# Agentic Commerce for Shopify: From Discovery to a Verified Order

Agentic commerce lets a shopping assistant help someone discover products, build a cart or complete a purchase. For a Shopify merchant, the useful decision is which buying path your store can support reliably. Separate product discovery from checkout availability, then verify the order, fulfillment, refund and measurement records. A product appearing in an AI answer does not establish that the customer can buy it correctly.

Canonical URL: https://www.martechdevs.com/blog/agentic-commerce-shopify

By [Jafrin Akter Alo](https://www.martechdevs.com/authors/jafrin-akter-alo)

Data Engineer and Co-founder at [Martechdevs](https://www.martechdevs.com) and [Growth and Analytics Partners](https://www.growthandanalytics.com)

[LinkedIn](https://linkedin.com/in/jafrinalo)

Last updated: 2026-10-25

Next step: [GEO vs AEO: What Changes in AI Search and What to Measure](https://www.martechdevs.com/blog/geo-vs-aeo)

A shopper asks for a lightweight carry-on bag that fits a particular airline. Your product appears, but the recommended variant is unavailable, the delivery promise is wrong, or a required personalization field disappears during checkout. More exposure makes each of those failures more expensive.

Our starting point is the purchase contract: the exact item, price, delivery conditions, buyer confirmation and resulting order. The AI interface is another route into that contract. The [ecommerce stack guide](https://www.martechdevs.com/blog/ecommerce-martech-stack) covers the wider architecture; this guide focuses on making the AI shopping route dependable.

Martechdevs can help connect the catalog, order and measurement checks through an [analytics and tracking implementation](https://www.martechdevs.com/analytics-tracking).

## Where do ChatGPT and Google shoppers actually pay?

Shopify shoppers referred by ChatGPT complete payment in the merchant's online-store checkout. Google AI Mode and Gemini can offer a Shopify-powered direct checkout for eligible stores and shoppers, subject to rollout and channel settings.

That distinction changes the implementation, customer experience and reporting. Use the channel-specific path rather than assuming every AI recommendation includes native payment. [Shopify's agentic storefronts overview](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts).

| Channel | Product-discovery route | Purchase route | Eligibility boundary to verify |
| --- | --- | --- | --- |
| ChatGPT | Shopify Catalog for eligible products | Store checkout in an in-app browser or a new browser tab | Store sells to US customers; the business may be based elsewhere |
| Google AI Mode and Gemini | Google Merchant Center, directly or through Google & YouTube | Direct checkout when available and enabled; otherwise store checkout | Store may be based anywhere; eligible shoppers are in the US, Canada or Australia |
| Microsoft Copilot | Shopify Catalog | Supported direct checkout or store checkout | Store sells to US customers; direct checkout is shown to US shoppers |

These are channel requirements, not a promise that your account or every product is approved. Google direct checkout is rolling out to shoppers in those three countries. Read the actual status in your admin and the relevant [ChatGPT](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/chatgpt), [Google](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/google) and [Copilot](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/microsoft) requirements.

### What happened to ChatGPT Instant Checkout?

OpenAI says its initial Instant Checkout did not provide the flexibility it wanted, so it shifted emphasis toward product discovery and merchant-owned checkout experiences. Its Shopify integration uses Shopify Catalog for product data. That supports the practical recommendation to prepare your existing checkout, rather than waiting for a universal in-chat replacement. [OpenAI's product-discovery announcement](https://openai.com/index/powering-product-discovery-in-chatgpt/).

There is a separate approved-partner boundary: OpenAI's developer guidance still describes Instant Checkout for approved partners, with merchant systems validating and accepting the order. Neither that developer route nor an earlier launch announcement establishes that an ordinary Shopify store has native ChatGPT checkout enabled. [OpenAI's commerce concepts](https://developers.openai.com/commerce/guides/key-concepts).

Shopify provides no ChatGPT direct-checkout toggle. Its ChatGPT page states that normal store checkout customizations and payment methods apply, with standard payment-processing fees and no additional ChatGPT selling fee. Treat that as the Shopify channel's commercial scope, not a price promise for a custom app or another platform. [Shopify's ChatGPT checkout and fees](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/chatgpt).

