazon moved view-based attribution for Sponsored Brands, Sponsored Display campaigns billed on viewable CPM and Amazon DSP campaigns serving Store inventory to a shopping-signal enhanced last-touch model with shorter windows, keeping the old 14-day logic available as "all views" metrics. The new impression notice does not say how view-based attribution will treat impressions under the on-screen rule.

Iqra Shakoor, co-founder of Tacosology, separated measurement from delivery. "A 63% drop in reported impressions doesn't necessarily mean less ad delivery. It means the measurement is becoming stricter," Shakoor wrote. In a follow-up reply to Jacobs, Shakoor set out the billing scenario: "If CPM billing uses the stricter count, advertisers could effectively pay more per reported impression even if total spend stays the same."

Seth Meyerowitz, an enterprise sales director whose headline lists earlier roles at Amazon and Google, questioned why the old method existed at all. "hmmm... seems like ad impressions should have ALWAYS counted after the ad loads. but that's just me," Meyerowitz wrote. Jacobs agreed: "in all of my programmatic experience that is the standard. I guess maybe this is a nuance of the Amazon app downloading the ad to a device. Either way it should have been addressed sooner."

That last explanation is speculation in the thread, not something the notice states.

Seven weeks ahead of Google

Amazon's January 1 date lands 47 days before Google's own display switch on February 17, 2027. Google announced its change by email to AdSense publishers on September 1, 2026, with a companion help article extending it to Ad Manager. Native, interstitial, app and video inventory were already compliant with begin-to-render under Google's description, so the change targets banner display on web, mobile web and connected TV.

On the buying side, Google's Display & Video 360 began moving display counting from on-download to begin-to-render in September 2025, and Campaign Manager 360 followed on November 6, 2025. A help page updated on August 21, 2026 then scheduled the billed unit in both tools to move to begin-to-render in February 2027, citing alignment with MRC standards to support continued accreditation. Google gave no figure beyond a slight decrease in billed volume.

The contrast with Amazon's 63% is stark, though the two changes are not the same. Google is moving to begin-to-render, which does not require visibility. Amazon's notice describes counting on appearance on screen, a stricter trigger that would remove every below-the-fold ad that renders but is never seen. That difference in definition may account for much of the gap between "slight" and 63%.

There is a further difference in who carries the change. Google's sell-side switch affects what publishers report as delivered on their own sites. Amazon is both publisher and seller of the Store inventory in question, and it is changing the count on its own pages.

Why it matters for Amazon's advertisers

The timing puts the change on top of a platform that is already in flux. Amazon retired the Amazon DSP name at its unBoxed event on September 29, 2026, folding the platform into Amazon Ads Agent and naming the display, video and audio product DVA. The new DVA+ campaign type, which brings together what was previously bought through Amazon DSP, Sponsored Display and Sponsored TV, entered a US closed beta with general availability targeted for 35 markets in Q1 2027. Advertisers moving budgets into DVA+ in early 2027 will be doing so under a new impression unit as well as a new interface.

The pricing of Amazon's ad inventory is also under legal scrutiny. On August 31, 2026, the Federal Trade Commission and 22 states sued Amazon, alleging an undisclosed soft reserve applied on top of its second-price auctions for Sponsored Products, Sponsored Brands and Display Ads. The complaint says Sponsored Products advertisers paid their own full bid on 79.1% of charged clicks in 2024. Amazon denies overcharging and says no advertiser pays more than its bid. A change to the unit behind display CPMs, announced without billing detail, arrives while the price of the click is in court.

Independent verification on Amazon's own pages has historically been limited, too. When Amazon DSP added Moat's pre-bid viewability targeting in January 2022, the feature did not work on Amazon owned-and-operated supply, only on third-party inventory.

The commercial stakes are large. Amazon reported advertising revenue of $19.64 billion for the third quarter of 2026, up 10.9% year over year, or 12% excluding currency effects. Amazon does not break out how much of that comes from display placements on its store, and the notice gives no indication of any expected revenue effect.

What remains unknown is the part advertisers will see on invoices. Will a thousand on-screen impressions cost the same as a thousand downloads did? Will bids, floors and auction reserves be recalibrated, as Jacobs assumes? Does the change reach DVA campaigns on third-party sites, or sponsored ads, at a later date? The notice answers none of these.

Timeline

Summary

Who: Amazon Ads, which sent the notice to advertisers, and Saul Jacobs, chief marketing officer at Elevate33, who posted it on LinkedIn. Advertisers running DVA display ads on the Amazon Store are affected, and commenters including Mansour Norouzi, David Buckley, Iqra Shakoor and Seth Meyerowitz raised questions about scope, attribution and billing.

What: Amazon will stop counting DVA display impressions on its store when the ad starts downloading (count-on-download) and will count them only once the ad appears on the shopper's screen. Amazon expects impression counts on these placements to fall by about 63% on average, with bigger drops lower on the page, and says advertisers in testing saw higher ROAS and lower CPA. The notice does not say whether billing changes.

When: The change takes effect on January 1, 2027. The undated notice became public through the LinkedIn post on November 3, 2026, about seven weeks before Google moves AdSense and Ad Manager display counting to begin-to-render on February 17, 2027.

Where: DVA display placements on the Amazon Store, including the homepage, product detail pages, search results pages and order pages. The notice does not mention DVA video or audio, third-party inventory or sponsored ads.

Why: Amazon's notice gives no reason for the change. The old method sits below the begin-to-render minimum the IAB and MRC set in 2017, and the scale of the expected drop suggests most display impressions counted on Amazon's store pages today are never seen on screen. For advertisers, the change resets CPM, click-through rate and year-over-year benchmarks, and leaves open whether the price of an impression on Amazon's own pages changes with the count.