What is agentic commerce? A guide for online stores

Agentic commerce is shopping where an AI agent such as ChatGPT, Gemini or Copilot finds, compares and sometimes buys products on a shopper's behalf. For an online store, that makes product data the new storefront: the agent picks from your feed, catalog and page markup before a shopper ever visits. This guide explains how AI shopping agents work, what Shopify now does by default, and gives a readiness checklist for data, feeds and checkout.

By Kanika Mathur, Head of Service Delivery

Reviewed by Kritika SaxenaPublished Sep 22, 2026Updated Sep 22, 202611 min read

Ecommerce

Key takeaways

The short version

Agentic commerce puts an AI agent between the shopper and your store. The agent reads your product data, matches it to the shopper's constraints, then sends them to a checkout or completes one in the AI surface.
The traffic is real and converts. AI-referred visits to US retail sites grew 393% year over year in the first quarter of 2026 and converted 42% better than non-AI visits in March 2026 (Adobe). Adobe's August 2026 update put the gap at 60% (Digital Commerce 360).
Product pages are the weak spot. Adobe scored US retail product pages at 66% machine-readable on average, the lowest of the main page types.
Shopify does much of the plumbing by default. Agentic storefronts connect eligible stores to ChatGPT, Google AI Mode and Gemini, Copilot and Meta, and ChatGPT sends shoppers to your own checkout (Shopify help center).
Checkout programs are still moving. OpenAI pivoted away from in-chat Instant Checkout in March 2026 (CNBC), so invest first in the data layer every surface reads.

What is agentic commerce?

Agentic commerce is shopping where an AI agent does part of the buying work for a person. The agent reads the request, searches product catalogs, compares options against the shopper's constraints, and then either sends the shopper to a checkout or completes checkout inside the AI surface. For an online store, it means your product data, not your homepage, is often the first thing a buyer sees.

The shift already shows up in traffic. Adobe, which measures more than 1 trillion visits to US retail sites, reports AI-referred traffic growing fast and converting better than other visits, a reversal from March 2025, when it converted 38% worse. In Adobe's companion survey, 39% of consumers said they had used AI for online shopping.¹¹

Year-over-year growth in AI-referred traffic to US retail sites, January to March 2026

Adobe Digital Insights, 2026

How much better AI-referred retail visits converted than non-AI visits in March 2026

Adobe Digital Insights, 2026

Average machine-readability score of US retail product pages, the lowest of the main page types

Adobe AI Content Visibility Checker, 2026

Two things separate this from ordinary search, and both follow from how AI agents plan and act. First, the shopper describes a need instead of typing keywords. CNBC's report on the March 2026 ChatGPT shopping update describes people uploading images or stating a budget, preferences and other constraints.⁸ Second, the agent reads structured data before it reads anything else. Shopify's own announcement says these AI surfaces favor structured data with clean attributes and real-time accuracy.⁴ A beautiful product page with thin attributes can lose to a plain one with complete data.

How AI shopping agents find and buy products

An AI shopping agent works in three steps. It gathers product data from platform catalogs, merchant feeds and crawled web pages; it matches products to the shopper's stated needs; then it hands off to a checkout. Where checkout happens depends on the surface: ChatGPT sends Shopify shoppers to your own checkout, while Google AI Mode, Gemini, Copilot and Meta can run checkout inside the AI channel where it is available and switched on.

A laptop showing a blurred grid of product thumbnails on a desk by a window, a second monitor with a blurred storefront grid, a plain cardboard box, a handheld barcode scanner, a mug and a warm lamp, no people — The agent reads your catalog data long before a shopper sees your storefront.

Step 1: where agents get product data

Platform catalogs. Shopify Catalog syndicates each product's title, description, options, images, price and availability to AI channels and keeps inventory and pricing updated continuously.²
Direct feeds. OpenAI accepts structured product feeds from approved partners and recommends a full feed once a day by file upload, with updates through the day by API.⁵ Google reads product data from Merchant Center, which it calls its central hub.⁹
The open web. Crawlers still matter. OpenAI states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers.⁷

Step 2: matching the request

Shopify's example is a shopper asking for navy running shorts; the agent searches and shows a selection that fits.¹ Real prompts add clauses ("waterproof, lightweight, arrives before the weekend"), and every clause is a filter. If your material, size range or delivery promise is missing from the data, the agent cannot match you on it.

