AI Shopping Assistants and the New Commerce Stack
> AI shopping assistants are software agents that accept a buyer's purchase intent in natural language, query merchant catalogs via machine-readable protocols, and complete purchases on the buyer's behalf — requiring merchants to expose structured product data and commerce APIs rather than relying solely on HTML storefronts.
---
What Are AI Shopping Assistants?
AI shopping assistants are AI-powered software systems — embedded in products like ChatGPT, Google AI Mode, and Gemini — that can interpret a buyer's natural language purchase intent and take action to fulfil it. Unlike traditional search, which returns a list of pages for a human to evaluate, a shopping assistant evaluates options programmatically, applies the buyer's criteria (price ceiling, product attributes, shipping speed, return policy), and presents a curated result set — or, where checkout protocols are in place, completes the purchase entirely within the conversational interface.
These assistants are not browsing the web the way a human does. They query structured data directly: product feeds, commerce APIs, and machine-readable manifests. The Agentic Commerce Protocol (ACP) is an open specification enabling a programmatic exchange between buyers, their AI agents, and sellers to complete a purchase — agents render the checkout UI, collect buyer selections and payment credentials, and keep customers informed in real time, while sellers keep their existing back-end models and payment processing. The Universal Commerce Protocol (UCP), co-developed by Google with industry partners including Shopify and Walmart, enables agents to discover merchant capabilities, negotiate available actions, and execute transactions without navigating a website.
The practical implication is that AI shopping assistants are a distinct commerce channel with their own data requirements — and merchants need to prepare for them specifically, just as they previously prepared separate feeds for Google Shopping or Meta Commerce Manager.
---
How AI Shopping Assistants Connect to Merchant Catalogs
AI shopping assistants connect to merchant catalogs through two primary mechanisms: structured product feeds and live commerce APIs.
Structured product feeds are the data foundation. For ChatGPT Shopping, this means a merchant-controlled feed in a format the system can index — providing accurate pricing and availability without relying on passive crawling. For ACP-based surfaces, this means a JSON-LD structured feed that exposes product identity fields (SKU, GTIN, MPN), pricing (including sale prices with start and end dates), trust signals (return policy, digital vs. physical status), structured variants, Q&A pairs, and structured reviews. The quality and completeness of this feed directly determines whether an AI assistant can confidently recommend the product.
Live commerce APIs handle the transactional layer. Under UCP, the protocol exposes a Catalog Search API for free-text queries with filters and pagination, a Catalog Lookup API for querying by item ID or barcode, a Cart API for creating and managing multi-item purchase sessions, and an Identity Linking API for connecting buyer accounts. Under ACP, the Checkout API handles session creation, updates, completion, and cancellation. These APIs allow an AI assistant to move from "has this product" to "can I purchase it right now" in a single programmatic sequence.
Both layers must be present for a merchant to be fully reachable by an AI shopping assistant. A merchant with a great feed but no checkout API can be discovered and recommended but not purchased programmatically. A merchant with a checkout API but poor product data may be reachable but will not surface well against merchants with complete, enriched catalogs.
---
Key Components: What the New Commerce Stack Requires
| Component | What It Is | Protocol | |---|---|---| | Machine-readable manifest | Declares merchant capabilities at `/.well-known/ucp` so agents know what the merchant supports | UCP | | Structured product feed | JSON-LD or compatible format with complete product fields, trust signals, and enrichment data | ACP / UCP | | Catalog Search API | Free-text search endpoint returning structured product results | UCP | | Catalog Lookup API | Retrieve specific products by item_id or barcode (GTIN/MPN) | UCP | | Cart API | Create, update, and manage multi-item purchase sessions | UCP | | Checkout Session API | Full checkout lifecycle: create, update, complete, cancel | ACP | | Payment credential relay | Secure token-based delegation from PSP (Stripe / Adyen / Braintree) | ACP | | Order webhooks | Merchant emits `order.created` and `order.updated` events to keep assistant in sync | ACP |
---
Why It Matters for Merchants
The commerce stack required for AI shopping assistants is meaningfully different from the stack required for traditional e-commerce channels. Adding a Google Shopping feed or a Meta catalog used to be the primary form of feed management. AI shopping assistants require structured APIs, machine-readable manifests, payment credential relay, and enriched content fields that most traditional feed tools do not generate.
Merchants who do not build this new stack are absent from a growing purchase surface. Merchants who do build it gain a channel where purchase intent is already resolved — the buyer has already decided to buy something fitting a description, and the agent is simply finding the best match. This is a high-intent context that is structurally different from display advertising, where merchants interrupt a browser, or even traditional search, where the buyer must still navigate to a product page and complete a checkout manually.
