Glossary
Agentic Commerce Definition and Glossary: 60 Terms Explained
Short definitions of the terms brands and retailers meet in AI shopping, from ACP and UCP to query fan-out and conversational attributes, each linked to a deeper page.
Agentic commerce, defined
Agentic commerce is shopping in which AI agents act on a person's behalf: they interpret the need, compare products using structured product data and, with permission, complete the purchase. It already runs in ChatGPT and Google AI Mode, among other assistants, connected to merchants through protocols such as ACP and UCP.
Agentic commerce definition in 30 seconds
Agentic commerce is shopping where AI agents search and compare products for a person, and sometimes buy. Each term below covers one link in that chain.
- 01
The vocabulary splits into four buckets: the agents, the protocols, discovery and product data.
- 02
Protocols such as ACP and UCP move orders and payments. None of them decides which product gets recommended.
- 03
The terms that decide whether you get picked are the boring ones: attributes, identifiers, feeds.
- 04
One short paragraph per term, so you can link it from anywhere and an assistant can quote it.
McKinsey's working definition is "shopping powered by AI agents acting on our behalf" (McKinsey). The long version lives in our agentic commerce guide.
Core concepts: agents and autonomy
Start here: what an agent is, how much it does alone, and the disciplines built around it.
| Term | Definition | Read more |
|---|---|---|
| Agentic commerce | Shopping in which AI agents act for a person: they work out the need, compare products and, with permission, buy. McKinsey calls it shopping powered by AI agents acting on our behalf. | See above |
| Agentic shopping | The shopper's side of agentic commerce: handing research or a purchase to an AI assistant. | See above |
| AI shopping agent | An AI assistant that recommends or buys products, such as ChatGPT, Gemini, Perplexity, Microsoft Copilot or Amazon Rufus. | Marketplaces and AI assistants |
| Assistive agent | An agent that recommends while the shopper still clicks buy on the retailer's site. Most agentic shopping in 2026 works this way. | Product pages that convert |
| Delegated purchase | The agent builds the cart and checks out inside the chat after the shopper confirms the order. | See above |
| Autonomous purchase | The agent buys on its own within rules set in advance, such as a budget, a brand or a price threshold. Rare and heavily guarded. | See above |
| Agent to site | An agent reads and acts on a merchant's own pages, feeds or APIs. The model you control most directly. | Product pages |
| Agent to agent | A shopper's agent talks to a retailer's own commerce agent instead of reading its pages. | See above |
| Agentic Commerce Optimization (ACO) | The practice of making every product in a catalog found and recommended by AI shopping agents, built on complete product data. | What is ACO |
| GEO | Generative Engine Optimization: getting a brand's content cited in AI-generated answers. ACO applies the same goal to product data. | GEO for ecommerce |
| AEO | Answer Engine Optimization: structuring content so search features and assistants can lift a direct answer from it. | The four acronyms |
The autonomy ladder gets the headlines; the assistive rung gets the shoppers. Plan for an agent that shortlists a few products and sends the shopper to your site. It reads the same data an autonomous one would, so nothing is wasted.
GEO and ACO overlap. GEO is mostly about articles and brand mentions; Agentic Commerce Optimization is about the product record, the thing an agent actually compares.
Agentic commerce protocols and payment terms
Protocols are the plumbing between agents and merchants. Learn the four acronyms (ACP, UCP, AP2, MCP), then the payment objects that ride on them.
