Product Data for Five AI Shopping Surfaces

AI traffic to US retail sites grew 393% year-over-year in Q1 2026 (Adobe), and by March 2026 that traffic converted 42% better than non-AI sessions — a complete reversal from March 2025, when AI traffic converted 38% worse. The merchants capturing that delta are the ones whose product data is structured, fresh, and aligned with the feed specifications that Universal Cart, ChatGPT Shopping, Perplexity Buy with Pro, Microsoft Copilot Checkout, and Alexa for Shopping (formerly Rufus) actually consume.

The problem is fragmentation. Five surfaces, three underlying protocols, and no single guide that maps the overlap. Most coverage is single-surface: an ACP walkthrough here, a Perplexity merchant signup guide there. What a Head of eCommerce actually needs is a matrix showing which fields matter where — and which fields are shared across all five surfaces so one feed-hygiene investment compounds.

This guide delivers that. It covers the three protocols powering the five surfaces, the per-surface readiness requirements, and the 26-item feed-readiness checklist — organized by owner and difficulty — that you can hand to engineering on Monday. For the Shopify shortcut (auto-enrollment in UCP, ACP, and ChatGPT via Catalog Broadcast), see Shopify merchants who inherit the UCP stack automatically. For the full UCP protocol deep-dive, see the complete Universal Commerce Protocol guide.

In the twelve months before May 2026, the AI shopping surface landscape went from a single early experiment (ChatGPT Instant Checkout, September 2025) to five live or imminent checkout surfaces. Each launched on a different protocol, with different feed spec requirements and different onboarding paths. The merchant who reads each surface's guide in isolation misses the most operationally important insight: the majority of required fields are shared.

Google's Shopping Graph contains 60 billion product listings with more than a billion shopping interactions per day — the largest index any AI shopping surface currently pulls against. Universal Cart, announced at Google I/O Day 1 on May 19, 2026 by Vidhya Srinivasan (VP/GM Ads & Commerce), is the consumer-facing product layered on top of the Universal Commerce Protocol. It rolls out in the US on Search and the Gemini app first, summer 2026; YouTube and Gmail follow.

The structural case for a single feed strategy: once you have clean GTIN, title, description, image, price, availability, brand, and category data at the SKU level, you are 70% of the way to eligibility on every AI shopping surface simultaneously. The per-surface work — native_commerce: true for Universal Cart, is_eligible_checkout: true for ChatGPT, the /.well-known/ucp business profile for direct UCP integrations — is additive, not duplicative. See the 2026 Google product-feed strategy baseline for the foundational feed-hygiene work.

The single most operationally useful mental model for this landscape is the protocol-behind-the-surface map. Merchants who understand which protocol each surface runs — and therefore which feed a single integration covers — can sequence their build effort rationally.

Google Universal Commerce Protocol (UCP). UCP is the open multi-vendor protocol Google co-developed with retail and payments partners. Universal Cart is the consumer product layered on top. A UCP business profile JSON manifest at /.well-known/ucp publicly declares server capabilities, transport bindings (REST or MCP — Model Context Protocol), and payment-handler configs. Required for direct UCP integrations. Shopify merchants can access UCP via Google MC connection without writing the manifest themselves. See UCP multi-item cart mechanics for the protocol internals.

Agentic Commerce Protocol (ACP). OpenAI and Stripe co-developed ACP, released September 2025 under Apache 2.0. The Stripe Shared Payment Token (SPT) is the new payment primitive — scoped to a specific merchant and cart total so ChatGPT never sees the buyer's payment credentials. Microsoft Copilot Checkout launched on ACP in January 2026. For the full protocol primer, see the Agentic Commerce Protocol (ACP) guide — OpenAI + Stripe's open standard.

Microsoft: the dual-protocol bridge. On April 21, 2026, Microsoft Merchant Center began consuming UCP feeds — making Microsoft Copilot the only major AI platform that ingests both ACP and UCP. One correct Merchant Center feed or UCP manifest now covers Universal Cart (via UCP) and Microsoft Copilot (via both), while a correct ACP feed covers ChatGPT Shopping and Microsoft Copilot's ACP path simultaneously. This is the strategic argument for fixing feed hygiene before surface-specific work: the protocols are additive.

