eCommerce AI Agents: Discovery to Checkout in 2026

AI agents are reshaping how consumers discover and purchase online — and the merchant-readiness gap between early movers and the field has never been wider. Adobe’s March 2026 data confirmed the pivot: AI-referred traffic converted 42% better than non-AI traffic, reversing the 38% deficit recorded just twelve months earlier. Five distinct AI shopping surfaces now compete for the same consumer intent, and they all draw from overlapping product-attribute requirements.

The stakes are structural. Universal Cart — Google’s cross-merchant cart primitive expected to debut tomorrow at I/O 2026 — would let shoppers add items from Search, Gemini, YouTube, and Gmail to a single persistent cart. ChatGPT Shopping already serves comparison research to Free, Go, Plus, and Pro subscribers. Amazon’s Alexa for Shopping (formerly Rufus, rebranded May 13) and Walmart’s Sparky are already inside third-party AI surfaces. The merchant that is not schema-compliant, GTIN-complete, and feed-accurate by H2 2026 is invisible to all of them.

This playbook maps the five active AI shopping surfaces, the Agentic Commerce Protocol that powers two of them, a cross-surface merchant-readiness checklist, and a prioritized H2 2026 action plan. Everything is framed for the day before Google I/O — the window when action still determines presence.

In March 2025, AI-referred traffic to US retail sites converted 38% worse than non-AI traffic — a meaningful deficit that led many merchandising teams to deprioritize AI-surface readiness. By March 2026, Adobe’s Digital Trends data showed the same metric at +42% in AI’s favor — a 80-percentage-point swing in twelve months.

This is the most under-reported pivot in eCommerce data right now. Most coverage focuses on traffic volume growth — AI traffic to US retail sites rose 393% YoY in Q1 2026 per Adobe — but conversion quality matters more than raw volume. AI shoppers arriving in 2026 are arriving with higher commercial intent: they have already researched, compared, and filtered through a conversational interface before landing on the product page. Adobe also reports that AI-referred visitors spend 48% longer on retail sites and browse 13% more pages per visit compared to non-AI traffic.

The interpretation matters for merchant prioritization. If AI traffic converted poorly, the case for feed investment was speculative. With AI traffic now converting better than non-AI, the business case for schema compliance, GTIN completeness, and ACP onboarding is immediate. Products that are AI-discoverable win now.

Pre-event leaks tracked by 9to5Google and retail-trade coverage describe Universal Cart as a cross-merchant persistent cart spanning Google Search, Gemini, YouTube, and Gmail. Leaked framing suggests that items added in any surface would persist across all of them — a single cart primitive across Google’s consumer touchpoints.

The Google I/O 2026 keynote opens tomorrow, May 19, at 10:00 AM PT. Pre-event signals indicate Universal Cart will be the Shopping-track headline. The underlying standard is the Universal Commerce Protocol (UCP) — Google’s open commerce standard already covered in depth in our multi-item cart mechanics deep-dive. UCP defines how merchant product objects, cart state, and checkout handoffs are structured for AI agent traversal.

Expected initial merchant coverage (per pre-event leak threads) includes select Shopify merchants and major US retailers. The expected rollout sequence is Search and Gemini first, in the US, with YouTube and Gmail to follow. From a merchant standpoint, the practical question is not “is Universal Cart announced?” — it is “are my products structured correctly to appear when it is?”

Agent Payments Protocol (AP2) is expected to be re-emphasized at I/O 2026 as the payment-layer counterpart to UCP — providing tamper-proof digital mandates between buyer, merchant, and payment processor. The combination of UCP (product layer) and AP2 (payment layer) would make Universal Cart a full-stack commerce primitive, not just a browsing feature.

The 2026 AI shopping landscape is no longer a single platform story. Five distinct surfaces now compete for consumer shopping intent — each with a different owner, protocol, and checkout mechanism, but all drawing from overlapping product-attribute requirements. Understanding where each surface sits helps merchants prioritize which readiness investments are universal vs. surface-specific.

The surfaces diverge most on checkout mechanism and merchant onboarding model. Universal Cart and ChatGPT Shopping / Copilot Checkout use compatible but distinct open standards (UCP vs ACP). Perplexity’s Shopify integration makes Shopify sellers automatically eligible without additional onboarding. Alexa for Shopping is Amazon-marketplace-exclusive — the most closed of the five. For merchant-readiness purposes, the universal baseline (Schema.org Product, GTIN, Merchant Center feed) serves four of the five surfaces; Alexa for Shopping requires Amazon Seller Central separately.

The Agentic Commerce Protocol (ACP) launched in September 2025, co-developed by OpenAI and Stripe as an Apache-2.0 open standard. Our full ACP guide covers the specification in depth; the key facts for merchant readiness are as follows.

ACP introduced the Shared Payment Token (SPT) — a new Stripe payment primitive that lets AI applications like ChatGPT initiate a purchase without ever seeing the buyer’s full payment credentials. Each SPT is scoped to a specific merchant and cart total, making it a single-use cryptographic mandate rather than a stored credential. The buyer authorizes once; subsequent purchases on the same surface reuse the SPT until it expires or the buyer revokes it.

Initial ACP launch merchants on ChatGPT were US-based Etsy businesses, followed by Shopify merchants including Glossier, Vuori, Spanx, and SKIMS. Per Stripe’s launch announcement, ACP is designed so that any merchant using Stripe as their payment processor can participate without bespoke integration work. Microsoft adopted the same ACP standard for Copilot Checkout, making ACP the de facto cross-platform protocol for AI-agent checkout as of mid-2026.