Step 3: the checkout hand-off

Where the sale closes for a Shopify store on each AI surface, at the time of writing
Surface	How products get in	Where checkout happens	Customer relationship
ChatGPT	Shopify Catalog, approved-partner feed, crawled pages	Your store checkout, in an in-app browser or new tab	Yours
Google AI Mode and Gemini	Merchant Center, or the Google & YouTube channel on Shopify	On Google with Google Pay, for select merchants using UCP	You stay seller of record
Microsoft Copilot	Shopify Catalog	Shopify-powered direct checkout, if activated	Yours
Meta surfaces	Facebook and Instagram by Meta channel	Shopify-powered direct checkout, if activated	Yours

Sources: Shopify¹² and Google.⁹

The protocols underneath

Agentic Commerce Protocol (ACP). An open standard, developed by OpenAI and Stripe, for commerce flows between buyers, AI agents and businesses. OpenAI's integration path starts with a structured product feed; onboarding feeds is limited to approved partners who apply through chatgpt.com/merchants.⁵
Universal Commerce Protocol (UCP). Google describes it as an open standard covering discovery, buying and post-purchase support, compatible with A2A, AP2 and MCP. Shopify says it co-developed UCP with Google.⁹⁴
Trusted Agent Protocol. Visa's framework, built with Cloudflare, uses agent-specific cryptographic signatures so merchants can tell a legitimate shopping agent from a malicious bot.¹²

What happened to in-chat checkout in ChatGPT

OpenAI announced Instant Checkout in 2025, letting users buy select items from retailers such as Etsy, Walmart and Shopify merchants inside ChatGPT. By March 2026 it had pivoted. CNBC reports that analysts said OpenAI struggled to onboard merchants, show accurate product data, and support multi-item carts or loyalty memberships.⁸

"We're allowing merchants to use their own checkout experiences."

OpenAI, quoted by CNBC, March 2026

The lesson: the checkout layer is still moving, while the data layer is what every surface reads. Start there.

Agentic commerce readiness checklist for your store

Readiness comes down to four layers: complete and accurate product data, product pages that machines can read, a feed or catalog connection to each AI surface, and published policies plus a checkout an agent can hand a shopper into. Work through them in that order. A checkout integration earns nothing if the agent never selects your product in the first place.

Start with the page layer, where Adobe's data shows stores are weakest. Adobe's AI Content Visibility Checker scores how much of a page a machine can read; product pages averaged 66%, so about a third of their content was not machine-readable.¹¹

Product pages are the least machine-readable pages on US retail sites

Average AI Content Visibility score by page type. 100% means a machine can read all of the page's content.

Data behind this chart
Page type	Average score
Product pages	66%
Store locator	73%
Category pages	74%
Homepages	75%
Customer service and help center	79%
Returns and exchanges	82%

Source: Adobe, AI traffic grows but retail sites lag in AI search visibility (April 2026).

Then work the checklist:

Agentic commerce readiness checklist
Layer	Check	Why it matters	Source
Data	One row per purchasable variant, each with its own ID, title, price, stock and image	OpenAI's feed requires nine fields on every variant row	OpenAI feed spec
Data	Titles under about 150 characters, factual descriptions under 5,000	OpenAI's guidance for concise, plain-text copy	OpenAI feed spec
Data	Availability always set explicitly	A blank or unrecognized value rejects the row	OpenAI feed spec
Data	Real GTINs where assigned, never invented	Both OpenAI and Google ask for genuine global identifiers	OpenAI, Google
Data	Attributes stored where the catalog can read them, or mapped	Catalog Mapping exists for metafields and title delimiters	Shopify
Data	Legal disclosures in the first 6,000 characters of the description	Shopify's rule for the ChatGPT channel	Shopify
Pages	schema.org Product markup with name, image and an Offer priced above zero with a currency	Required for Google merchant listings	Google
Pages	Offer markup adds availability, return policy and shipping details	Recommended properties that answer shopper constraints	Google
Pages	Key product facts in page text, not only in images	Product pages are the least machine-readable page type	Adobe
Feeds	Merchant Center feed current and error-free	Merchant Center is Google's hub for AI Mode and Gemini	Google
Feeds	Outside Shopify, apply for ChatGPT feed access; send a full feed daily plus API updates	Feed onboarding is limited to approved partners	OpenAI
Access	robots.txt allows OAI-SearchBot; decide GPTBot separately	Blocked sites are not shown in ChatGPT search answers; changes take about 24 hours	OpenAI
Policies	Terms of service, privacy policy and return policy published	Required for Shopify's ChatGPT channel	Shopify
Policies	Return window and shipping price filled in the feed	A blank shipping price reads as unknown, not free	OpenAI feed spec
Checkout	Bot protection tested against legitimate shopping agents	Visa warns bot detection can block real agentic purchases	Visa

Sources: OpenAI⁵⁶⁷, Shopify²³, Google⁹¹⁰, Adobe¹¹ and Visa.¹² The content side of the same problem is generative engine optimization, and agent-ready catalogs sit alongside the other ways stores use AI.

