A practical guide to making product data accurate, discoverable, and ready for the next shopping journey.
A customer asks an AI assistant: “Find me a waterproof jacket in medium, under $150, that can arrive before Friday.”
Your store sells one. It is in stock. The product page looks excellent. But the assistant recommends a competitor.
The problem may have nothing to do with your design or your price. Perhaps your product feed calls the jacket water-resistant while the page calls it waterproof. Perhaps the medium variant has no distinct identifier. Perhaps a cached feed says it is out of stock. Or perhaps the assistant finds the product but cannot establish whether it ships to the customer’s location.
That is the new commerce challenge: a product must be understandable and trustworthy outside the storefront where you designed its experience.
In October 2026, Google launched Universal Cart in the US across Search and the Gemini app, allowing shoppers to collect products from multiple retailers and use supported checkout paths. Shopify has also introduced Agentic Storefronts for eligible merchants. These developments make AI-assisted product discovery a present operational question, although availability and purchasing features still vary by channel, merchant, and region. Google’s Universal Cart announcement, Shopify’s Agentic Storefronts documentation
The question for a retailer is no longer simply, “Does our website rank?” It is:
Can a shopping system identify the correct product, verify the current offer, and hand a customer to a reliable purchase path?
Three gates between your catalog and a sale
“AI-ready” is often treated as a single setting. It is not. A store has to pass three separate gates.
| Gate | What it means | What can go wrong |
|---|---|---|
| Discoverable | A channel can access an eligible product through a catalog, feed, or indexed page. | The product is excluded, disapproved, inaccessible, or absent from the relevant source. |
| Understandable | Its attributes accurately answer the shopper’s request. | Missing variants, vague descriptions, inconsistent specifications, or outdated availability make the match uncertain. |
| Purchasable | The shopper can reach an appropriate checkout with a valid price and fulfillment option. | The offer changes, the variant is unavailable, shipping cannot be confirmed, or the checkout path is not supported. |
Passing one gate does not guarantee the next. Nor does eligibility guarantee that an AI channel will recommend a particular product. Shopify explicitly notes that each channel decides what to display for a shopper’s request. Shopify product-discovery guidance
This is why “install an AI shopping plugin” is not an adequate commerce strategy.
The architecture behind an accurate recommendation
A useful mental model is:
Commerce system or PIM → product data mapping → storefront and structured data → channel feeds and catalogs → AI-assisted discovery → merchant checkout or supported channel checkout → order and attribution
A PIM is a product information management system. Many smaller retailers use their commerce platform itself as the source of product information instead. Either approach can work. What matters is that the same purchasable item has consistent facts throughout the chain.
Consider a running shoe with three sizes and two colours. The parent product may describe the shoe, but the customer buys a specific variant. That variant needs a stable identity, the right image, its own availability, and a price in the appropriate currency. If the product page, feed, and checkout disagree, the retailer has a data problem before it has an AI problem.
Google recommends providing both product structured data on pages and a Merchant Center feed to help it understand and verify products. Its Merchant Center guidance also warns that mismatches between feed, landing-page, and structured-data prices or availability can cause product issues. Google product structured-data guidance, Google Merchant Center mismatch guidance
The product truth contract
Before optimizing for any AI channel, define what must remain consistent for every sellable item:
| Product fact | Questions to settle |
|---|---|
| Identity | Do the store, feed, and integrations refer to the same SKU and variant? |
| Description | Are material, dimensions, compatibility, and other decision-making attributes accurate? |
| Offer | Do price, currency, sale conditions, and customer-visible price agree? |
| Availability | Does the channel reflect stock changes quickly enough for the business? |
| Fulfillment | Are shipping regions, costs, delivery expectations, and restrictions clear? |
| Policies | Can a shopper find the applicable returns and support information? |
| Destination | Does the product URL open the correct item and preserve the selected variant? |
Call this a product truth contract: not a new industry standard or a piece of markup, but an internal agreement between merchandising, engineering, and operations about which system owns each fact and how discrepancies are resolved.
It is the least glamorous part of agentic commerce. It may also be the most valuable.
