Amazon Ads’ unBoxed event this week yielded three announcements at the intersection of AI and advertising.
What’s new
Amazon announced on Tuesday:
- Amazon Ads Agent: A unified platform bringing sponsored ads, DVA+ and new Full-Funnel Campaigns together with conversational planning, targeting and analytics. Ads Agent and Full-Funnel Campaigns roll out over the coming months; DVA+ begins rolling out in late October. Press release.
- Branded Conversations and Review Requests: Brands can provide information for sponsored shopping conversations on Alexa and run campaigns to gather customer reviews. Branded Conversations launches in the U.S. in January 2027; Review Requests launches in November 2026. Press release.
- Sponsored Services: A partnership with Yelp connects Amazon shoppers with local service professionals through ads offering calls or quote requests. Available now for eligible U.S. Yelp advertisers in home and auto services. Press release.
Q&A with Amazon
An Amazon spokesperson answered tipsheet’s questions by email:
tipsheet: What explanations will advertisers receive for Full-Funnel Campaigns’ spending decisions between Amazon-owned and third-party inventory?
Amazon: Full-Funnel Campaigns runs on Amazon-owned properties from sponsored ads on Amazon[dot]com to Prime Video ads. It also runs placements across the open internet and third-party publishers. Full-Funnel Campaigns will optimize placements based on where the brand is most likely to reach new-to-brand customers and drive long-term sales.
tipsheet: What controlled testing options will advertisers have to verify that Full-Funnel Campaigns generates incremental sales?
Amazon: Full-Funnel Campaigns provides a unified view of performance across all formats in a single report. Advertisers see the standard metrics they already use today — impressions, clicks, sales — plus expanded new-to-brand metrics and Long-Term Sales.
In beta, advertisers saw 67% higher long-term return on ad spend and 29% lower cost to gain a new-to-brand customer, compared to running the same ad formats in separate campaigns.
tipsheet: Can Conversational Briefs include instructions governing which products Alexa recommends, beyond the product information and use cases Amazon has described?
Amazon: No. Sponsored Brands prompts are the entry point into Branded Conversations. Sponsored Brands prompts show up across Amazon[dot]com and across the open internet, as questions or information bubbles that customers can click. Only when a customer clicks a prompt (which is an ad itself), then it opens a Branded Conversation within Alexa for Shopping that continues the conversation about that brand, using information provided by the brand. Branded Conversations will not show up in Alexa for Shopping without first clicking a sponsored prompt.
tipsheet: What triggers an advertiser charge in Branded Conversations—the initial prompt click, subsequent interactions or another event?
Amazon: Advertisers pay on a cost-per-click basis only on the conversation-starter prompt. Follow-up turns within the conversation are free.
tipsheet: Can Amazon or advertisers exclude purchasers from Review Requests campaigns based on returns, complaints or other signals of dissatisfaction?
Amazon: No. Review Requests appear to customers who purchased the product. Customers are never offered anything in exchange for their review, and advertisers cannot edit or remove reviews. The reviews that result carry the verified-purchase badge and are subject to Amazon’s standard review policies.
Read
- “How Amazon is streamlining ad buys across multiple formats with AI” (Sept. 29) – Marketing Dive
- Melissa Burdick, President and co-founder of Pacvue, a commerce media platform and Amazon Ads partner, provided a thorough event breakdown here. (Sept. 29)
From tipsheet: Amazon is giving AI more control over where ads run and creating a paid entry point into shopping conversations. Its answers clarify how advertisers pay for Branded Conversations, but leave open how they can inspect Full-Funnel Campaigns’ spending decisions and independently test whether the resulting sales are incremental.
PLATFORMS
Safari ‘block’ puts ad infrastructure in focus
Apple’s latest operating system is blocking The Trade Desk from serving ads on Safari, according to AdExchanger’s James Hercher.
The iOS 27 update added advertising identity services including UID2, ID5 and LiveRamp to Apple’s blocklist. It also blocked The Trade Desk’s core ad request and delivery domain, extending the disruption beyond identity matching to ad delivery itself.