## What do UCP and ACP change, and do you need to build either?

Universal Commerce Protocol (UCP) and Agentic Commerce Protocol (ACP) define ways for commerce systems and agents to exchange information. An existing Shopify merchant should first use the supported channel controls; building a protocol integration is a separate engineering decision.

UCP describes capabilities such as catalog search, carts, checkout and order management. Businesses publish supported capabilities and versions; platforms negotiate the capabilities they can use. A capability in the standard is therefore not evidence that a particular channel supports it. [UCP core concepts](https://ucp.dev/documentation/core-concepts/).

ACP supports OpenAI commerce experiences, including product discovery. For an eligible Shopify store, the Catalog connection already supplies product data to ChatGPT. A second feed or custom shopping agent needs a specific unmet requirement before it earns a place in the stack. [OpenAI's discovery integration](https://openai.com/index/powering-product-discovery-in-chatgpt/).

Google explicitly distinguishes a native-checkout integration from a cart handoff to the merchant's website, and warns that its surfaces do not support every UCP feature. Its direct integration requires approval before going live. [Google's UCP implementation paths](https://developers.google.com/merchant/ucp/guides/overview).

For merchants using Google's UCP integration hub, access is gradually available in the US and checkout is limited to selected merchants. Google directs businesses using a platform or service provider to discuss that provider's integration. That hub's US early-access scope is separate from Shopify's channel requirements. An eligibility attribute in a feed is not a shortcut around onboarding. [Google's integration-hub eligibility](https://developers.google.com/merchant/ucp/guides/merchant-center/overview).

Ask a proposed implementation partner three questions:

1. **Which missing customer action requires custom work?** Name the unsupported journey rather than buying a generic “AI-ready” package.
2. **Which system owns the transaction?** Identify the checkout, payment processor, order record and post-purchase updates.
3. **What proves this particular channel supports the requirement?** Request the applicable account access, product limits and acceptance results.

This keeps the [industry-specific stack decision](https://www.martechdevs.com/blog/martech-stack-by-industry) grounded in the buying journey. Protocol adoption alone does not improve a wrong product description or an unfulfillable order.

Shopify's separate Agentic plan serves merchants retaining another ecommerce platform. It requires external product URLs and additional setup; direct checkout starts deactivated. ChatGPT order history for that path stays in the external store's analytics, not Shopify. Do not apply that setup to an existing Shopify online store by default. [Agentic plan setup](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/agentic-plan-setup).

## Which eligibility and product-data checks come first?

Check store eligibility, product eligibility and channel eligibility separately. A valid product feed cannot fix an ineligible store, and an eligible store can still contain products that the channel cannot sell.

In Shopify, inspect **Sales channels → Agentic → Sources → Shopify Catalog**. If the store is ineligible, use the displayed reason before editing product copy. Catalog requirements include a supported plan, a non-private storefront, identifiable published product URLs, a title, an image and a positive price. Account standing and excluded product categories also matter. The separate Agentic plan has an exception for merchants retaining an external storefront. [Shopify Catalog requirements](https://help.shopify.com/en/manual/shopify-catalog/requirements).

Complete the store's terms, privacy and return/refund policies and review the supplemental channel terms. Do not remove a legitimate B2B access restriction simply to make a wholesale catalog discoverable to consumers. Shopify's agentic storefronts support direct-to-consumer sales; the product's intended buyer remains part of the design. [Shopify discovery and B2B boundaries](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/products).

### Which product facts should the team reconcile?

Choose a small set of products that exposes your actual complexity: multiple sizes, a sale price, low inventory, an awkward shipping destination and an item with a meaningful restriction. Compare the same variant across the source system, feed, landing page and checkout.