Shopify: audit the data behind the automatic access
For an eligible Shopify store, Agentic Storefronts can make products available to participating AI channels. Shopify’s admin provides channel controls and information about catalog eligibility. However, “available to a channel” should not be read as “guaranteed to appear in a recommendation.” Shopify Agentic Storefronts
Start with Sales channels → Agentic in Shopify admin. Review:
- Whether the store and the intended products are eligible.
- Which channels are active and what each one supports for your store.
- The product count available through Shopify Catalog.
- Whether important attributes are stored in standard fields, metafields, or custom structures that need mapping.
- Whether products intended only for wholesale customers could accidentally expose D2C pricing.
The last point deserves particular care. Shopify cautions that custom B2B access or pricing implemented through third-party apps or theme modifications may not always be recognized as B2B-only by its catalog systems. Shopify Catalog and B2B guidance
Boxinall’s practical recommendation: choose a sample of your most commercially important variants and compare what the customer sees on the product page with what your catalog mapping supplies. Test sale items, low-stock items, bundles, and products with custom attributes—not only the five simplest products in the store.
WooCommerce: treat the feed as an integration, not a checkbox
WooCommerce gives merchants considerable control over their product model. That flexibility also means responsibility for feed quality, scheduled jobs, plugin compatibility, caching, and variant mapping often sits with the store operator.
The official Google for WooCommerce integration can sync product data to Google Merchant Center. In WordPress, merchants can review feed and account issues under Marketing → Google for WooCommerce → Product Feed. Missing required fields, identifiers, or category-specific attributes can prevent products from syncing or being approved. WooCommerce product-feed documentation
A sensible WooCommerce audit asks:
- Are scheduled product-sync jobs completing?
- Are variations represented as the actual purchasable items?
- Do custom attributes map to the fields a destination expects?
- Does a cache show yesterday’s price or stock status?
- Do the feed, rendered product page, and checkout show the same offer?
- Are Merchant Center issues reviewed regularly, rather than only during a campaign launch?
Important: synchronizing WooCommerce products to Merchant Center does not, by itself, mean the store has enabled every AI shopping feature or native checkout experience. Product discovery and checkout integrations have different requirements.
Custom stores: build for interoperability before a protocol
For a custom or headless store, there may be no platform-managed catalog to do the mapping for you. The engineering work begins with a dependable product model and clear ownership of updates.
A strong baseline includes:
- Crawlable product pages with stable, consistent URLs.
- Valid
ProductandOfferstructured data where applicable. - A Merchant Center feed with accurate product and offer details.
- Variant-level identifiers and images.
- Clear regional pricing and fulfillment rules.
- A dependable path from a recommended offer to the correct checkout state.
- Monitoring for feed failures and mismatches.
Do not confuse two emerging integration paths. Google’s Universal Commerce Protocol (UCP) concerns agentic commerce on Google’s supported surfaces; its merchant documentation describes onboarding and checkout requirements. OpenAI’s Agentic Commerce Protocol (ACP) has its own product-feed and checkout specifications; direct feed onboarding and Instant Checkout are currently subject to partner approval. They are not one universal switch a developer can turn on. Google UCP guide, OpenAI merchant onboarding, OpenAI checkout guidance
For most teams, the first investment should be reliable product data, not a rushed custom checkout integration.
The Boxinall Product Truth Test
Here is a diagnostic we would use to turn “Are we AI-ready?” into something a commerce team can investigate. It is a proposed internal test framework, not an official score from Google, Shopify, or OpenAI.
Select 20 products that matter to the business: bestsellers, complex variants, recent launches, discounted items, and products with shipping restrictions. For each, test five conditions.
| Test | What to verify | Score |
|---|---|---|
| Eligibility | The item is approved and available in each intended channel or feed. | 0–2 |
| Identity | The correct SKU and variant can be distinguished from similar items. | 0–2 |
| Truth | Price, currency, availability, and key specifications match across systems. | 0–2 |
| Fulfillment | Shipping and policy information is clear for the target market. | 0–2 |
| Handoff | The destination opens the correct purchasable item and checkout path. | 0–2 |
A score of 0 means missing or wrong; 1 means incomplete or inconsistent; 2 means verified. The maximum is 10 per product.