The architectural choice matters: ACP sits at the payment layer, while UCP (Universal Commerce Protocol) sits at the product and cart layer. The two are complementary — a merchant optimized for both UCP (Google surfaces) and ACP (OpenAI + Microsoft surfaces) covers four of the five major AI shopping destinations with a single technical investment. The OpenAI buy-in-ChatGPT announcement remains the primary source for the ACP consumer-facing design.

The most actionable insight from mapping all five AI shopping surfaces is how much their merchant-readiness requirements overlap. Schema.org Product JSON-LD, Google Merchant Center feed quality, and GTIN completeness are required or recommended by every surface except Alexa for Shopping (which routes through Amazon Seller Central instead). Investing in feed quality is the universal multiplier — the same data fuels Universal Cart, AI Mode shopping, Google Shopping ads, schema-based AI Overview citations, and the per-protocol product objects in ACP.

Per industry coverage of Google I/O 2026 sessions, schema-tagged product pages are cited 3.1x more frequently in AI-generated shopping results. Per Google Merchant Center’s own GTIN documentation, products with GTINs can receive up to 40% more clicks than products without them — Google uses GTINs to match products to its database, pull reviews, compare prices, and surface products in more query contexts. Learn more about required Merchant Center product attributes and the broader Google Shopping product-feed strategy for 2026.

The checklist below maps nine core readiness items across the five AI shopping surfaces. “Required” means absence likely excludes the product from that surface. “Recommended” means absence reduces visibility or ranking without hard exclusion. The full merchant AI-readiness self-assessment covers implementation specifics for each item.

While Universal Cart and ACP represent the emerging standard layer, the largest retail platforms have already shipped proprietary AI shopping agents with measurable adoption. These cases illustrate the scale of the opportunity and the platform-specific nuances merchants need to understand.

The vendor picture reveals two distinct strategies. Amazon and Walmart built proprietary agents tightly coupled to their own inventory, then expanded outward into third-party surfaces (Walmart Sparky into ChatGPT/Gemini). OpenAI and Microsoft built protocol-first (ACP) and opened enrollment to any merchant on compatible payment infrastructure. Google’s Universal Cart takes the protocol-first approach at the product layer (UCP) while retaining Google’s existing merchant-onboarding infrastructure (Merchant Center). Shopify’s Magic + Sidekick suite spans both — Shopify merchants are auto-eligible for Perplexity and ACP-native for ChatGPT checkout.

For most mid-market merchants, the priority ranking is clear: fix Merchant Center feed quality and Schema.org Product first (covers Universal Cart + AI Mode + Google Shopping), then onboard to ACP via Stripe or Shopify (covers ChatGPT + Copilot), then monitor Perplexity eligibility (auto for Shopify merchants). Amazon Seller Central is a separate track.

The AI shopping data landscape in mid-2026 has three reliable primary-source anchors: Adobe’s Digital Trends Consumer Report, Adobe’s quarterly retail traffic reports (sourced via TechCrunch), and vendor-published usage figures from Amazon, Walmart, and Shopify. Each category carries different attribution confidence.

Adobe-sourced data (highest confidence): Q1 2026 AI traffic to US retail sites grew 393% YoY per TechCrunch’s report on Adobe data. March 2026 AI conversion was +42% vs non-AI, reversed from -38% in March 2025. Holiday 2025 AI traffic grew 693% YoY (November-December 2025). AI agents influenced approximately $262B of global online spend in Holiday 2025 — roughly 20% of the $1.29T total. AI-referred visitors spend 48% longer on retail sites and browse 13% more pages per visit. 39% of consumers have used AI for online shopping; 85% say it improved their experience.

Vendor-reported data(treat as directional): Amazon reports 250M+ Alexa for Shopping users, 149% MAU growth YoY, and $10-12B in incremental annualized sales. Amazon also reports a 60% lift in purchase-completion likelihood for users of the assistant — none of these figures are independently audited. Walmart reports approximately 50% of app users have tried Sparky since its June 2025 launch. Shopify-attributed industry coverage cites 15x growth in AI-search orders between January 2025 and January 2026.

What the data projects forward:the structural insight from the Adobe conversion data is not merely “AI traffic is growing” — it is that AI-referred shoppers are arriving with materially higher commercial intent than non-AI traffic. If the conversion differential continues compounding at even a fraction of the 2025-to-2026 rate, AI-referred traffic could represent a disproportionate share of retail revenue by the end of H2 2026. This is the case for treating AI-surface readiness as a revenue investment rather than a marketing experiment.

Given the five-surface landscape and the convergent feed requirements, the merchant action plan for H2 2026 has a clear priority stack. Not all actions are equal — Schema.org Product + GTIN + Merchant Center feed quality is the universal multiplier that covers four of the five surfaces simultaneously. ACP onboarding covers two more. Amazon Seller Central is orthogonal.

The sequencing matters. Feed quality and Schema.org are infrastructure investments that compound over time — the sooner they are in place, the more indexing lead time is available before Universal Cart’s expected summer 2026 rollout in Search and Gemini. ACP onboarding is a one-time enrollment that unlocks two major surfaces simultaneously. Analytics baseline setup is low-effort and high-return: without a pre-launch baseline, it is impossible to measure the revenue impact of any of the above investments.

For teams planning H2 2026 eCommerce investment, the framing is simple: the five AI shopping surfaces are not separate channels requiring separate budgets — they are one investment in feed quality and schema compliance, expressed across five distribution points. The merchant AI-readiness self-assessment covers implementation specifics for each checklist item with estimated effort and owner assignments.