Conversational ads converge: the model writes the copy from your feed

Three platforms announced or shipped feed-powered, AI-generated ad formats within a seven-week window in mid-2026, and the mechanic is the same across all three: a language model reads your product catalog at query time and writes the copy that the shopper sees. Adobe Analytics found that AI-referred shoppers during Prime Day 2026 (June 23-26) completed purchases at a rate 40% higher than visitors from any other digital channel¹, a complete reversal from Prime Day 2025, when AI-directed traffic converted 23% worse than alternatives. The implication for merchants is direct: feed quality is now ad copy quality, and those are the same audit.

Three platforms, one mechanic: the model writes the ad from your feed, not from your creative.

What actually happened during Prime Day 2026 to make AI advertising matter now?

Amazon moved Prime Day from its traditional July timing to June 23-26, 2026, the same day it launched Alexa+ Agentic Ads on Echo Show devices.² Total US online spending across the event reached $26.4 billion, up 9.3% year-over-year. AI-directed traffic to US retail sites grew 89% year-over-year during that four-day window¹, after already growing 235% in the first five months of 2026 alone.¹²

The conversion reversal from 2025 to 2026 is the most significant data point in that Adobe report. In 2025, shoppers arriving via AI channels were worse buyers than shoppers arriving via search or email. In 2026, they were better buyers by 40%. Something changed in the quality of the referral, and the most plausible explanation is that conversational interfaces are now doing real query-intent matching before sending shoppers onward, rather than simply linking out from a chatbot response. Whether that matching is accurate depends on what the model can read about your products.

How does each platform pull from the feed to write the ad?

The three platforms share the same underlying principle but differ in how the feed connects to the ad surface, what copy the model generates, and what the merchant controls.

Amazon launched Alexa+ Agentic Ads on Prime Day opening day (June 23, 2026), initially on Echo Show devices with Papa Johns and concert ticketing via Ticketmaster as launch partners.³ The format allows a shopper to view an ad, ask questions, review options, and complete a purchase entirely within a single Alexa conversation, with no redirect to a separate website. Alexa+ generates a conversational product pitch and handles the Q&A within the session.⁴ The exact feed schema powering Agentic Ads is not yet publicly documented by Amazon. Amazon rebuilt its broader ad stack in late 2025 into four products: Campaign Manager (merged Ads Console and DSP), Ads Agent (AI campaign manager), Creative Agent (rebranded Creative Studio for streaming), and a Full-Funnel Tool.⁶

Google announced four new Gemini-powered ad formats at Marketing Live 2026, native to AI Mode in Search.⁷ The format most relevant to feed management is AI-powered Shopping ads: Gemini pulls the advertiser’s most relevant products from their Google Merchant Center feed and writes a custom explainer at query time describing why that specific product fits that shopper’s specific need.⁸ Both Conversational Discovery ads and AI-powered Shopping ads include an independent AI explainer as part of the ad. Gemini evaluates and synthesizes product information from the feed and displays that context alongside the advertiser’s creative, labeled Sponsored. No advertiser-written copy is required. Merchant Center feed quality, product titles, descriptions, attributes, and specifications, is the direct input from which Gemini constructs the explainer.

OpenAI opened ChatGPT Ads Manager self-serve to all US advertisers on May 5, 2026, with no minimum spend requirement, then launched product feed ads in the Ads Manager beta on June 11, 2026.⁹ ChatGPT product feed ads automatically generate sponsored placements from product names, images, and catalog attributes.¹⁰ Ads appear beneath ChatGPT responses, labeled Sponsored. The feed format is compatible with existing Google Shopping catalogs, so brands with a Merchant Center feed can participate without rebuilding from scratch. OpenAI’s developer documentation specifies Boolean eligibility flags at the product level: is_eligible_search, is_eligible_checkout, and is_eligible_ads control which products appear in which surfaces.¹¹ A product missing is_eligible_ads: true is invisible in sponsored placements regardless of bid or budget.