Perplexity and Amazon. Neither operates on UCP or ACP. Perplexity accepts CSV feeds following the Google Merchant Center specification (per agency synthesis; Perplexity has not published a single canonical feed spec document). Amazon and Alexa for Shopping operate on Amazon's proprietary catalog — there is no protocol to integrate; the lever is the listing quality in Seller Central. For how all four protocols fit together, see how MCP, A2A, ACP, and UCP fit together.

Universal Cart launched at Google I/O Day 1 (May 19, 2026). It is a cross-merchant cart spanning Search, Gemini, YouTube, and Gmail. The Buy button — the thing every merchant wants — is not automatic. Per Google Merchant Center's UCP-powered checkout documentation, verbatim: "Only product listings using the native_commerce product attribute will display the 'Buy' button for this checkout experience."

native_commerce is a boolean attribute set per SKU in your Merchant Center feed. Products without it participate in discovery — they appear in Universal Cart search results and Gemini shopping cards — but users cannot complete checkout without leaving the surface. The distinction matters for conversion: the Buy button keeps users in-surface through Google Pay; the redirect sends them to the merchant PDP and loses the conversion momentum the AI surface built.

Ineligible categories for native_commerce: true include financial products, recurring billing, personalized goods with custom fulfillment, age-restricted or prohibited content, pure digital goods and services, final-sale products, and products requiring special shipping. If your catalog includes regulated SKUs, the UCP consumer_notice attribute is required — with sub-attributes consumer_notice_type (enum: legal_disclaimer / safety_warning / prop_65) and consumer_notice_message (max 1,000 chars, accepts <b>, <br>, and <i> HTML tags only).

The 2026 Merchant Center spec update introduced a 500×500 px image floor (warnings began April 14, 2026; enforcement starts January 31, 2027). Until enforcement, Google accepts 100×100 (non-apparel) and 250×250 (apparel) as a transition floor and auto-optimizes some sub-500 images. Best practice is 1,500×1,500 or higher. Per the updated spec, new optional attributes include video_link, handling_cutoff_time, minimum_order_value, and loyalty_program_label. For the full Merchant Center 2026 spec, see Google Merchant Center Help — Product Data Specification 2026 update.

OpenAI publishes the official ChatGPT product feed spec at developers.openai.com/commerce/specs/feed/. The core required fields are: item_id (max 100 chars, must be stable), title (max 150 chars, no all-caps), description (max 5,000 chars, plain text, factual — no marketing language), url (must return HTTP 200), brand (max 70 chars), image_url (JPEG or PNG, HTTPS preferred), price (ISO 4217 currency code), availability (enum: in_stock / out_of_stock / pre_order / backorder / unknown), seller_name, seller_url, target_countries and store_country (ISO 3166-1 alpha-2), plus the OpenAI integration flags is_eligible_search and is_eligible_checkout (the latter requires the former).

Conditionally required fields: availability_date (ISO 8601, required if pre_order), seller_privacy_policy + seller_tos (required if is_eligible_checkout: true), dimensions_unit (required if any dimension field is provided), and item_weight_unit (required if weight is provided).

Recommended optional fields that meaningfully improve discoverability: gtin (8–14 digits), mpn (max 70 chars), product_category (> separator), sale_price with start and end dates, group_id and variant_dict for variant families, additional_image_urls, video_url, model_3d_url (GLB/GLTF preferred), return policy fields, and structured review data including q_and_a objects and per-review title / content / rating objects.

The delivery model is push, not pull. Merchants push files via encrypted HTTPS to an OpenAI-provided allow-listed endpoint. Merchants may push updates as often as every 15 minutes to keep pricing and inventory near-real-time. This cadence advantage is significant for flash sales and dynamic repricing: a Google MC scheduled feed runs daily; ChatGPT's push model can surface a price change within a quarter-hour. The leading rejection reasons are pricing mismatches between feed and on-site PDP, missing required fields, malformed data types, and marketing language in description fields. For the merchant-focused walkthrough, see the ChatGPT Shopping merchant onboarding guide. For background on OpenAI's pivot away from in-surface Instant Checkout (March 2026), see OpenAI's pivot away from in-surface Instant Checkout.

Perplexity Buy with Pro. Perplexity launched Buy with Pro on November 18, 2024 for US Pro subscribers, with in-surface checkout powered by PayPal and free shipping funded by Perplexity (not the merchant). The Perplexity Merchant Program is open to any merchant that ships products to the US — no minimum revenue or product count — and applications take under five minutes via the signup form at perplexity.ai.