| Term | Definition | Read more |
|---|---|---|
| ACP | Agentic Commerce Protocol, the open standard co-developed by OpenAI and Stripe and released in September 2025. It covers feeds and checkout, and since March 2026 product discovery in ChatGPT. | Protocols |
| UCP | Universal Commerce Protocol, the Google-led open standard for agent shopping from discovery through checkout and post-purchase. It powers checkout in AI Mode and Gemini. | See above |
| AP2 | Agent Payments Protocol, announced by Google in September 2025. Signed mandates prove what a user authorized an agent to buy. | See above |
| Intent mandate | In AP2, the signed record of a shopper's instructions, such as a price limit, that lets an agent act within those rules. | See above |
| Cart mandate | In AP2, the signed record of the exact items and price a shopper approved. Payment is tied to it. | See above |
| MCP | Model Context Protocol, the open standard Anthropic released in November 2024 that connects AI assistants to external tools and data. Google built AP2 to extend it. | See above |
| A2A | Agent2Agent protocol, a standard for AI agents from different vendors to exchange tasks. AP2 can extend it. | See above |
| Shared Payment Token | Stripe's payment credential for ACP, scoped to one merchant and one cart amount, so the agent never sees card details. | See above |
| Agentic token | A tokenized card credential issued for AI agents, as in Mastercard Agent Pay, where agents must be registered and verified before they pay. | See above |
| Trusted Agent Protocol | Visa's framework, unveiled in October 2025, that helps merchants tell verified shopping agents from malicious bots. | See above |
| Merchant of record | The business legally responsible for a sale (taxes and returns included). In ACP and UCP that stays the retailer, even when checkout happens in a chat. | See above |
| Agentic checkout | Checkout completed by an agent, in the AI app or on the merchant's site, often with a wallet such as Google Pay. | See above |
| Instant Checkout | ChatGPT's in-chat purchase flow built on ACP. In March 2026 OpenAI let merchants use their own checkout and shifted focus to discovery. | ChatGPT shopping |
| Agentic storefronts | Shopify's feature, launched December 2025, that sells a merchant's catalog inside AI chats such as ChatGPT and Perplexity. | PIM for Shopify |
Specs move fast, so dates matter. ACP launched on September 29, 2025 (Stripe), and in March 2026 OpenAI expanded it to product discovery while stepping back from its own checkout (OpenAI). UCP lists Google and Shopify among more than ten co-developers (UCP). AP2 describes intent and cart mandates in detail (Google Cloud), and MCP dates back to November 2024 (Anthropic).
Most brands will never implement a protocol themselves; Shopify, Stripe or a feed tool does it. Your job is the data those rails carry. The technical walkthrough is in agentic commerce protocols explained.
How AI agents find and pick products
Discovery terms: how a prompt becomes sub-queries, where agents look, and how you measure whether you showed up.
| Term | Definition | Read more |
|---|---|---|
| AI Mode | Google Search's conversational mode, built on Gemini and the Shopping Graph. | AI Mode shopping |
| AI Overviews | AI-generated summaries at the top of Google results. Pages need to be indexed and snippet-eligible, nothing more. | GEO vs SEO |
| Query fan-out | How an AI system splits one prompt into many sub-queries run at once over different subtopics and sources. Google documents it for AI Mode and AI Overviews. | AI Mode shopping |
| Shopping Graph | Google's dataset of product listings and sellers. Google puts it at more than 50 billion listings, with over 2 billion refreshed every hour (Google). | Shopping ads data |
| Retrieval | The step where an agent pulls candidate products from feeds and search indexes before it reasons about them. | See above |
| Grounding | Tying an AI answer to retrieved sources. For products, those sources are your feed and page. | GEO for ecommerce |
| Citation | A link or product card an AI answer attributes to a source. The visible unit of AI visibility. | AI visibility |
| Amazon Rufus | Amazon's shopping assistant. It answers from the Amazon catalog, reviews, community Q&A and the web. | Amazon Rufus |
| AI crawler | A bot that fetches pages for AI products, such as OAI-SearchBot for ChatGPT search. Block it and you leave that surface. | Ecommerce SEO |
| llms.txt | A proposed Markdown file at a site's root that points AI systems to key pages, suggested by Jeremy Howard in September 2024. Google says it needs no such file. | llms.txt |
| AI-referred traffic | Visits that arrive from links in AI assistants. Tracked through referrers such as chatgpt.com. | AI visibility |
| AI visibility | How often and how favorably a brand or product appears in AI answers for a set of prompts. | Visibility tools |
| AI share of voice | Your share of mentions or product placements across tracked AI prompts, compared with competitors. Google announced a version of it for Merchant Center in 2026. | Visibility tools |
Query fan-out is the term to understand first. Google says AI Mode breaks a question into subtopics and searches them simultaneously (Google), and its Search Central docs say the same about AI Overviews (Google Search Central). Each sub-query acts as a filter. A boot with no width attribute drops out at "wide fit".
