Amazon Ads Agent: Full-Funnel Buying Meets Auto-Creative

At unBoxed 2026 on September 29, Amazon Ads announced that the unified platform it built by merging its demand-side platform and ads console now has a name — Amazon Ads Agent — and that agentic AI sits underneath all of it. Alongside the rename came something more concrete: Full-Funnel Campaigns are now generally available to all advertisers in the United States, and a new programmatic format, DVA+, starts rolling out in late October.

The rename is the least interesting part. The interesting part is that Amazon has now put campaign planning, optimisation and creative generation into the same flow — the same direction Google took with vertical video buying in DV360 a day earlier. Two of the largest ad platforms spent the last week of September automating the buying layer and baking asset generation into it. That puts the remaining manual work in a very specific place.

What actually shipped

Amazon Ads Agent brings three campaign types into one platform: sponsored ads, Display, Video and Audio, and Full-Funnel Campaigns. Inside it, advertisers can build manually, hand optimisation to Amazon's AI, or mix the two.

Full-Funnel Campaigns — previewed at last year's unBoxed, now available to all US advertisers — put sponsored ads, display, video and streaming TV into a single campaign. You supply the budget, products and creative; Amazon's AI plans, executes and continuously optimises. Signals flow across formats, so upper-funnel investment is fed into lower-funnel performance rather than sitting in a separate report. Reporting is unified, and it carries new-to-brand and long-term sales metrics next to the familiar impressions, clicks and sales.

DVA+ collapses Sponsored Display, Sponsored TV and programmatic campaigns into one buying experience, powered by Brand+ for audience identification and Performance+ for outcome optimisation. Amazon places the ads across streaming TV, online video, display and audio on its own properties and the open internet. Experienced programmatic buyers keep advanced settings — targeting, deals, supply sources, frequency caps, bidding.

The conversational layer is the connective tissue: plain-language media planning with audience sizing, reach forecasting and budget allocation recommendations; AI targeting; natural-language analytics; and one-click recommendations for Sponsored Products. Amazon says these roll out over the coming months.

The number Amazon put next to it

Amazon reports that advertisers who used natural language to apply campaign targeting recommendations saw more than 25% additional unique customers while lowering cost-per-impression by more than 10% on average over the past year. That is Amazon's own measurement of its own recommendation feature — directional, not a forecast.

The case studies published alongside it are worth reading the same way. A dog-treat brand, Canine Naturals, ran a full-funnel campaign for 206 days and attributed 41% of its sales to new-to-brand shoppers. Cuisinart, with agency Global Overview, piloted Amazon's generative creative tooling on food-processor products in Q2 2026 and reported an 18-point increase in detail-page view rate, a 14% reduction in cost per detail-page view and a 10-point lift in branded search rate against traditionally produced assets. Two brands, two formats, vendor-reported — useful as a signal about direction, useless as a benchmark for your account.

Where the constraint moves now

Read the Amazon announcement next to the DV360 one and a pattern is hard to miss: the platforms are quietly taking over media planning and asset assembly at the same time.

Full-Funnel Campaigns will generate "TV-quality video and display images from existing product pages," with advertisers approving creatives before launch. That is the same URL-and-product-data-to-ad path Prizmad runs, now embedded in a retail media console. When that capability is available inside the platform that also does the buying, the marginal creative asset stops being the scarce thing. Anyone can produce volume. What stops being free is a distinct, defensible reason for a shopper to stop — a claim, an offer, a hook that is not the same promise as last week's cut.

Three consequences worth planning around:

1. Your product page is now the creative brief. An agent that builds assets from product data can only say what the data supports. Thin attributes, missing use cases and vague descriptions produce generic ads — and the agent has nothing else to work with.
2. Approval is the only remaining choke point. Amazon explicitly keeps an advertiser approval step before launch. That review is where claims, on-screen text, pricing and legal lines get caught — treat it as a real gate, not a formality.
3. Variant supply is the differentiator, again. One campaign spread across streaming TV, video, display and audio, optimised by AI, rewards a queue of genuinely different claims over re-cuts of a single idea. Name assets by claim so creative-level reporting and new-to-brand metrics have something to attach to.

If you are planning creative for Q4, this is the practical version of the maths: a finished AI UGC ad runs about 1,000 tokens in Prizmad — $39 for 3,000 tokens on Launch, $99 for 8,000 on Starter (≈$12 per ad), $249 for 35,000 on Pro. That is the full cost breakdown if you want to model a creative queue rather than a single ad.

The honest caveats

• Region and timing. Full-Funnel Campaigns are GA in the United States only. DVA+ begins rolling out in late October and continues through the end of the year. The conversational features are "over the coming months" — no dates.
• This is Amazon Ads, not a general ad platform. If you do not sell on Amazon or buy through Amazon DSP, today's announcement changes nothing in your account.
• The numbers are Amazon's. The 25% and 10% figures come from Amazon's own measurement of its own feature; the Canine Naturals and Cuisinart results are individual vendor-published case studies. None of it predicts your return on ad spend.
• Automation does not pick your offer. No agent knows whether your discount, your bundle or your guarantee is the reason someone should buy. It can test the versions you give it — it cannot invent the reason.
• This is a buying and assembly upgrade. A weak hook on a perfectly optimised full-funnel campaign is still a weak hook.

Read the primary announcement rather than the summaries if you are deciding whether to test. And note the lineage: the buy side has been commoditising all year — Amazon extending campaigns into ChatGPT, Google unifying vertical video buying — and each step leaves the same thing in your hands. Not budget. Not placement. The claim, and how many distinct versions of it you can put in front of people.