Use this merchant-owned inspection worksheet. It is an acceptance method, not a list of universally required feed fields.

| Product fact | Evidence to compare | Failure worth catching | Owner |
| --- | --- | --- | --- |
| Product and variant identity | Stable item ID, selected options and landing URL | A blue medium bag opens the black large variant | Catalog operations |
| Price and currency | Current offer, sale conditions and checkout amount | Expired promotion or wrong market currency | Merchandising |
| Availability | Sellable inventory and actual order acceptance | An unavailable size is still offered | Inventory operations |
| Product suitability | Dimensions, material, compatibility and limitations | Marketing copy implies a use the product cannot support | Product owner |
| Delivery and returns | Destination, shipping charge and applicable policy | Attractive delivery promise disappears at checkout | Operations |
| Images and descriptions | Actual variant images and visible product facts | Generic imagery conceals a material difference | Content owner |

Google's product specification calls for stable IDs, accurate variant attributes and consistency with the landing page. Missing or conflicting values can cause disapprovals or incorrect displays. Use Merchant Center's issue details to find the actual failed attribute instead of adding repeated keywords. [Google's product data specification](https://support.google.com/merchants/answer/7052112?hl=en).

For example, a hypothetical “Cabin Bag 20” title leaves the unit, dimensions and permitted use unclear. A useful description states the bag's actual dimensions, weight, capacity and material, then explains that airline limits must be checked for the intended ticket. Do not invent airline approval because an AI shopping prompt asks for it.

### What if the useful details live in metafields?

Map the field that contains the real product information. Shopify Catalog Mapping can source title, description and category from product attributes, metafields or metaobject references. It also supports grouping separate products into combined listings and choosing whether to send eligible compare-at prices. Mapping changes can take time to process. [Shopify Catalog Mapping](https://help.shopify.com/en/manual/shopify-catalog/mapping).

Inspect both a populated and an empty metafield before switching the source for a whole catalog. Check a legitimate variant group and two different products that happen to share a similar name. A cleaner preview is not a successful mapping if it merges products customers should compare separately.

Keep the Google feed mapping separate from the Catalog mapping. Record the destination and owner for each transformation. A fix visible in one preview does not prove that the other feed changed.

### Do schema and AI discovery files replace a product feed?

No. Product-page structured data and discovery files serve different jobs from channel feeds and checkout eligibility. Keep visible offers and `Product`/`Offer` markup consistent, then validate the markup with Google's Rich Results Test. Valid markup can make a page eligible for merchant listing experiences; it does not guarantee a placement. [Google's merchant listing guidance](https://developers.google.com/search/docs/appearance/structured-data/merchant-listing).

Shopify automatically serves `/agents.md`, `/llms.txt` and `/llms-full.txt` with store discovery information. These do not replace Shopify Catalog. Inspect the existing output before buying an app or overriding a template, and keep policy links accurate. [Shopify's discovery files](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/products).

## How do you activate channels without breaking the buying journey?

Activate the purchase path only after its required product, checkout and operating behavior passes a controlled test. Discovery can be useful while a more restrictive direct-checkout path remains unsuitable.

For a Shopify online store, use this sequence:

1. **Record the current channel state.** Capture eligibility, discovery access, direct-checkout availability and the person responsible for changes.
2. **Choose who manages future enrollment.** Shopify's default “Allow Shopify to manage for me” includes supported direct checkout and future channel enrollment. Turn it off when you need individual controls, then save the intended settings.
3. **Confirm the relevant product source.** For Google, verify the Merchant Center feed and shipping configuration; for Catalog channels, inspect Catalog eligibility and mapping.
4. **Test the actual buying path.** Use representative products, buyer locations, delivery options and discounts. Record expected behavior before testing.
5. **Approve the route and rollback action.** Keep the channel on the store-checkout path when essential requirements do not pass.

The controls live under **Sales channels → Agentic**. Access requires the store owner or the relevant Products-view and Agentic staff permissions. [Managing agentic storefronts](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/agentic-home).

Google direct checkout excludes subscriptions, bundles, customizable products and B2B-only products. Local delivery and pickup are also unsupported, and some checkout blocks may not appear. A store relying on a required input must verify its collection rather than assume the online checkout's UI transfers intact. [Google checkout limitations](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/google).

Copilot has its own corresponding checkout restrictions, including unsupported product types and potentially missing customization blocks. Review that channel separately instead of carrying Google's result across by analogy. [Copilot checkout limitations](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/microsoft).

A hypothetical engraving store illustrates the decision. If the destination checkout cannot collect the engraving text and preserve it on the order, direct checkout has not passed. Keep a supported store-checkout route, verify the personalization there and reconsider direct checkout only when the missing behavior is supported and tested.

