Amazon Q2's AI advertising claims don't line up

Amazon gave buyers one clean number in Q2: advertising services revenue was $19.81 billion, up from $15.69 billion a year earlier. That is the part of the Q2 story that can be tied to the income statement. In the same release, Amazon also said advertisers using Ads Agent saw an 8% lower CPM and a 6% lower CPA. Those are not the same kind of claim, and they should not be budgeted the same way. [1]

The problem is not that Amazon Ads is weak. It is the opposite: Amazon’s ad business is now commercially large enough that a performance claim placed near the revenue line can borrow credibility from it. The revenue line is checkable. The Ads Agent lift claim is vendor-reported, and the release does not publish the sample size, spend level, account type, measurement period, or attribution method behind the 8% and 6% figures. [1]

There is also a second Amazon-controlled surface with a different magnitude. Amazon’s Ads Agent product page cites a 2025 U.S. beta in which 65% of beta advertisers saw delivery improvements, with average CPM reductions of 18% and CPA reductions of 16%. [2]

Amazon does not present those two sets of figures as a contradiction. The comparison is the buyer’s problem: one Amazon source says 8% lower CPM and 6% lower CPA in the Q2 release; another says 18% lower CPM and 16% lower CPA on the Ads Agent page. Until the denominator and method are visible, the safe reading is that these are separate vendor claims, not a transferable forecast for next month’s media plan.

Put each number in the right bucket

That separation matters because the phrase “AI advertising impact” can collapse three different questions into one. Did Amazon’s ad business grow? Yes. Did Amazon sell more AI-related infrastructure and services through AWS? The AWS line is a better place to look for that than the advertising-services line. Did Ads Agent reduce CPM or CPA in a way a buyer can underwrite before testing? That is still unproven from the public materials. [1]

The Ads Agent claim is missing the parts finance will ask for

“Advertisers using Ads Agent saw lower CPM and CPA” sounds operational. It is not operational enough. A buyer cannot translate that into a budget assumption without knowing which advertisers were included, whether the accounts were already managed by Amazon teams, how much they spent, whether DSP activity was required, what period was measured, and whether attribution changed during the test.

The missing ACoS number is especially noticeable for Amazon sellers. CPM and CPA can improve while ACoS gets worse if the campaign mix, conversion value, retail readiness, or attribution window changes. Feedvisor’s review of Ads Agent notes that Amazon published no ACoS lift figure and says the beta participants were managed, DSP-active advertisers operating at scale. That makes the beta informative, but not representative of a typical seller deciding whether to hand campaign operations to the agent. [4]

The difference between the Q2 release and the product-page beta claim also changes the adoption decision. An 8% CPM reduction and a 6% CPA reduction might justify a controlled test. An 18% CPM reduction and a 16% CPA reduction could shift a roadmap, staffing plan, or managed-service conversation. The public record does not yet tell buyers which figure is closer to the result they should expect in their account.

For a broader tool-by-tool read, the existing Amazon AI marketing tools audit is the better place to compare Ads Agent against other Amazon AI surfaces. Here, the issue is narrower: the Q2 earnings release does not publish enough methodology to turn the Ads Agent lift line into an expected outcome.

Amazon’s own growth story points somewhere else

Andy Jassy’s account of Amazon Ads growth is not centered on Ads Agent. He points to Sponsored Products, conversational shopping in Alexa+ and Alexa for Shopping, Prime Video, on-site placements, and sold-out sports inventory around Thursday Night Football, NBA, WNBA, and NASCAR. Those are plausible growth engines. They are also mostly inventory, placement, demand-capture, and shopping-interface stories, not proof that an AI campaign agent caused the advertising-services increase. [3]

That is the stronger argument for Amazon Ads, and it is also the reason the AI-tool conclusion should be narrower. Sponsored Products can keep scaling because advertisers still want high-intent search and product-detail-page demand. Prime Video and sports can add premium supply that brand and upper-funnel budgets can buy. Alexa+ can create new conversational shopping moments. None of that requires treating Ads Agent’s public lift figures as the cause of the 26% advertising-services growth.

