Jeff Brokaw · Chief AI and Marketing Officer · Published · Updated
Your January marketing report might show that Amazon impressions fell off a cliff.
Before your AI dashboard tells you to cut the campaign, check what Amazon changed.
Amazon expects an average decline of about 63% in reported impressions for affected Display, Video and Audio (DVA) display placements on the Amazon Store when it changes the way those impressions are counted on January 1, 2027, according to a notice to advertisers reviewed by Digiday. The change moves the count from when an ad starts downloading to when it appears on a shopper's screen. It affects display placements on Amazon's own store, including the homepage, product detail pages, search results and order pages. That is an average across affected placements, not a forecast for an individual account. PPC Land also quoted the notice; I have not independently reviewed the original advertiser notice.
Amazon also says advertisers in its testing saw ROAS rise and CPA fall, according to PPC Land. The notice as quoted gives no test size, period or methodology. A lower impression count alone does not explain either result.
The bigger problem is what happens when your dashboard compares December and January as if the definition never changed. An AI system can write a very confident explanation for that chart. It won't necessarily know that the ruler changed.
What is changing in Amazon's impression reporting?
For affected DVA display placements on the Amazon Store, the reported notice says Amazon plans to count an impression when the ad appears on screen rather than when it begins downloading. Digiday reports that Amazon expects the new method to produce roughly 63% fewer reported impressions on average. ET BrandEquity's reporting says Amazon reports viewable impressions separately and told the outlet that viewability and attribution remain unchanged.
Don't apply that 63% to every Amazon ad product or every campaign. The reports describe affected display placements on Amazon's store, not a blanket reduction across all sponsored ads, off-site inventory and other channels. Sponsored Products and Sponsored Brands are not named in the notice as reported by PPC Land. The quoted notice does not say whether billing changes. Ask your account team which definition and billing unit apply to your placements and reports.
One more distinction matters: an ad appearing on screen is not proof that a person noticed it. The reported counting change and separate viewable-impression measurements answer different questions. For example, Amazon defines viewable impressions in certain Display and Sponsored Brands reports using specific visibility and duration thresholds. Don't assume those thresholds describe the reported 2027 count.
Your CPM can jump even if the bill doesn't
CPM is spend per thousand reported impressions. If spend holds steady and the number of counted impressions falls, CPM rises. You don't need a more expensive auction for the math to change.
Here's an illustration, not an Amazon campaign result:
| Before the counting change | After, assuming a 63% reduction | |
|---|---|---|
| Spend | $10,000 | $10,000 |
| Reported impressions | 1,000,000 | 370,000 |
| Spend per 1,000 reported impressions | $10.00 | $27.03 |
The second calculated CPM is roughly 2.7 times the first because the denominator changed. This example assumes identical spend and uses the reported average decline for simple arithmetic. It does not predict your campaign's actual impression count, billing, billed CPM, auction price or performance.
Click-through rate can also change when its impression denominator changes. If a report still records 10,000 clicks, the illustrated rate goes from 1% to about 2.7%. That doesn't prove the creative suddenly got better. It means the rate needs to be read alongside its definition and the actual click data.
This is the part I want in the client conversation before the first January report goes out. A new number without the context is an invitation to make the wrong budget decision.
Amazon says ROAS went up. Show me the test.
The notice quoted by PPC Land makes another claim: Amazon says advertisers in testing saw return on ad spend rise and cost per action fall. The quoted notice doesn't disclose the number of advertisers, test period or methodology.
That's a result I'd want to investigate, not a promise I'd put in a client forecast. Amazon supplies the platform and is reporting the test. Show me the comparison, the spend and conversion data, and whether campaigns changed alongside the count.
A lower impression count alone doesn't make ROAS go up or CPA go down. ROAS compares attributed sales with spend; CPA compares spend with actions. If those inputs stay the same, those ratios stay the same. January Digital's Jared Smith told Digiday that the reporting shift shouldn't move ROAS or CPA. Amazon's reported test and that mathematical point are different claims. We don't have enough detail to explain the test result.
Google changes publisher-side counting in February
On February 17, 2027, Google says Ad Manager and AdSense publisher-side impression measurement will move banner display inventory on desktop web, mobile web and connected TV from count-on-download to begin-to-render. Google's definition counts when the creative has loaded and started rendering. Its documentation says native, app and video inventory already uses or complies with begin-to-render.
Amazon's reported on-screen count and Google's begin-to-render count are separate changes with different scopes. Don't apply Amazon's 63% estimate to Google. Google gives eligible Ad Manager publishers a way to compare download and begin-to-render metrics in historical reports. If you have access, use your own data to estimate the difference.
If your reporting includes affected Amazon figures and Google publisher-side banner metrics, annotate each change where it applies. Do not assume every advertiser dashboard adopts Google’s publisher-side definition on February 17. For buyer-side tools, Google separately estimates February 2027 for its Display & Video 360 and Campaign Manager 360 billed display-impression change; that notice does not set February 17 for those tools. Check each report before comparing channels or handing the data to an AI summary.
The AI report is only as good as the instructions and data behind it
In the AI go-to-market system I built, a salesperson reviews and sends prospect emails; an interested reply is the gate for creating or enriching associated CRM records. An open or click is tracked separately from buyer interest. The public demo uses fictional records and illustrative figures. That system was not built for this Amazon change. It illustrates the decision I care about here: keep the event, its meaning and the action it permits distinct.
A reporting assistant might see impressions down, CPM up and a target missed. If the report is configured to explain variances against December, it may describe the change as deteriorating delivery. It might even recommend pausing the placement.
Give the system a measurement-change record alongside the campaign data:
- What changed: The counting method described in the advertiser notice.