The same logic applies to a required buyer acknowledgement, delivery constraint or eligibility rule. An extra upsell is optional; information required to fulfill the order is not.

![Select the channel, product and buyer market, then repair failed discovery eligibility or data checks. Supported and suitable direct checkout routes to a direct purchase test; other cases test the merchant checkout handoff. Both routes must pass purchase and post-purchase checks before approval. Failed checks hold expansion and return to route selection and eligibility.](https://www.martechdevs.com/assets/blog/agentic-commerce-shopify.svg)

Diagram connections:

- Select channel product and buyer market to Discovery eligibility and data pass?. 
- Discovery eligibility and data pass? to Repair eligibility or product data: No. 
- Repair eligibility or product data to Select channel product and buyer market. 
- Discovery eligibility and data pass? to Direct checkout supported and suitable?: Yes. 
- Direct checkout supported and suitable? to Test merchant checkout handoff: No. 
- Direct checkout supported and suitable? to Test direct checkout and order creation: Yes. 
- Test merchant checkout handoff to Purchase and post-purchase checks pass?. 
- Test direct checkout and order creation to Purchase and post-purchase checks pass?. 
- Purchase and post-purchase checks pass? to Hold expansion and repair the failed route: No. 
- Hold expansion and repair the failed route to Select channel product and buyer market. 
- Purchase and post-purchase checks pass? to Approve route and reconcile observed orders: Yes

Direct-checkout support and suitability are both required for the Yes branch. Passing discovery sends the product into a buying-path test, not straight into release. Any repair that changes the product, market or route repeats the relevant eligibility checks.

## Who owns fulfillment, refunds and customer permissions?

The merchant retains responsibility for the customer relationship and post-purchase experience. Shopify records AI-channel orders with channel or referrer attribution; the channel's interface does not replace the merchant's operational record. [Shopify's order ownership](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts).

Check the full handoff: accepted order, payment state, fulfillment instructions, dispatch, tracking and customer support. A successful payment with no fulfillable order is a failure. So is an order that reaches the warehouse without the option the buyer selected.

Refunds and physical returns are separate operations. Shopify allows full or partial refunds, including refunds without a return. If returned-item reasons matter, create the return before refunding the items; a refund-only note is not a recorded return reason. Inspect the effect on payment, inventory and customer notification independently. [Shopify refund behavior](https://help.shopify.com/en/manual/fulfillment/managing-orders/refunding-orders).

A custom Google UCP integration also has a post-purchase delivery obligation: order updates must reach Google. Its guide distinguishes fulfillment events from financial adjustments and describes retry handling with an unchanged idempotency key for the same event. That is work for the custom integration owner, not an instruction for every Shopify merchant to build a second webhook pipeline. [Google's order lifecycle](https://developers.google.com/merchant/ucp/guides/overview/order-lifecycle).

For customer data, review what the selected route shares. Shopify says direct-checkout channels receive relevant order details, while they do not receive the store's full customer database or other-channel order history. [Agentic storefronts data sharing](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/data-privacy).

Keep transactional service and promotional eligibility separate in your operating rules. The email address needed for a delivery update does not, by itself, establish permission for every marketing destination. Preserve the actual preference evidence and use a [customer preference centre](https://www.martechdevs.com/blog/customer-preference-centre) before adding the customer to campaigns.

## How do you measure AI-channel sales without inventing a conversion rate?

Measure AI-channel orders from the order record, and measure website conversion only among the website sessions that could produce that conversion. Keep discovery observations, referral evidence and direct-checkout orders identifiable rather than forcing them into one denominator.

Shopify's Agentic reporting includes sales, orders, online-store sessions and online-store conversion. Sales combines referral and direct-checkout sales. Its Catalog search preview is directional because channels can rerank the results. Neither view reveals every shopper's private conversation or the complete path to purchase. [Shopify's reporting and preview scope](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/agentic-home).

Google's direct checkout does not fire Google Analytics or custom client-side pixels; Shopify describes server-to-server checkout-started and checkout-completed pixels instead. Do not expect a browser purchase event from that path. [Google's pixel boundary](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/google).

