AI shopping agents ecommerce checklist

AI shopping agents can compare products, interpret a shopper’s requirements and help them move towards a purchase. For an ecommerce business, the immediate challenge is not building a futuristic checkout. It is making sure an AI system can find accurate information about what you sell.

This AI shopping agents ecommerce checklist gives Northern Ireland retailers a practical way to inspect that foundation. It does not promise that any platform will recommend a product. It helps you find the gaps that could make a product difficult to understand, compare or buy.

What can an AI shopping agent use today?

There are two broad routes. An AI shopping service may read publicly available product pages, or it may receive structured catalogue information through a merchant or platform integration.

OpenAI says its shopping research can use public product information and merchant data supplied through the Agentic Commerce Protocol. Its current merchant guidance for product discovery covers structured feeds containing titles, descriptions, images, prices and availability. Direct feed onboarding is currently limited to approved partners, and merchants need to confirm market support for their integration.

OpenAI also says Shopify catalogue data is already integrated into ChatGPT and individual Shopify merchants do not need extra work for that connection. That does not guarantee that a particular product will appear. Relevance and the accuracy of the underlying catalogue still matter.

Google supports a similar mix of website and feed data. Its product structured data guidance recommends product markup, a Google Merchant Center feed, or both. Google says using both can maximise eligibility for shopping experiences and help it understand and verify the information.

The sensible first step is therefore a data and website check, not a custom integration.

The AI shopping agents ecommerce readiness check

Choose five products that represent the awkward parts of your catalogue, not only the easiest listings. Include a variant, a sale item, an out-of-stock item, a product with delivery restrictions and one of your most important sellers. Check each one against the following areas.

1. Can the product be identified clearly?

An agent needs more than a clever product name. The page and catalogue should state what the item is, who made it and which exact version is for sale.

For each sample product, check:

the product has a stable, unique ID or SKU;
the title distinguishes the important variant, such as size, colour or capacity;
brand, model and recognised identifiers such as GTIN are present where they genuinely apply;
variants are grouped without giving different items the same identity;
the product URL is stable and opens the intended item or selected variant.

Do not invent identifiers for products that do not have them. The goal is consistent identity, not filling every possible field.

2. Does the description answer a specific request?

A shopper may ask for a waterproof jacket for a particular activity, a gift within a budget or an appliance that fits a measured space. An agent can only match those constraints when the relevant attributes are stated plainly.

Check whether the description includes material, dimensions, compatibility, intended use, care information and other facts that affect the buying decision. Put important attributes in structured fields as well as readable page copy where your platform allows it.

OpenAI’s current product feed specification requires core discovery fields and supports variants, product attributes, shipping and returns. Google likewise warns that missing or incorrect variant details and conflicts between a feed and website can limit product eligibility in its Merchant Center product data specification.

Avoid unsupported superlatives and blocks of supplier copy that do not explain the actual item. A precise description is more useful than a longer vague one.

3. Do price and availability agree everywhere?

Compare the product page, cart, structured data and any shopping feed. The current price, sale price, currency and stock status should agree. Test what happens when a sale starts, an item sells out or a pre-order becomes available.

This is a maintenance question as much as a setup question. OpenAI recommends keeping full product feeds current and updating changing records, while Google’s product guidance highlights conflicting feed and website data as a common cause of display and eligibility problems.

Record where each value originates and how quickly it updates. If staff change the website but an old spreadsheet continues feeding another channel, the business has two sources of truth.

4. Are delivery and returns unambiguous?

For a Northern Ireland retailer, location matters. An agent should not have to guess whether a price includes delivery to Belfast, whether an item can be shipped across the Irish Sea or how long a customer has to return it.

Check that the customer can find:

delivery regions and exclusions;
charges, thresholds and realistic timeframes;
collection options, where offered;
the returns window, method, costs and exceptions;
contact details for questions or problems.

Make the policy page agree with product-level messages and the checkout. Google supports shipping and return information in merchant listings, while OpenAI’s feed model includes shipping and returns fields. Clear policies also reduce uncertainty for human shoppers, so this work remains useful even where an AI platform does not consume every field.