How do I get my products shown in ChatGPT?

Three routes, and most stores should use two:

On Shopify: nothing to install. Products are discoverable through Shopify Catalog by default, as long as you sell to US customers, your products are Catalog-eligible, you accept the supplemental terms, and your terms, privacy and return policies are complete.³
Not on Shopify: apply to OpenAI for product feed access, or use Shopify's Agentic plan to add products to Shopify Catalog while keeping your existing store.⁵¹
Everyone: allow OAI-SearchBot in robots.txt and make product facts readable on the page, because crawled pages remain a discovery path.⁷²

Getting recommended, not merely listed, comes from the data rows above. Products in prohibited categories, such as age-restricted goods or weapons, are not surfaced at all.⁵³

Shopify specifics: agentic storefronts and Catalog

On Shopify, agentic storefronts is active by default for eligible stores and connects your products to ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Meta, mostly through Shopify Catalog, managed under Sales channels, then Agentic, in the admin. The work is mostly data quality, policies, and choosing which channels and products to include.

In March 2026 Shopify opened ChatGPT selling to millions of merchants through Agentic Storefronts, and said thousands of merchants were already selling in Copilot.⁴ What to review, channel by channel:

ChatGPT is a referrer, not a checkout. Shoppers buy on your own store checkout in an in-app browser, so your branding, payment methods and checkout customizations carry over. Shopify lists no fees beyond standard payment processing.³
Google, Copilot and Meta can check out in the channel. With direct checkout activated, customers complete a Shopify-powered checkout without leaving the AI channel; on Google it is still rolling out. Google receives your products through the Google & YouTube channel or Merchant Center.¹²
Custom data needs mapping. If titles, categories or specs live in metafields, metaobjects or tag prefixes, use Shopify Catalog Mapping so agents get the right values.²
Agent discovery files come free. Every store serves /agents.md, /llms.txt and /llms-full.txt with store name, sitemap and policies, editable through Liquid templates.²
B2B pricing can leak. Agentic storefronts are direct-to-consumer only. Shopify excludes B2B catalogs it can detect, but custom setups that hide prices until login may still show up.²

Shopify controls worth knowing before you change anything
Action	Effect
Turn off Catalog access for ChatGPT	Can take up to 7 days to stop sharing; crawled pages may still appear
Set a product to Unlisted	Hides it from AI channels, but also from sitemaps, Google search and store search
Block a crawler in robots.txt.liquid	Advisory only, and does not affect Catalog syndication
Use the Agentic plan (non-Shopify stores)	Keeps your existing store; direct checkout is off by default

Sources: Shopify help center.¹²³ ChatGPT shoppers land on your own store, so speed still decides the sale; our guide to speeding up a Shopify store covers that. Weighing platforms? See how Shopify, WooCommerce and Magento compare.

How to measure agentic commerce

Treat agentic commerce as its own channel: segment AI referrals, track orders by AI channel, and compare conversion and engagement with your non-AI baseline. Beyond the conversion benchmark above, Adobe found AI-referred visits in March 2026 engaged 12% more and stayed 48% longer than non-AI visits. Your own numbers matter more than any benchmark.

What to track, and where
Metric	Where to get it	What it tells you
Orders and revenue by AI channel	Shopify admin channel or referrer attribution	Whether agents are closing sales, not just sending visits
AI referral sessions	Your analytics, grouped by AI assistant referrer	Volume trend and which assistants send buyers
Conversion and engagement versus non-AI	Your analytics	Whether AI visitors behave like Adobe's benchmark
Rejected or incomplete feed rows	Feed or Merchant Center diagnostics	Products the agent cannot show at all
Product page readability	A machine-readability audit of top product pages	Content the agent cannot see on your pages
Share of answers	A fixed set of buyer prompts, run monthly	Whether you are named for the prompts that matter

Attribution sources: Shopify.¹⁴ In March 2026 AI-referred visitors also browsed 13% more pages per visit.¹¹ Adobe's August 2026 update, reported by Digital Commerce 360, put the conversion gap at 60% and AI traffic growth at 62% year over year in July 2026.¹³ One caution: a shopper who researches in an assistant and later types your URL will not show up as an AI referral.