A server-side signal existing somewhere does not prove that your GA4 property receives a correctly configured purchase. Inspect the destination, consent handling, identifiers and duplicate controls of any actual integration. Use the [Shopify-to-GA4 reconciliation guide](https://www.martechdevs.com/blog/ga4-shopify-purchase-mismatch) for the order-to-event investigation and [AI traffic classification](https://www.martechdevs.com/blog/track-ai-traffic-ga4) for identifiable website referrals.

### What does a two-path report look like?

Consider this hypothetical cohort in one currency and reporting window. There are 240 identifiable assistant-referred website sessions. Twelve distinct sessions complete checkout, producing twelve orders. Another six orders complete through direct checkout, without a website session in this dataset.

Each order contains USD 60 of merchandise after discounts, excluding tax and shipping. One direct-checkout order later receives a full USD 60 merchandise refund within the same observation window.

| Purchase path | Orders | Merchandise before refunds, USD | Merchandise refunds, USD | Merchandise after refunds, USD |
| --- | ---: | ---: | ---: | ---: |
| Store checkout after observed referral | 12 | 720 | 0 | 720 |
| Direct checkout in AI channel | 6 | 360 | 60 | 300 |
| Combined observed orders | 18 | 1,080 | 60 | 1,020 |

The website conversion rate is 12 completed-checkout sessions divided by 240 website sessions, or 5%. Dividing all eighteen orders by those sessions produces 7.5%, an invalid mixed-path conversion rate. The direct-checkout path has no comparable session denominator in this example.

```text
Website checkout conversion = 12 / 240 = 5%
Observed merchandise after refunds = (18 × USD 60) - USD 60
Observed merchandise after refunds = USD 1,020
```

The refunded order remains an order that completed checkout. Its later financial outcome belongs in the refund and value measures. Do not silently remove it from one historical conversion numerator while leaving the original session cohort unchanged.

Store the order ID, channel evidence, purchase path, currency, amount definition and refund identifiers. Deduplicate at the order level before joining analytics events; a repeated event is not another purchase. Preserve an unknown path when the evidence is missing. Avoid deriving it from a generic product name or an unverified AI label.

This report establishes associated orders and merchandise value. It does not prove incremental sales. To assess budget, include implementation and operating costs, margin and the evidence limits described in [measuring AI search ROI](https://www.martechdevs.com/blog/measure-ai-search-roi).

## What must pass before you expand the channel?

Expand when the required purchase and post-purchase cases pass, someone owns the exceptions, and the order report can explain its totals. Use a small representative catalog first; more products should not multiply a known failure.

The following cases are proposed merchant acceptance tests. Record the actual result and evidence for each channel rather than treating a checkbox as a completed test.

| Case | Expected result | Evidence to retain | Failure response |
| --- | --- | --- | --- |
| Exact variant and valid destination | Buyer receives the selected sellable item and accurate total | Variant, checkout and order records | Repair mapping or shipping rule |
| Unavailable variant | Buyer cannot complete an invalid purchase | Observed availability and order outcome | Investigate freshness and stock handling |
| Changed price or expired discount | Final accepted amount matches the disclosed offer | Before/after offer and payment values | Correct offer synchronization |
| Required option or acknowledgement | Required information survives on the order | Captured field and fulfillment view | Keep a supported store-checkout route |
| Failed or abandoned payment | No order is treated as paid without confirmation | Payment state and order state | Repair status interpretation |
| Repeated confirmation or integration retry | One real order remains one counted purchase | Order ID and duplicate evidence | Repair deduplication |
| Partial fulfillment or cancellation | Operations and customer updates show the actual state | Fulfillment and support records | Fix the post-purchase handoff |
| Partial refund | Correct payment adjustment, inventory decision and report value | Refund ID and linked order adjustment | Repair refund reconciliation |

Use a payment-provider-supported test path where available. A real controlled transaction needs a named owner, a known amount and an agreed refund process; do not assume test mode reproduces every live channel behavior.

Before signing off:

## Checklist

- [ ] Eligibility is confirmed for the store, product, buyer market and purchase route.
- [ ] The chosen feed contains the correct variant, price, availability and policies.
- [ ] Required checkout behavior passes on the actual channel path.
- [ ] Failed payment, retry and fallback cases produce the intended outcome.
- [ ] Fulfillment, support and refund ownership is explicit.
- [ ] Customer preferences reach downstream systems without being inferred from a purchase.
- [ ] Orders, refunds and website sessions use compatible definitions.
- [ ] Expansion has an owner, a review trigger and a reversible channel setting.