5. Can authorised crawlers reach the useful pages?

Open an important product page without signing in and confirm that the main description, price and availability are available in the delivered page. Check that product URLs return a successful response, use a consistent canonical URL and appear in an XML sitemap.

Review robots.txt, firewall rules and bot protection with your web partner. OpenAI’s crawler documentation says OAI-SearchBot is used to surface websites in ChatGPT search and recommends allowing that crawler and its published IP ranges. It also explains that OAI-SearchBot and GPTBot controls are independent, so a business can permit search crawling while opting out of model-training crawling.

Do not switch off security controls indiscriminately. Confirm the official crawler identity and allow only the access the business intends.

6. Is the product data expressed in a standard format?

Inspect the page’s Product structured data with a validation tool and review diagnostics in Google Search Console and Merchant Center where those services are used. For a purchasable product, Google recommends merchant listing markup and supports details including price, availability, shipping and returns.

Then inspect the product feed produced by your ecommerce platform. A useful feed should include one accurate record per purchasable item or variant, current images, stable URLs and consistent identifiers. Fix errors in the source catalogue rather than repeatedly patching exports.

Structured data and feeds improve machine readability, but they are not a visibility guarantee. Each platform decides eligibility, relevance and presentation.

A simple scoring method

Score each of the six areas for your five sample products:

0: missing, blocked or contradictory;
1: partly present, but manual interpretation or correction is needed;
2: complete, current and consistent across the page, feed and checkout.

The maximum is 12 points per product. Treat the score as a prioritisation aid, not an industry benchmark. A low score in price or delivery accuracy may be more urgent than several minor description gaps.

Start with repeated failures. If four products use inconsistent size names, repair the catalogue rule before rewriting individual pages. If stock changes reach the website immediately but the feed updates weekly, fix the update path before adding more destinations.

Our plain-English guide to agentic commerce explains the wider change in shopping discovery. If you want a structured review of product data, technical visibility and the customer journey, see our Agentic Commerce Readiness service. A broader AI marketing audit may be a better starting point when ecommerce is only one of several marketing priorities.

What to do after the check

Turn the findings into a short sequence of work:

correct customer-facing errors in price, stock, delivery or returns;
establish one maintained source for product and variant data;
repair structured data and existing platform feeds;
monitor diagnostics and sample products after updates;
consider additional platform integrations only when the foundation is dependable and the relevant market is supported.

Recheck the sample after a normal catalogue change, not only immediately after the repair. That is how you learn whether the process stays accurate when the team is busy.

Frequently asked questions

Do we need a direct OpenAI product feed?

Not necessarily. OpenAI says shopping research can use publicly available product information, while Shopify catalogue data is already integrated for Shopify merchants. Direct product feed onboarding is currently for approved partners, and market support must be confirmed. Start by making the public product pages and existing catalogue accurate. Consider a direct feed only when it is available and solves a defined data problem.

Does product schema guarantee that AI shopping agents will show our products?

No. Structured data helps systems understand a page, but it does not guarantee crawling, eligibility, ranking or recommendation. It should accurately match the visible product and the feed. Treat it as one part of a dependable product-data system, alongside accessible pages, current inventory, clear policies and a buying journey that works when a customer arrives.

Is this different from ecommerce SEO?

There is substantial overlap. Clear product titles, useful descriptions, crawlable pages, stable URLs, structured data and accurate Merchant Center feeds already support search and shopping visibility. AI shopping readiness adds attention to conversational product attributes, agent-specific crawler access and emerging feed integrations. Improve the shared foundations first instead of running a separate speculative optimisation project.

How often should a retailer repeat the check?

Test after changes to the ecommerce platform, catalogue structure, feed app, delivery policy or bot protection. Also review a small sample on a regular schedule that matches how often prices and stock change. The important measure is how quickly incorrect information is detected and corrected, not how often a checklist is completed.

Want to know what an AI shopping service can understand about your store? Book a free call about an Agentic Commerce Readiness review →

Sources checked on 24 September 2026: OpenAI developer guidance and Google Search Central and Merchant Center documentation, linked above. Platform access, market support and presentation can change. The checklist and scoring method are our practical guidance, not a visibility guarantee or reported client result.