Do not publish the average score as a claim that a store will “rank in AI.” The score is useful because it exposes which operational failure to fix. A store that scores highly on eligibility but poorly on truth may need catalog reconciliation. One that scores highly until handoff may have a variant URL or checkout-state problem.
Then run a small set of real customer-style prompts, such as:
“Find a machine-washable waterproof jacket in medium under $150.”
Do not test only the exact product name. Customers describe needs, constraints, and use cases. Record whether the correct product appears, whether the assistant describes it accurately, and whether the purchase destination is valid. Results can vary between sessions and channels, so use the prompts as diagnostic observations, not a deterministic ranking test.
The unexpected test: change one fact and trace it
Most audits look at a static product. Commerce systems are not static.
Change the stock status or sale price of a test variant in the source system, then trace when that change appears on the page, in structured data, in the feed, and in each channel dashboard that provides visibility. Record the delay and any failures.
This change-propagation test can reveal a broken scheduled job, stale cache, bad mapping, or feed rejection that a one-time schema check would miss. It also tells the team where monitoring belongs.
A practical 30-day readiness plan
Week 1: Establish the baseline. Identify the intended markets and channels. Select the test catalog. Document which systems own product identity, pricing, stock, and shipping data. Check current feed approvals and product-page markup.
Week 2: Fix product truth. Repair missing attributes, incorrect variant mappings, conflicting prices, stale inventory, and broken product URLs. Prioritize high-revenue products and errors that could mislead customers.
Week 3: Verify the platform path. Review Shopify Agentic eligibility and mapping, WooCommerce sync and Merchant Center issues, or custom feed and structured-data pipelines. Investigate checkout integrations only where the merchant and market are eligible.
Week 4: Test and measure. Run the Product Truth Test again. Document remaining failures. Set an owner for feed monitoring and a process for new-product launches, promotions, and stock changes.
The point is not to declare the store “AI-optimized” on day 30. It is to give the business a dependable way to identify and reduce preventable errors.
What should you measure?
Do not judge this work solely by whether an AI assistant mentioned your brand during a few manual prompts. Track the evidence you can actually obtain:
- Percentage of priority products eligible in the intended catalog or feed.
- Product disapprovals and their causes.
- Price and availability mismatch rate.
- Time taken for important product changes to propagate.
- AI-channel referrals or orders where the platform provides attribution.
- Conversion and contribution margin from attributable traffic, where measurable.
Channel reporting will differ, and some discovery cannot be attributed cleanly. Treat reported AI referrals as an observable signal, not a complete count of every AI-influenced purchase.
Three questions merchants keep asking
Do we need to build our own shopping agent?
Usually, no. This guide is about making your existing products intelligible to shopping experiences customers already use. A customer-facing agent on your own store is a separate product decision with its own support, accuracy, and return-on-investment requirements.
Will structured data guarantee recommendations?
No. Structured data helps systems interpret product information; it does not guarantee placement. Google also states that there is no special schema required simply to appear as a supporting link in its AI search features. Product feeds, merchant eligibility, relevance, and the channel’s own selection process remain distinct considerations. Google’s AI features guidance, Google’s product-data guidance
Should we move platforms to become “AI-ready”?
Not on that argument alone. Shopify offers a managed path for eligible merchants, while WooCommerce and custom stores can also provide structured, accurate product data. A platform move should be justified by the store’s broader needs—operations, flexibility, ownership, cost, and growth—not a promise of AI visibility.
The opportunity is operational, not magical
AI shopping will change how some customers discover and compare products. It will not make poor product information harmless.
The retailers best positioned for this shift will know exactly which item they sell, what it costs, whether it is available, where it can be delivered, and how a customer can buy it. They will also know when those facts stop agreeing across systems.
That is the work behind being found.
Boxinall Softech builds commerce experiences and integrations, including Shopify storefront work for 98 Degree North. If your team wants to assess its product data, catalog integrations, or checkout path, contact Boxinall with your platform, approximate catalog size, and target markets. We can start with the product-truth questions before proposing a build.