Hercher reports:
“And now alarm bells are going off at The Trade Desk.
‘Adsrvr[dot]org is The Trade Desk’s core ad request and delivery domain, not identity,’ writes Ian Meyers, The Trade Desk’s senior director of engineering, in an urgent request [on WebKit’s bug tracker] to Apple’s WebKit team.”
Apple’s WebKit team is investigating. It remains unclear whether Apple will reverse the block. Neither Apple nor The Trade Desk commented before publication.
Trusted Server to the rescue?
Meanwhile, IAB Tech Lab CEO Anthony Katsur argues on X the incident makes the case for its Trusted Server initiative, which moves advertising functions into publisher-controlled server infrastructure.
In a follow-up exchange on the same X thread, he says, “This 100% would have prevented what happened to TTD. Trusted Server can proxy all 3P calls on the server side today.”
Read: “Apple’s Latest Operating System Blocks The Trade Desk From Serving Ads On Safari” (Sept. 29) – AdExchanger
From tipsheet: Katsur’s response points to a reason publishers may want more of the advertising workflow running on infrastructure they control. If a browser update can cut off a major buyer’s ad delivery, where those functions execute has commercial consequences. Moving them onto publisher-controlled servers makes infrastructure a question of control as well as reliability.
Related: “Index Exchange CEO Andrew Casale: Every DSP will containerize” (Sept. 30) – tipsheet
PLATFORMS
OpenAI wants “max automation from day zero”
Analyst Eric Seufert quizzes OpenAI monetization product and engineering leader Asad Awan about ChatGPT Ads in the latest Mobile Dev Memo podcast.
Among the highlights, Awan explains some of the thinking behind the ChatGPT Ads stack, which aims to “start from max automation from day zero” (formatted as bullets for readability):
“We want to start with knowing:
- what AI can do to select a really good ad
- what AI can do to make it presentable to the user
- how do we maximize the ranking versus narrow targeting
- how do we simplify bidding so that it is not like 50 types of bidding systems.
Of course we have a max bid, but we have an autobidder now as well which is doing really well, finds you cheap opportunities. Maxing out towards automation because over time every system learns that automation is the best way to deliver actual advertiser value.”
His longer-term vision for the advertiser’s role is prompt-based: “At some point in the future, same as what Codex did for coding, you declare what you want, you explain what you want and how it is done becomes irrelevant.”
My favorite exchange concerned what OpenAI is optimizing for with ads. Awan identified a priority order:
- User trust
- User value
- Advertiser value
- Revenue
Listen: “Podcast: OpenAI’s advertising ambitions (with Asad Awan)” (Sept. 30) – Mobile Dev Memo (subscription)
More ChatGPT ads updates: “OpenAI expands ChatGPT Ads with bulk product campaigns and deeper reporting” (Sept. 30) – Search Engine Land
SELL SIDE
Compensating publishers for AI answers
Building on news from two weeks ago regarding Google’s AI contribution — or compensation — pilot for publishers, The Information reported on Tuesday:
“Google started testing the program involving its AI Overviews, AI Mode and Gemini chatbot less than a year ago, according to a person close to Google. Payments have varied widely. One publisher that joined a few months ago has earned about $50,000 to $60,000 so far, which doesn’t represent that much of its business, the publisher said. Another that joined early on is earning more than $1 million a year, which represents a bigger slice of its revenue, according to people familiar with the program. Smaller sites have earned less than $1,000 over several months, according to those people.”
Read: “Google Is Paying About 100 Digital Publishers for AI Overviews” (Sept. 29) – The Information (subscription)
COMMERCE MEDIA
Grocery AI brings brand influence into the basket
“Oftentimes when people think of using AI to order, they think of a chat window popping up,” DoorDash co-founder Andy Fang said at last month’s Groceryshop event.
But shoppers often ignore that window because they don’t know what to ask, he explained. DoorDash is finding more useful entry points in familiar tasks: reordering usual purchases and finding alternatives to out-of-stock products.
Reporting from the event, Modern Retail describes how retailers are building agentic shopping assistants around everyday needs.