The prompt test is cheap and telling. Write ten prompts in your buyers' words, run them monthly in ChatGPT, Gemini and Copilot, and record whether you appear and at what price. Wrong prices or missing variants point straight to a feed problem.

Risks of agentic commerce for online stores

The main risks are wrong data shown at scale, protocol churn, bot defenses that block legitimate agents, and less visibility into who the customer is. Most controls are the data and policy work the readiness checklist already asks for.

Risks and the control for each
Risk	What it looks like	Control
Stale price or stock	An assistant quotes a price or size you no longer have	Full daily feed plus intraday updates, or a platform catalog that syncs continuously
Protocol churn	A checkout program changes shape, as ChatGPT's did in March 2026	Invest in data first; keep checkout on a platform you control
Agents blocked by bot defenses	Legitimate agent purchases fail at your firewall	Test bot rules; watch signed-agent standards such as Visa's protocol
Less customer visibility	You see the order but not the person behind the agent	Stay merchant of record; offer account linking and loyalty where supported
Wrong products exposed	B2B prices or restricted items appear in consumer answers	Use native B2B catalogs; review prohibited categories
Slow or partial opt-out	Products keep appearing after you remove Shopify Catalog access	Allow up to 7 days for Catalog sharing to stop; crawled pages persist

Sources: OpenAI onboarding⁵, Shopify²³⁴, CNBC⁸, Google⁹ and Visa¹². Google's UCP checkout and Shopify's agentic storefronts both keep you as seller of record, so the order data stays yours.

If your catalog sits in custom fields, a headless stack or several platforms, the data work is the real project, and our ecommerce AI development team can help you scope it.

Frequently asked

Agentic commerce questions

What is agentic commerce?

Agentic commerce is shopping where an AI agent does part of the buying work for a person. The agent interprets a request, searches product catalogs and feeds, compares options against the shopper's budget and preferences, and then either sends the shopper to a merchant checkout or completes checkout inside the AI surface. For a store, it means structured product data becomes the first thing a buyer sees, often before your own website.

How do AI agents buy products?

They gather product data from platform catalogs such as Shopify Catalog, merchant feeds such as Google Merchant Center, and crawled pages, then match products to the shopper's request. The purchase happens in one of two places. ChatGPT sends Shopify shoppers to the merchant's own checkout in an in-app browser. Google AI Mode, Gemini, Copilot and Meta can run checkout inside the channel for merchants who activate it, where available; on Google it is limited to select merchants for now. The merchant stays the seller of record.

How do I get my products shown in ChatGPT?

On Shopify, products are discoverable through Shopify Catalog by default if you sell to US customers, your products are Catalog-eligible, you accept the supplemental terms and your terms, privacy and return policies are complete. Off Shopify, apply to OpenAI for product feed access through chatgpt.com/merchants or use Shopify's Agentic plan. In every case, allow OAI-SearchBot in robots.txt and keep product facts readable on the page.

How do I get my products recommended by ChatGPT shopping?

Being listed is not the same as being picked. The agent can only match you on constraints your data answers, so fill in every variant with its own price, stock and image, keep titles concise and descriptions factual, set availability explicitly, add real GTINs, and publish return and shipping terms. Then test monthly with the prompts your buyers actually use and fix whatever price, variant or attribute comes back wrong.

Do I need to implement ACP or UCP myself?

Usually not on a hosted platform. Shopify connects stores to ChatGPT, Google AI Mode and Gemini, Copilot and Meta through agentic storefronts. You would implement the protocols directly if you run a custom or headless stack and want checkout on Google's surfaces, which currently requires UCP and select-merchant approval, or a deeper ChatGPT integration beyond a product feed. Either way, clean product data comes first.

Can I stop AI agents from listing my products?

Partly. On Shopify you can remove Shopify Catalog access for ChatGPT or Copilot. If you turn off Catalog access, for example for ChatGPT, it can take up to 7 days to stop sharing, and products can still appear through crawled pages. Setting a product to Unlisted hides it from AI channels but also from sitemaps, Google search and your store search. Blocking crawlers in robots.txt is advisory and does not stop Shopify Catalog syndication.

Is agentic commerce traffic worth the effort?