Perplexity accepts CSV feeds following the Google Merchant Center specification, delivered via SFTP or secure server. Merchants with an existing Google Shopping feed have a direct path to enrollment. Critical: Perplexity treats products without GTINs as effectively invisible — GTINs are the primary deduplication and matching key. Equally critical: Perplexity drops products with stale pricing or out-of-stock signals from results entirely. Feed accuracy is a participation requirement, not a ranking lever. Note: Perplexity has not published a single canonical merchant specification document; the requirements above are synthesized from agency guides (Alhena, AI Advantage) and should be treated accordingly.

Amazon Alexa for Shopping (formerly Rufus).Amazon rebranded Rufus to Alexa for Shopping on May 13, 2026 — eleven days before this post publishes. The underlying intelligence and Bedrock model stack (Anthropic Claude Sonnet, Amazon Nova, and Amazon's custom model) carry over intact. There is no separate Alexa for Shopping feed. The surface pulls from the Amazon product detail page, A+ Content, reviews, and Q&A. The entire merchant lever is the quality of the existing Amazon listing. Amazon main-image standards require a pure-white RGB(255,255,255) background, the product filling at least 85% of the frame, no text, logos, or watermarks, and a 1,000-px minimum on the longest side (2,000–3,000 px recommended; max 10,000 px / 10 MB; JPEG preferred; sRGB color space — CMYK is rejected).

Microsoft Copilot Checkout. Microsoft Copilot Checkout launched January 8, 2026 on Copilot.com in the US, with PayPal, Shopify Payments, and Stripe as payment partners. Launch partners included Urban Outfitters, Anthropologie, Ashley Furniture, and Etsy inventory. Shopify merchants are auto-enrolled (opt-out window only); non-Shopify merchants apply via the Copilot Checkout merchant form. Microsoft added UCP-feed support on April 21, 2026 — the first major AI platform to ingest both ACP and UCP. Microsoft reports 53% more purchases within 30 minutes for Copilot-assisted journeys and 194% higher purchase intent (both vendor-reported, not independently audited). For the Walmart-in-ChatGPT context that parallels the Copilot Checkout model, see the Walmart-in-ChatGPT discovery and checkout walkthrough.

The table below is the artifact a Head of eCommerce hands to engineering. Each row is a readiness item; columns map it to all five AI shopping surfaces (Universal Cart, ChatGPT Shopping, Perplexity, Microsoft Copilot Checkout, and Alexa for Shopping); the last two columns show the internal team owner and implementation difficulty. Sources: the OpenAI ACP feed spec, Google Merchant Center UCP-powered checkout help, Google Merchant Center 2026 spec update, and the Shopify Agentic Commerce guide.

The matrix below renders the 26 items across three logical groups: Foundation (items 1–10), Media (items 11–14), Operational (items 15–20), and Protocol-specific (items 21–26). Items 1–10 are the shared core where one investment covers all five surfaces.

For the 4.6+ million merchants on Shopify, the feed-protocol landscape is materially simpler than it appears. As the Shopify Agentic Commerce guide states: "Shopify Catalog provides the same schema fields automatically for all stores on Shopify." This means:

ChatGPT Shopping (ACP).As of March 2026, products in Shopify Catalog are automatically discoverable in ChatGPT for US merchants. Per-channel direct-checkout toggles live in Settings > Sales Channels. No separate ACP feed engineering is required for discovery; the is_eligible_checkout toggle controls checkout eligibility.

Microsoft Copilot Checkout (ACP + UCP). Shopify Payments is a Copilot Checkout payment partner. Shopify merchants are auto-enrolled in Copilot Checkout with an opt-out window only — no application required.

Universal Cart (UCP). Shopify simplifies the Google MC connection, which is the onboarding path to UCP. However, native_commerce: true is a per-SKU attribute that must still be set explicitly — Shopify's auto-enrollment does not bypass the Buy-button gate. The merchant must opt each eligible SKU into native commerce in the Google MC data feed.

What Shopify does not handle automatically: Amazon (requires Seller Central enrollment), Perplexity (requires a manual application and separate feed setup), and the /.well-known/ucp business profile JSON for merchants who want direct UCP integration rather than routing through Merchant Center. For the full Shopify-to-AI surface picture, see selling inside ChatGPT, AI Mode, and Copilot from a Shopify store and making Shopify products visible inside AI chats.