- Kroger’s assistant turns meal ideas, dietary requirements and even handwritten shopping lists into baskets.
- Giant Eagle’s recipe generator lets shoppers add ingredients directly to their carts.
Kroger also sees opportunities for CPG partners to reach shoppers through retail media advertising and recipe content within that contextual shopping experience.
Read: How Kroger, Giant Eagle and DoorDash are building their agentic AI assistants (Sept. 29) – Modern Retail
From tipsheet: The advertising opportunity depends on making the assistant useful enough that shoppers let it help select products. A dinner suggestion that becomes a basket creates an opening for brand influence at the moment of choice. For CPG advertisers, the question is where they can pay to participate and how that paid influence interacts with shopper preferences and product availability.
PLATFORMS
AppLovin says AI needs holiday head start
With Q4 officially underway today (October 1), AppLovin is preparing clients — including users of its self-serve AppLovin Ads platform — for Black Friday and Cyber Monday (BFCM).
AppLovin’s Sue Deng and Shirley Deng shared new analysis yesterday of BFCM 2025, making the case for starting holiday campaigns early, partly to give the buying model time to learn before demand peaks.
In their blog post, they write: “If you are new to AppLovin, launch in October to give the model time to learn and creatives time to gather signals.”
Advertisers that reached twice their October spending level two to three weeks before BFCM recorded a 576% median lift over October in “D7 attributed revenue,” i.e. revenue attributed to ad clicks on a seven-day basis. That was nearly double the 296% lift among advertisers that reached the same spending threshold one week before or during Thanksgiving week.
AppLovin calls the comparison “directional, not causal,” noting that advertisers who hit that spending threshold earlier (“scalers”) may also have anticipated stronger demand.
Read: “BFCM 2026: Start Earlier, Scale Smarter” (Sept. 30) – AppLovin
From tipsheet: With AppLovin Ads having launched in June, this is also a lesson for clients in how to use the self-serve platform: give Axon time to learn before scaling.
AGENTS
Runway connects AI creative to ad performance
Runway is moving beyond video generation into the advertising workflow, according to Adweek’s Trishla Ostwal.
Reporting on the launch yesterday, Ostwal writes:
“Runway Ads launches today inside the company’s web app. The agent connects to a brand’s accounts on Meta, Google, and TikTok. AI development platform Lovable is the launch partner.
The agent can recreate existing ads, spin out variations, localize them for other markets, and publish the human-approved versions. It then pulls the performance data back from those platforms to inform the next round of creative.”
Read: Runway’s AI Video Agent Creates and Measures Ads Across Platforms (Sept. 30) — Adweek
More: “Creating with Runway Agent” – Runway Help
Related: “AI video startup Runway raises $315M at $5.3B valuation, eyes more capable world models” (Feb. 10) – TechCrunch
From tipsheet: Two developments stand out.
- Runway is closing the feedback loop between creative production and ad performance across platforms, giving results from one round of ads a way to inform what gets made next.
- Runway and Higgsfield are expanding beyond video generation into advertising workflows. That puts AI creative companies in a position to compete for more of the advertiser’s daily work, from production to publishing and analysis.
LLMs & CHATBOTS
Developments
- AI agent ordering: “Skip The App: Now You Can Just Text DoorDash” (Sept. 30) – DoorDash
- AI watermarking in biology: “Introducing SynthID Bio” (Sept. 30) – Google DeepMind
- Paying for AI: “Musk’s SpaceXAI Considers Overhaul of Pricing for Grok, X Users” (Sept. 30) – Bloomberg (subscription)
MORE
- Tubi AI: In a research paper published in May, Fox-owned Tubi details how its TubiFM foundation model combines recommendations and search ranking to improve content discovery and reduce latency (Sept. 29) – LinkedIn | Research paper
- Meta connecting: “Business Agent Platform docs” (Sept. 18) – Meta Developer site
- Case study: “Kantar turns advertising intelligence into action with Microsoft AI” (Sept. 28) – Microsoft
- India: “ASCI sets rules for AI-generated ads, mandates disclosure for synthetic content” (Sept. 29) – Storyboard 18