On Adobe's data it is growing fast and converting well. Traffic from AI sources to US retail sites grew 393% year over year in the first quarter of 2026, and in March 2026 those visits converted 42% better than non-AI traffic, after converting 38% worse a year earlier. Adobe's August 2026 update put the gap at 60%. Treat that as a sector benchmark and measure your own AI channel orders against your non-AI baseline.

Kanika Mathur

Head of Service Delivery, Resourcifi

Kanika Mathur is Head of Service Delivery at Resourcifi. She leads the engineering pods that scope, build and run client software, from mobile apps to AI systems, and she reviews the process and figures in our engineering guides for accuracy.

Resourcifi on LinkedIn →

Kritika Saxena

Reviewer. Head of Growth and Strategy, Resourcifi

Kritika Saxena is Head of Growth and Strategy at Resourcifi. She owns growth, marketing and AI search strategy for the company and its clients, and she reviews the marketing guides we publish for accuracy.

Resourcifi on LinkedIn →

Sources

Shopify Help Center, Shopify agentic storefronts (AI channels (ChatGPT, Google AI Mode and Gemini, Copilot, Meta), active by default, ChatGPT as referrer with in-app browser checkout, direct checkout on other channels, order attribution in admin, customer relationship ownership, Agentic plan direct checkout off by default, navy running shorts example).
Shopify Help Center, Shopify Catalog and product discovery for agentic storefronts (Catalog attributes and continuous updates, discovery via crawling and feeds, Catalog Mapping, robots.txt advisory and open-web only, agents.md and llms.txt files, Unlisted side effects, D2C only and B2B leak risk).
Shopify Help Center, Selling on ChatGPT (ChatGPT eligibility (US customers, Catalog-eligible, terms, three store policies), no fees beyond payment processing, legal disclosures in first 6,000 characters, 7-day opt-out lag, checkout customizations carried over).
Shopify, Millions of merchants can sell in AI chats (March 2026 rollout to millions of merchants, Agentic plan, merchant of record, ChatGPT referral attribution, thousands of merchants selling in Copilot, UCP co-developed with Google, AI surfaces favor structured data).
OpenAI, Agentic Commerce: Get Started (product feed onboarding limited to approved partners via chatgpt.com/merchants, full daily feed plus API updates, prohibited product categories; ACP definition (the get-started doc links to agenticcommerce.dev, which describes ACP as an open standard developed by Stripe and OpenAI)).
OpenAI, Agentic Commerce product feed spec: Products (nine required fields per variant row, 150 and 5,000 character guidance, blank availability rejects the row, GTIN rules, return and shipping fields, blank shipping price means unknown).
OpenAI, Overview of OpenAI Crawlers (OAI-SearchBot controls appearance in ChatGPT search answers, GPTBot set independently, about 24 hours for robots.txt changes).
CNBC, OpenAI revamps shopping experience in ChatGPT after struggling with Instant Checkout (Instant Checkout history and March 2026 pivot to discovery, onboarding and data accuracy struggles, multi-item carts and loyalty, OpenAI quote on merchants' own checkout, shopper criteria and image search).
Google Merchant Center Help, About the Universal Commerce Protocol (UCP) and UCP-powered checkout feature on Google (UCP definition and compatibility with A2A, AP2 and MCP, select merchants, seller of record, Google Pay checkout in AI Mode and Gemini, Merchant Center as the product data hub).
Google Search Central, Merchant listing (Product, Offer) structured data (required schema.org Product properties (name, image, offers with price above zero and currency), recommended availability, return policy, shipping details and GTIN).
Adobe, AI traffic grows but retail sites lag in AI search visibility (393% Q1 2026 AI traffic growth, 42% better conversion in March 2026 versus 38% worse in March 2025, 39% of consumers used AI to shop, 12% engagement, 48% longer, 13% more pages, readability scores by page type (product pages 66%), 82.5% and 54.2% homepage range; corroborated by TechCrunch, 16 April 2026).
Visa, Visa Introduces Trusted Agent Protocol: An Ecosystem-Led Framework for AI Commerce (bot detection mistakenly blocking legitimate agentic transactions, visibility into the consumer behind the agent, agent-specific cryptographic signatures, built with Cloudflare).
Digital Commerce 360, Adobe: AI-referral traffic spending, converting more than others (Adobe Analytics update of August 2026: AI-referred retail visits converting 60% better than non-AI traffic, AI-referral traffic up 62% year over year in July 2026).