This is not a semantic objection. If the growth came from more inventory, more sponsored product demand, sports sellout, or calendar timing, then the operating decision is about channel mix and auction pressure. If the growth came from Ads Agent improving efficiency, the operating decision is about automation adoption. Those lead to different tests, different budget controls, and different conversations with finance.

AWS is the cleaner AI revenue line

Amazon reported AWS revenue of $42.2 billion, up 37%. That is the cleaner place to discuss booked AI demand because cloud infrastructure, model access, and AI services show up closer to AWS revenue than to the advertising-services line. It still does not answer whether Ads Agent improves campaign economics. It answers a different question: whether Amazon’s cloud business is growing while enterprise AI demand is running through cloud platforms. [1]

Even the company-level profit number needs sorting. Amazon’s net income was reported at $62.6 billion, but that figure was inflated by a $53.4 billion Anthropic-related non-operating gain. That can matter to investors reading the quarter. It should not be dragged into an argument about whether an Amazon Ads automation product lowered CPA in a buyer’s account. [5]

The same discipline applies to stack risk. Ads Agent running through Amazon’s broader AI stack is relevant to tool reliability, but it is not evidence of performance lift. The site’s Nova deprecation note is useful context on exposure to model and infrastructure changes, especially because Ads Agent’s stack position is a separate question from whether its public CPM and CPA claims are transferable.

Calendar context belongs in the margin, not the center

Prime Day shifted into June 2026, putting part of that event inside Q2, and Amazon’s Q3 guidance included an approximately 400-basis-point headwind tied to that shift. That is worth noting because it affects how cleanly the 26% advertising-services growth can be compared across quarters. It does not solve the Ads Agent question either way. [1]

A full normalization analysis would ask how much of the ad-revenue increase came from event timing, retail demand, Prime Video supply, sports inventory, Sponsored Products, and pricing. The public Ads Agent figures still need their own proof. Calendar movement can explain part of revenue growth; it cannot validate a CPA claim.

The adoption standard for buyers

The practical standard is simple: do not reject Ads Agent because the public claims are incomplete, and do not adopt it because the advertising-services line grew. Treat Ads Agent and Full-Funnel Campaigns as tools that need account-level validation before they become planning assumptions.

Before moving meaningful budget, buyers should ask for the sample, spend level, account type, period, attribution method, campaign mix, and primary metric behind any lift claim. If the claim is based on managed, DSP-active advertisers at scale, that should be stated before applying it to a seller running Sponsored Products with a different margin structure. If the metric is CPA, the test still needs ACoS, ROAS, conversion value, and delivery quality checks.

The pattern is not unique to Amazon. Platform AI claims often arrive as blended lift statements before the named-account evidence catches up. The parallel in Alphabet’s AI strategy for Google Ads is useful for the same reason: platform-reported lift can be directionally interesting and still fail the test for budgeting until the account type and method are clear.

Amazon’s ad business grew, and that matters. The Q2 release does not prove Ads Agent caused that growth, and the public Ads Agent lift figures are not ready to be treated as expected outcomes. Use them as claims to test, not as numbers to put into the forecast.

Last reviewed: Aug. 1, 2026, against Amazon’s July 30, 2026 Q2 earnings release. This tracker should be rechecked when Amazon reports Q3 results in late October 2026, because the Prime Day calendar shift, Q3 ad growth, and any updated Ads Agent methodology could change the verification picture. [1]

References

1. Amazon.com Announces Second Quarter Results — Amazon Investor Relations, July 30, 2026
2. Ads Agent — Amazon Ads
3. Amazon CEO Andy Jassy on Amazon Ads growth in Q2 2026 earnings — About Amazon
4. What Is Amazon Ads Agent? — Feedvisor
5. Amazon Advertising Up 26% In Q2, Profits Soar To $62.6B On Anthropic Gain — MediaPost