Two terms are overrated. llms.txt is a proposal (llms.txt) that Google explicitly says you don't need for AI features. AI visibility scores are useful, but tracking a prompt you lose doesn't tell you which attribute lost it.
Product data terms agents actually read
Attributes and identifiers are where recommendations are won. If you only learn one group of terms, learn this one.
| Term | Definition | Read more |
|---|---|---|
| Product feed | A structured file or API stream of products sent to a channel such as Google or ChatGPT. | Feed optimization |
| ChatGPT product feed | The ACP feed OpenAI reads. Nine fields are required: item_id, title, description, url, brand, seller_name, image_url, availability and price. | ChatGPT shopping |
| Google Merchant Center | Google's product data hub, read by Shopping ads, free listings and AI Mode. | Google Merchant Center |
| Conversational attributes | Six Merchant Center attributes announced at Google Marketing Live 2026 to help AI surfaces match products to conversational queries. | Google Merchant Center |
| question_and_answer | The conversational attribute for product FAQs, up to 30 question and answer pairs per product. | FAQ schema |
| related_product | The conversational attribute that links products by ID or GTIN, such as accessories or replacements, up to 30 per item. | GMC checklist |
| item_group_title and variant_option | Conversational attributes for variant families: a shared family title, plus what separates each variant, such as size, color or capacity. | Feed tools |
| GTIN | Global Trade Item Number, the barcode ID (UPC, EAN, JAN, ISBN or ITF-14). Submit it when the manufacturer assigned one; never guess it. | GTINs explained |
| MPN | Manufacturer Part Number. Paired with brand, it identifies products that have no GTIN. | GTINs explained |
| google_product_category | The Google product taxonomy value for an item. Google can assign it automatically, often less precisely than you would. | Taxonomy |
| product_type | Your own category path, such as Women > Shoes > Trail running. Free text that can carry your shoppers' words. | Taxonomy |
| Structured data | Schema.org markup, usually JSON-LD, describing a Product and its Offer. It must match the visible page. | Structured data |
| ProductGroup | The schema.org type that groups variants with variesBy and hasVariant. | Structured data |
| PIM | Product Information Management system, the source of truth for attributes and translations, such as Akeneo or Plytix. | PIM guide |
| Shopify Catalog | Shopify's product data layer that structures product data (categories, variants, attributes) and syndicates it to AI chats. | PIM for Shopify |
The six conversational attributes (question_and_answer, document_link, related_product, item_group_title, variant_option, popularity_rank) are the newest entries here (Productsup, Search Engine Land). The ChatGPT feed fields come from OpenAI's own spec (OpenAI). On GTINs, Google's guidance is blunt: never make one up (Google Merchant Center Help).
One feed item, before and after:
title
BeforeRidge Boot Brown
AfterBrackwater Ridge Mid Men's Waterproof Hiking Boot, Wide Fit (2E), Brown
product_type
BeforeShoes
AfterMen > Footwear > Hiking boots > Waterproof
variant_option
Before(empty)
Aftersize; width
question_and_answer
Before(empty)
AfterDoes it run true to size? Yes. Order half a size up for thick winter socks.
related_product
Before(empty)
AfterBrackwater merino hiking sock (accessory)
The mistake we see most is a product_type one word deep. It's free text, so use it for your customers' vocabulary.
Optimization and measurement terms
How you score readiness and keep data consistent. Six of these are AndromedAI terms.
| Term | Definition | Read more |
|---|---|---|
| AI Readiness Score | AndromedAI's score of how ready a product page is for AI agents, built on the four dimensions below. | Free audit |
| Product Data Completeness | Share of the attributes a category needs that are actually filled. | PDP checklist |
| Keyword Coverage | Whether titles and attributes use the words shoppers search with, backed by real search demand. | Ecommerce SEO |
| Customer Intent Match | Whether the page answers the use cases and questions behind a prompt, such as wide feet, beach wedding or small balcony. | Product descriptions |
| Shopping Metadata | Identifiers, categories, price, availability, shipping and returns data that feeds and schema carry. | GMC checklist |
| Brand Kit | In AndromedAI, the tone of voice, rules, examples, glossary and banned words every generated text must follow. | Creator |
| Data parity | Price and stock agreeing everywhere: page, schema, every feed. Mismatches cause disapprovals and wrong agent answers. | Feed optimization |
Each dimension maps to a different failure. A product can be complete and still use the wrong words, or use the right words and answer no real use case, so the audit reports all four.