The sharpest structural difference across the three is in what is publicly documented about feed schema. OpenAI has made the eligibility flags explicit at the field level, which is useful for merchants doing a targeted audit. Amazon and Google exercise similar eligibility logic through their existing catalog health systems, but the exact field-level controls are embedded in broader seller tooling rather than published as a discrete spec.

Is the AI-generated copy actually writing from the feed, or pre-indexing at ingestion time?

The honest answer is that the precise generation mechanism is not publicly documented by any of the three platforms. Google’s Marketing Live materials describe Gemini generating an explainer “per query” describing why a specific product fits a specific shopper’s need. OpenAI’s product feed documentation describes pulling from “product names, images, and catalog attributes.” Amazon’s Agentic Ads launch page describes the consumer conversation flow without disclosing the merchant-side feed schema or the moment of generation.

My read, and I label it as an inference, is that the distinction between query-time generation and retrieval from a pre-indexed representation matters less than merchants typically assume. Whether the model generates fresh copy at inference time or retrieves from a semantically rich index it built during ingestion, the accuracy ceiling is the same: the richness and correctness of the feed attributes that went in. A poorly titled product with a thin description and missing specification attributes will surface poorly either way. The feed is the upstream constraint.

What does this mean for merchants who haven’t audited their catalog data?

The audit question has changed. Historically, feed quality affected shopping ad performance through a fairly legible mechanism: better titles and attributes improved match rates and Quality Score. The improvement was real but marginal at the edge. In the AI-generated copy model, feed quality sets the ceiling for what the model can say about your product in a conversational context. A product page with no specification data cannot generate a specific answer to “does this fit a 6-quart Dutch oven.” A product with a generic title like “Blue Sneaker Men” cannot be coherently differentiated from a competitor in a head-to-head Gemini explainer.

Adobe’s own data adds a technical dimension to this: up to 46% of retailer websites are not machine-readable by AI crawlers.¹² That figure applies to organic AI discovery, but the feed dependency in paid AI advertising creates the same gap through a different mechanism. If your catalog attributes are thin, the model has nothing to work with, regardless of whether the entry point is a crawl or a feed upload.

The merchant audit that this moment calls for is not a campaign audit. It is a catalog audit, run against the attributes the model needs to construct a coherent conversational pitch: product titles that include brand, type, and differentiating attributes; descriptions that answer likely conversational questions about fit, compatibility, material, and use case; specification fields populated consistently across the full catalog, not just on hero SKUs; and, for OpenAI specifically, eligibility flags set correctly per product.

Should ecommerce merchants treat this as urgent, or is this still early-stage signaling?

The urgency varies by where you sit in the competitive set. OpenAI’s Ads Manager is self-serve and open to all US advertisers as of May 5, 2026, with no minimum spend requirement. The product feed format is compatible with existing Google Shopping feeds, which means any retailer already managing a Merchant Center feed can participate with modest setup effort. The financial signal is notable: OpenAI’s Ads Manager reached $100 million in annualized revenue in under six weeks of the self-serve launch, which indicates demand from advertisers at scale, not just early testers.

Google’s AI-powered Shopping ads are in active rollout following the Marketing Live 2026 announcement. Amazon’s Agentic Ads launched on Prime Day with a limited partner set. The access curve is expanding, and merchants who have done the feed work will have a structural advantage when full access opens, because the audit takes time that cannot be compressed at campaign launch.

The larger shift, and the one that warrants treating this as more than a platform update, is what Amazon’s Agentic Ads model represents architecturally. In that surface, a shopper can view an ad, ask product questions, review options, and complete a purchase inside a single Alexa conversation with no redirect to a website. If that pattern scales, the product detail page ceases to be the primary conversion environment for a portion of AI-referred demand. Merchants whose only quality-assurance mechanism is the PDP are measuring the wrong thing.