Review the channel again when a required checkout app, product type, market, fulfillment rule or feed mapping changes. A passed test for an ordinary single-item order does not cover a later subscription launch.

Bring the channel settings, representative products and one disputed order to the [$99 martech audit](https://www.martechdevs.com/pricing); those artifacts give the investigation a concrete starting point.

## Frequently asked questions

### Can a Shopify business outside the US use ChatGPT product discovery?

Yes, a store outside the US can qualify if it sells to US customers and meets Shopify's other ChatGPT requirements. Google's Shopify channel separately allows stores in any country, with direct checkout rolling out to shoppers in the US, Canada and Australia. Confirm the shopper's access and product eligibility; neither channel's requirements guarantee availability for every buyer.

### Do we need our own AI shopping bot to participate?

No. An eligible Shopify store can use its supported agentic storefronts without first building a custom bot. Start by verifying the catalog and purchase path already available to your account. Consider custom development when a specific customer action or integration is missing and you can own its ongoing operation. A separate agent introduces additional testing rather than replacing merchant readiness.

### Does turning off Shopify Catalog access hide every product from AI?

No. Removing Catalog access controls that distribution route; products may still be found through web crawling or other feeds. Blocking crawlers also does not stop an enabled Catalog feed. Choose the control for the actual route you want to restrict. Shopify's Unlisted product status has wider search-discovery consequences, so assess those effects before using it as a blanket response.

### Can a headless Shopify store use the same AI-channel dashboard?

Do not assume it can. Shopify says agentic storefronts performance analytics are not yet available for headless stores. Establish the order and referral evidence your own storefront can retain, then reconcile it with Shopify records. Keep unobserved discovery unknown. A custom analytics dashboard should explain its coverage and missing paths rather than recreate unavailable native metrics by assumption.

### Does an AI-channel purchase authorize lifecycle marketing?

An order record alone is not sufficient evidence for your campaign eligibility rules. Keep the service messages needed to fulfill the purchase separate from optional promotional communication, and preserve the actual preferences captured on the buying path. Check the destination's requirements and applicable consent rules before activation. If a required preference was not collected, do not manufacture it from the order source.

## Sources

1. [Shopify: Agentic storefronts](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts)

2. [Shopify: Selling on ChatGPT](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/chatgpt)

3. [Shopify: Selling on Google AI Mode and Gemini](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/google)

4. [Shopify: Selling on Microsoft Copilot](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/microsoft)

5. [OpenAI: Product discovery in ChatGPT](https://openai.com/index/powering-product-discovery-in-chatgpt/)

6. [OpenAI: Commerce concepts](https://developers.openai.com/commerce/guides/key-concepts)

7. [UCP: Core concepts](https://ucp.dev/documentation/core-concepts/)

8. [Google: UCP implementation paths](https://developers.google.com/merchant/ucp/guides/overview)

9. [Google: UCP integration-hub eligibility](https://developers.google.com/merchant/ucp/guides/merchant-center/overview)

10. [Shopify: Catalog requirements](https://help.shopify.com/en/manual/shopify-catalog/requirements)

11. [Shopify: Catalog and discovery controls](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/products)

12. [Google: Product data specification](https://support.google.com/merchants/answer/7052112?hl=en)

13. [Google: Merchant listing structured data](https://developers.google.com/search/docs/appearance/structured-data/merchant-listing)

14. [Shopify: Catalog Mapping](https://help.shopify.com/en/manual/shopify-catalog/mapping)

15. [Shopify: Managing agentic storefronts](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/agentic-home)

16. [Shopify: Refunds](https://help.shopify.com/en/manual/fulfillment/managing-orders/refunding-orders)

17. [Google: Order lifecycle](https://developers.google.com/merchant/ucp/guides/overview/order-lifecycle)

18. [Shopify: Data sharing](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/data-privacy)


19. [Shopify: Agentic plan setup](https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/agentic-plan-setup)