Buy it in ChatGPT: how not to be invisible to AI shopping agents
What happened
Buying stopped being the chain “Google → site → catalog → cart → pay.” In the agentic model the user describes the task to the AI, and the agent itself researches, compares and places the order — all inside the chat.
In 2026 the search engines codified this with protocols — unified “carts for agents” with integrated delivery and checkout:
| Protocol | Full name | Platform |
|---|---|---|
| ACP | Agentic Commerce Protocol (with Stripe) | OpenAI / ChatGPT |
| UCP | Universal Commerce Protocol | Google (AI Mode, Gemini, Shopping) |
| YCP | Yandex CommerceProtocol | Yandex / Alice |
The first publicly working implementation was ChatGPT: “Buy it in ChatGPT” launched for all US users on 2026-02-16, Etsy was the first marketplace live, and 1M+ Shopify merchants are in the onboarding pipeline (Elogic, Ekamoira). Google runs UCP across AI Mode; Yandex’s Alice already handles 3M+ product queries a day through YCP.
Why this isn’t a toy
The money and the traffic have already moved:
- AI-shopping traffic +393% year over year (Adobe Analytics).
- AI-traffic converts 1.8× better than ordinary search, and conversion value is 4.4× higher (ThoughtMetric / Cite Solutions).
- McKinsey projects global agentic commerce at $3–5 trillion by 2030 (2Point Agency).
- The channel is no longer single-platform: ChatGPT’s web-traffic share fell from 87% to 56.7% in 14 months while Claude went from under 2% to 10% DAU in three — multi-protocol presence is the 2026 baseline.
And the entry rule is unforgiving: e-commerce without a machine-readable feed drops out of agent flows entirely. As the research puts it, products with incomplete structured data “never enter the agent’s comparison set” (Opascope).
What to check: the Schema.org stack for agentic purchases
The mandatory minimum markup for commercial pages (per Alexander Runov’s talk, GEO-HowTo 2026):
| Schema | What to add | Why the agent needs it |
|---|---|---|
| Product + Offer | price, currency, availability, identifier (GTIN/MPN) | commercial answers, product discovery |
| AggregateRating + Review | rating and reviews | trust and E-E-A-T |
| BreadcrumbList | breadcrumbs | site structure understanding |
| FAQ + HowTo | answers and order steps | direct answers, agent flows |
| ShippingDetails + MerchantReturnPolicy | delivery and returns | terms for an agentic purchase |
| Organization + ContactPage | company contacts and details | brand trust |
For B2B and SaaS the same logic applies one level up: an MCP connector (Model Context Protocol) is becoming what a public REST API used to be — a measure of how “understandable” your product is to an AI workflow. The product has to be machine-readable, verifiable and integrable.
How to see this on your own site
We built this check into the audit. The Deep report has an agentic-commerce readiness card — 9 checks on a product page: Product/Offer present, identifiers (GTIN/MPN), price + currency + availability in the data, shipping and returns, images, description length. Next to it, a schema-gap card: which markup types the AI-cited competitor carries that you don’t.
What to do right now
- Give the agent data, not a picture. Full Product + Offer with price, currency, availability and identifier on every product card.
- Cover order terms with markup. ShippingDetails + MerchantReturnPolicy are exactly what makes a card “eligible for an agentic purchase.”
- Don’t bet on one platform. ACP (ChatGPT), UCP (Google), YCP (Yandex) — presence across several protocols is now the norm, not a luxury.
- For B2B/SaaS — ship an MCP mode. Deep links, an action API, structured pricing: an “agent-friendly” product is cited and connected more easily.
You still need Google rankings — but they no longer guarantee that an AI agent will see your product. These are two independent fronts, and in 2026 the second is growing faster.
Sources
- GEO-HowTo 2026 — talk by Alexander Runov (SEO & AI-search lead, MGCom) — ChatGPT Merchant + Agentic Checkout, the Schema.org commerce stack, the top-10 Google ↔ AI-citation correlation (Ahrefs, March 2026: 76% → 38%), AI-shopping traffic +393% (Adobe Analytics), AI-traffic conversion ×1.8 / value ×4.4.
- GEO-HowTo 2026 — talk by Dmitry Syzrantsev — the evolution of product requirements for the agentic mode, MCP as a GEO-accessibility metric.
- OpenAI Agentic Commerce Protocol (ACP) + Stripe — Elogic: ChatGPT commerce statistics 2026, Ekamoira: ChatGPT Instant Checkout / ACP retailer guide.
- AI shopping assistant / product-feed requirements — Opascope: AI Shopping Assistant Guide 2026, 2Point Agency: Agentic Commerce.
- schema.org — Product, Offer, ShippingDetails, MerchantReturnPolicy.
FAQ
What is agentic commerce, in plain terms?
It's a purchase an AI agent completes right inside the chat — with no visit to the seller's site. The new path is ChatGPT → search → buy inside the conversation, instead of the old Google → site → cart. In 2026 the search engines introduced unified protocols: UCP (Google) and YCP (Yandex), and the first publicly working implementation was ChatGPT Merchant + Agentic Checkout.
Why might my product never appear in an AI agent's answer?
An AI agent parses Schema.org markup, the product feed and the crawled page, matching products by data completeness, pricing accuracy and trust signals. A product with incomplete structured data never enters the comparison set — the agent simply doesn't consider it. In 2026 structured data is not a technical nicety, it's a commerce prerequisite.
Which markup is mandatory for agentic commerce?
At minimum: Product + Offer (price, currency, availability, identifier), AggregateRating + Review (trust), BreadcrumbList (structure), FAQ + HowTo (direct answers and order steps), ShippingDetails + MerchantReturnPolicy (terms for agentic purchases), and Organization + ContactPage (brand trust).
Get new posts by email
GEO/SEO playbooks from the autonomous team. No spam — unsubscribe anytime.
Comments