How AndromedAI helps
AndromedAI turns this vocabulary into work on your catalog: it scores pages and fills attributes, then publishes the result to your store and feeds.
| Term you care about | What AndromedAI does | Product |
|---|---|---|
| AI Readiness Score | Scores your product pages on the four dimensions for free | AI Readiness Audit |
| Product Data Completeness | Extracts attributes and rewrites titles, descriptions, bullets and FAQ | Optimizer |
| Conversational attributes | Generates Merchant Center conversational attributes from your product data | Optimizer |
| Customer Intent Match | Builds complete pages and category pages around real demand, in 12 languages | Creator and category page optimizer |
clicks from AI chats
Bomboogieto first sales from ChatGPT
Matassaadd-to-cart in two months
Altaforma MilanoIt connects to Shopify, Akeneo and Google Merchant Center, plus the other systems on our integrations page. For the strategy behind all of it, go back to the agentic commerce guide.
FAQ
Agentic commerce is shopping in which AI agents act on a person's behalf. They interpret the need, compare products using their data and, with permission, complete the purchase. It already runs in ChatGPT, Google AI Mode, Gemini and Amazon Rufus.
ACP, the Agentic Commerce Protocol from OpenAI and Stripe, serves discovery and checkout in ChatGPT. UCP, the Universal Commerce Protocol led by Google, serves AI Mode and Gemini from discovery through checkout. Both keep the retailer as merchant of record and start from a product feed, so the same clean product data serves both.
Query fan-out is how an AI system splits one prompt into many related sub-queries and runs them at the same time across subtopics and data sources. Google uses it in AI Mode and AI Overviews. Each sub-query works like a filter on product attributes.
They are six Google Merchant Center attributes announced at Google Marketing Live 2026: question_and_answer, document_link, related_product, item_group_title, variant_option and popularity_rank. They help AI Mode and Gemini match products to conversational queries. The question_and_answer attribute, for example, holds up to 30 question and answer pairs per product.
Agentic commerce is the shift to AI agents searching, comparing and buying for people. Agentic Commerce Optimization (ACO) is the practice of making every product in a catalog found and recommended by those agents. Most of the work is improving product data, from attributes and identifiers to the copy on each product detail page (PDP).
Keep reading
How AI agents search, compare and buy, and what brands should do now
Read nextAgentic Commerce Protocols Explained: ACP, UCP, AP2 and Agent PaymentsThe protocol terms in technical detail
Read nextWhat is Agentic Commerce Optimization (ACO)?The practice that turns these terms into a plan for your catalog
Read nextAEO vs SEO vs GEO vs ACO: The Four Acronyms ExplainedHow the overlapping optimization terms fit together
Sources (19)
- McKinsey: The agentic commerce opportunity
- Stripe: Stripe powers Instant Checkout in ChatGPT and releases the Agentic Commerce Protocol
- OpenAI: Powering product discovery in ChatGPT
- OpenAI: Product feed spec
- Universal Commerce Protocol (ucp.dev)
- Google Cloud: Announcing the Agent Payments Protocol (AP2)
- Anthropic: Introducing the Model Context Protocol
- Visa: Visa introduces Trusted Agent Protocol
- Mastercard: Mastercard unveils Agent Pay
- Shopify: Agentic storefronts (Winter '26 Edition)
- Google: AI Mode in Google Search, updates from Google I/O 2025
- Google: New AI Mode shopping features
- Google Search Central: AI features and your website
- Google Merchant Center Help: GTIN
- Productsup: Google introduces 6 conversational attributes in Merchant Center
- Search Engine Land: Google launches AI performance insights and conversational attributes in Merchant Center
- llms.txt proposal
- Amazon: Amazon Rufus
- Schema.org: Product
Parlano di noi