- Where it applies: The specific account, ad products and placements you've confirmed are affected.
- When it starts: January 1, 2027, as reported; confirm the effective date for your account.
- What it affects: The impression count and ratios that use it, including CPM and click-through rate.
- What remains to verify: Spend, auction and billing behavior, clicks, attributed sales, placement mix and any other changes in the account.
- What happens next: A person reviews the explanation before it reaches a client or triggers a bid or budget change.
Keep the link to the original notice with the record. Give the reviewer access to the underlying report, not just the AI's summary. If the notice says something different for your account, use that evidence and update the record.
This is the same reason I separate what different systems say about a customer in Why Your CRM, GA4 and Ad Platforms Disagree. A number is only useful when you know what it counts and which decision it can support.
Build a bridge between the old report and the new one
First, save the current definition of the metric and the affected report configuration. Record the campaign, placement, dates, filters and attribution settings. If Amazon provides a comparable viewable-impression measure for your account, pull it alongside the existing count and check its definition and availability. Don't relabel it as the new count unless Amazon confirms they are equivalent.
Second, mark January 1 on every chart and exported dataset that combines the old and new definitions. An annotation on a dashboard helps the person reading it. The AI generating commentary also needs the change in its source context or instructions; a note buried in a slide may never reach it.
Third, compare like with like as far as the data allows. Keep a pre-change baseline, then build a new baseline from post-change results. Review the affected placements separately from unaffected ones. When seasonality, budgets, targeting, creative or auction conditions change at the same time, record those changes too. A simple before-and-after comparison can't isolate the measurement change from everything else happening in the account.
Finally, talk about the business outcome. Spend, clicks, attributed orders and sales can add context, but their definitions and attribution windows matter too. Amazon's 2026 view-attribution update describes a separate change for Sponsored Brands, Sponsored Display campaigns billed on a viewable-impression basis, and Amazon DSP campaigns serving Store ads. It is a separate 2026 change, not a new 2027 attribution change. Check which attribution rules apply to the campaigns and dates in your report. Don't quietly treat a change in attributed sales as proof of incremental sales.
What should an agency tell a client?
Tell them before the report lands. I'd put the explanation next to the metric and then have the buyer walk through it:
According to reports of an Amazon advertiser notice, the platform is changing how it counts impressions for DVA display placements on the Amazon Store starting January 1, 2027. Amazon expects the reported count to fall on average, though our account may differ. We are separating the affected placements and establishing a new baseline. We'll review spend, clicks and sales alongside the revised impression figures before recommending budget changes.
Ask Amazon or the account representative which unit bills your campaign, whether billing changes, and how the affected report defines an impression. Record the answer. Don't tell the client that nothing changed in delivery until you've checked.
A good agency earns its fee by explaining what the numbers can and cannot support. If a client pays you to manage the budget, “the AI said CPM was up” is a pretty lousy answer for moving it.
Run a one-week test before the change
You don't have to wait until January to see whether your reporting workflow handles a definition change.
Take one existing client report. Keep the original data unchanged. In a separate test copy, label a set of figures as illustrative and change the impression count while holding spend and clicks constant. Add a short measurement-change note with a link to the reporting source. Ask your normal reporting workflow to explain the variance, then compare what it produces with and without that note.
Have a human reviewer check four things:
- Does the report identify the definition change and the affected placements?
- Does it explain the CPM and click-through-rate math without claiming the auction or creative caused the change?
- Does it identify what needs checking, including spend, placement mix and sales attribution?
- Does it hold the budget recommendation for review when the data cannot support a causal conclusion?
Record the time spent preparing and correcting each version, plus any missing or misleading claims. Use the results to fix the reporting workflow before January.
Which tools can do this?
Export placement-level data where available from Amazon Ads reporting, keep the old and new definitions in a Google Sheet or data warehouse, and connect the data to your reporting tool. If you use n8n, its human approval step can pause selected AI tool actions; choose your own approval point and check plan and access requirements.
My reporting article covers the decision brief, and The Human Gate That Isn't a Rubber Stamp covers the review that has to happen before an automated recommendation becomes an action.
Pull up the report your team plans to send in January. The counting rules are scheduled to change, so find every chart, alert and AI summary that compares impressions or CPM across those dates. Put the definition next to the number and give the reviewer the source before anyone uses it to move money.
Does an impression appearing on screen mean someone saw the ad?
No. The reported new Amazon count concerns whether an affected ad appears on screen. It does not establish that a shopper paid attention. Viewable-impression metrics use their own definitions and measurement methods, so check the specific report and accreditation before comparing them.
Should advertisers change their Amazon budgets because of the new count?
Not on the impression-count change alone. Check the scope in your own account, then review spend, placements, clicks, sales attribution and any changes to campaign settings. A budget change may be warranted by the full evidence, but a revised counting method is not enough on its own to explain the business result.
Sources
- Digiday: Buyers grapple with incoming Amazon reporting changes, November 16, 2026. Reports the advertiser notice, affected store placements and Amazon's average estimate. The notice was not independently accessed for this article.
- PPC Land: Amazon drops off-screen ads from impression counts, November 3, 2026. Quotes the notice and reports its testing claim; the full advertiser notice was not independently inspected.
- ET BrandEquity: Amazon's ad metric reset, November 17, 2026. Reports scope and Amazon's statement about separate viewable-impression metrics.
- Google AdSense: Begin-to-render impression counting. First-party details on the February 17 change and covered inventory.
- Amazon Ads: Campaign performance metrics. Defines viewability for specified Display and Sponsored Brands metrics; it does not define the reported new DVA count.
- Amazon Ads: View attribution updates for Amazon Store ads, January 2026. Describes a separate attribution change and its scope.