A clarification that most "schema for AI shopping" guides blur over: the Schema.org types that benefit merchant product pages are different from the schemas appropriate for an agency or publisher site like this one. On your product pages, the following schema types are appropriate and recommended by AI shopping surfaces.

Product with Offer. The canonical product JSON-LD block. Required fields for AI surface discoverability: name, description, image, brand, and an Offer block with price, priceCurrency, and availability (using the schema.org availability enum). Perplexity ingests Product, Offer, Review, and AggregateRating markup directly — structured data on the PDP improves Perplexity discoverability independently of the CSV feed. For the complete Schema.org type reference, see the complete Schema.org type reference.

AggregateRating and Review. On merchant PDPs, AggregateRating and Review schema are appropriate and recommended by Perplexity, Universal Cart, and ChatGPT's optional review fields. The ChatGPT ACP feed supports per-review objects with title, content, minRating, maxRating, and rating. Note: these schema types are appropriate on merchant product pages; they are forbidden on this site (digitalapplied.com) per our site's structured data policy. The distinction matters because several AI shopping guides recommend adding these types globally — that advice applies to merchant PDPs, not to agency sites.

GTIN in structured data. The Product schema supports gtin, gtin8, gtin12, gtin13, and gtin14 properties. For Perplexity (where GTIN absence means invisibility) and ChatGPT (where GTIN is strongly recommended), GTIN in structured data reinforces the feed-level GTIN and provides an additional ingestion path. For the post-I/O structured data implications, see the post-I/O structured-data updates that pair with this feed-readiness checklist.

The 26-item checklist is the full scope; the 30-day plan below is the practical sequence for a merchant starting from a well-maintained Google Shopping feed. If your Google MC feed is clean and your images are at 1,500×1,500 or higher, you are already 60% of the way to multi-surface readiness.

Week 1 — Foundation audit (Merchandising lead). Run a feed diagnostic against the required-field list in the matrix above. Specifically: (1) verify GTIN coverage per SKU — this is the highest-leverage single fix for Perplexity and ChatGPT discoverability; (2) check title length (≤150 chars, no all-caps); (3) audit description fields for marketing language — ChatGPT rejects feeds with marketing copy in description; (4) verify availability fields are real-time or near-real-time (Perplexity drops stale inventory); (5) confirm price in feed matches on-site PDP — ChatGPT rejects price mismatches. See the merchant AI-readiness self-assessment for a structured diagnostic.

Week 2 — Protocol enrollment (Merchandising + Ops). (1) For Shopify: verify Google MC connection is active; toggle ChatGPT and Copilot checkout in Settings > Sales Channels; set native_commerce: true on eligible SKUs for Universal Cart Buy-button access. (2) For non-Shopify: apply to the Perplexity Merchant Program (under five minutes); apply to Copilot Checkout via the Microsoft merchant form; confirm Merchant Center feed is UCP-ready. (3) For Amazon sellers: audit main-image compliance — pure white background, product ≥85% frame, 1,000-px minimum — and review A+ Content completeness.

Week 3 — Media and schema lift (Photo/Media + SEO/Dev). (1) Upgrade primary images to 1,500×1,500 or higher if not already there; add lifestyle and additional-angle images; (2) implement Product + Offer + AggregateRating JSON-LD on PDPs (Perplexity ingests directly; reinforces GTIN for ChatGPT); (3) for ChatGPT ACP, configure the push endpoint and test a 15-minute feed update cycle for high-velocity SKUs.

Week 4 — Tracking and attribution (Analytics). (1) Add per-surface UTM parameters to Perplexity (strongly recommended during Perplexity onboarding), Copilot Checkout, and Universal Cart; (2) configure Amazon Attribution tags for Alexa for Shopping traffic; (3) set up a dashboard segment for AI-referred sessions to track the conversion delta that Adobe measured at +42% for March 2026. For the broader analytics picture, our analytics services team helps merchants build AI-surface attribution models.

The forward-looking projection: the five surfaces active today are likely to be joined by more as Microsoft's dual-protocol adoption proves out the model. Merchants who build their feed hygiene and schema markup correctly in 2026 are compounding across surfaces that don't yet exist. The AI shopping platform landscape in H2 2026 and 2027 will almost certainly add surfaces; the universal feed quality investment is the most durable bet. For the complete agentic-commerce SEO preparation playbook, see the broader agentic-commerce SEO preparation playbook. Our eCommerce services team builds these multi-surface feed strategies for clients across Shopify, BigCommerce, and custom platforms.