The competitive advantage in agentic commerce is catalog quality. A product with complete attributes, accurate availability, structured variants, and answered Q&A pairs is far more interpretable by an AI agent than a product with a single image and a brief description. Merchants who invest in their data layer build durable discoverability.
---
FeedBridge Relevance
FeedBridge directly addresses the new commerce stack requirements for merchants who want to be reachable by AI shopping assistants. On the product data side, FeedBridge's AI enrichment engine operates across eight verticals — food, electronics, apparel, beauty, home, health, digital, and other — generating the intent tags, persona targeting arrays, use case descriptions, AI Q&A pairs, and voice snippets that AI assistants use to evaluate product fit. FeedBridge also generates ACP-compliant JSON-LD feeds for ChatGPT Shopping and implements the full UCP protocol stack including the `/.well-known/ucp` manifest, Catalog Search API, Catalog Lookup API, Cart API, and Identity Linking API.
Beyond protocols and feeds, FeedBridge gives merchants tools to stand in for an AI assistant's perspective: the AI Chat Simulator previews how an AI assistant would present their products, the Custom GPT Builder generates system instructions and knowledge files merchants can use to deploy a product-aware GPT, the Gemini Gem Builder provides a Google AI Studio integration guide, and the WhatsApp Bot Kit provides bot instructions and conversation flows. All of these tools are live on the platform as of April 2026. Merchants can start with the free AI Readiness Checker at feedbridge.ai/score to see how their current catalog measures against the data requirements AI shopping assistants expect.
---
Frequently Asked Questions
Q: Do I need to implement both ACP and UCP to work with AI shopping assistants? A: ACP and UCP serve different assistants and different protocols. ACP is required for ChatGPT Instant Checkout. UCP is the protocol for Google AI Mode. A merchant who wants to be reachable on both surfaces will need both. They share common data requirements — structured product data, trust signals, complete catalog fields — so the underlying catalog work overlaps significantly.
Q: Can I use my existing Shopify or WooCommerce feed for AI shopping assistants? A: A standard Shopify or WooCommerce product export will not meet the full requirements for ACP or UCP. Both protocols require enrichment fields — intent tags, Q&A pairs, structured variants, trust signals — that standard platform exports do not include. Merchants typically need a layer of AI enrichment and protocol feed generation on top of their existing catalog.
Q: How does an AI shopping assistant know what my return policy is? A: Return policy information must be encoded in structured fields in the product feed or trust signal layer. ACP supports `accepts_returns`, `return_deadline_days`, and related fields. If these are not present, the agent cannot surface the return policy to the buyer, which may reduce the buyer's confidence in completing the purchase.
Q: Are custom AI assistants (GPTs, Gemini Gems) the same as ChatGPT Shopping or Google AI Mode? A: No. Custom GPTs and Gemini Gems are merchant-deployed AI assistants for their own storefront or catalogue — they help buyers navigate a specific brand's products. ChatGPT Shopping and Google AI Mode are platform-level assistants that query across multiple merchants via open protocols. Merchants benefit from building both: a custom assistant for their owned properties and protocol compliance for platform-level surfaces.
Q: What happens when an AI assistant queries a product that is out of stock? A: Under ACP, the merchant's checkout session endpoint returns an error message with code `out_of_stock` as part of the `messages` array in the checkout session response. The agent surfaces this to the buyer. For discovery and evaluation to remain accurate, inventory availability must be kept current in the product feed. Real-time inventory sync is a known roadmap gap in FeedBridge — this feature is planned and not yet live.
---
Related Topics
Parent hub: Agentic Commerce Foundations
Related concepts:
- What Is Agentic Commerce?
- How Agentic Commerce Changes Online Buying
- What Merchants Need for Agentic Commerce Readiness
Breadcrumb:
---
Source Documentation
| Claim | Source | Source Class | Reference | |---|---|---|---| | ACP: agents render checkout UI, collect credentials, sellers keep back-end | agenticcommerce.dev/docs | T1 – Official ACP Docs | https://www.agenticcommerce.dev/docs | | UCP co-developed by Google; enables agents to discover, negotiate, transact | Shopify Engineering | T1 – Official UCP Docs | https://shopify.engineering/UCP | | UCP APIs: Catalog Search, Catalog Lookup, Cart, Identity Linking | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md | | ACP Checkout API: create, update, complete, cancel sessions | OpenAI ACP Checkout Spec | T1 – Official ACP Docs | https://developers.openai.com/commerce/specs/checkout/ | | ACP PSPs: Stripe, Adyen, Braintree | OpenAI ACP Checkout Spec | T1 – Official ACP Docs | https://developers.openai.com/commerce/specs/checkout/ | | FeedBridge Custom GPT Builder, Gemini Gem Builder, WhatsApp Bot Kit — live | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md | | Real-time inventory sync is a roadmap gap, not yet live | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md |