Prepare Product Data for AI Shopping Chat
A practical source map and test set for product questions that depend on the right variant, custom field or policy.
A shopper asks whether the blue jacket in medium is suitable for wet commutes. Your product description says the jacket is water resistant. A size chart lives on another page. A custom field lists the fabric, and stock belongs to a particular variant. Before an AI shopping assistant can give a useful answer, those facts need to refer to the same item.
Prepare your product data by mapping each important shopper question to its source, keeping variant facts separate, and testing what happens when a fact is missing. Adding more description text will not fix an assistant that cannot access the information the answer requires.
Start with the questions your shoppers actually ask
Collect a small set of real questions from your store's search terms, support conversations and product enquiries. Remove personal information before using them in a test. Choose questions that change a buying decision: fit, dimensions, compatibility, care, price, availability or a relevant policy.
Separate a discovery request such as “a jacket for wet commutes under $150” from a specific fact such as “is the blue medium available?” The first may need clarification and several product facts. The second needs the correct variant and a current availability source. A confident general description does not answer both.
If you use Shopify, its Search & Discovery reports distinguish searches with no results from searches with results but no clicks. Those reports cover search-results-page activity and exclude predictive search. Use them as one source of questions, alongside support records, rather than treating them as a complete record of shopper needs.
Build a question-to-source map
For each question, write down the fact needed, where it lives, who maintains it, and whether the assistant can actually read it in your configuration. Here is an illustrative source map for the jacket example. These are test requirements, not a claim about a particular store or integration.
- “Which jacket suits wet commutes?” Needs material and weather-use information from approved product specifications. The merchandising team confirms the wording. The answer should explain the supported trade-off and ask about missing preferences.
- “Does this come in blue, medium?” Needs the exact combination of option values and its variant identifier. The catalog owner confirms the available combinations. The answer should distinguish that variant from the parent product.
- “Will medium fit me?” Needs the relevant size chart and a clear explanation of how the brand measures. The answer should ask for the information needed to use that chart and avoid a guaranteed fit.
- “Can I buy it now?” Needs the relevant current availability and sale conditions. The store's inventory owner confirms the source and update path. If the assistant cannot verify them, the answer should direct the shopper to the product page for confirmation.
- “Can I return it?” Needs the applicable published returns policy and any product-specific exception. The policy owner confirms the current version. The answer should link to that policy and avoid inventing an exception.
Mark a source “unavailable” when access is unverified. “The information exists in our admin” and “this assistant can use it” are separate findings.
Keep product and variant facts separate
In Shopify, a variant represents a combination of product options, such as a particular size and color. Shopify's variant guidance also explains that inventory can be managed per variant. A parent description saying “available in blue” is not enough to establish that blue, medium is available.
Use consistent identifiers, option names and units in the data you supply. Check whether a measurement describes the item, the packaging or the recommended wearer. Keep a product's general description separate from variant-specific facts that could contradict it.
Then check the destination. If a recommendation points to a product page, can the shopper identify the recommended option there? If the experience offers a cart action, verify the selected variant throughout that action. Showing the right product family is only part of the test.
Check custom fields and third-party options explicitly
Some important facts sit outside ordinary descriptions. Shopify metafields can hold specialized information such as dimensions, care instructions and size charts. An options app or external file may hold additional configurations.
List those sources before choosing an assistant. Ask which fields and files the proposed setup can read, how values map to a product or variant, and how changes reach the assistant. Check the answer with a real example from your own catalog.
Do not assume that an app's access to products includes every custom field or third-party configuration. Likewise, file-upload support alone does not establish correct product matching, update frequency or the ability to replace an obsolete value. Confirm each part with the implementation owner.
Test difficult answers before a pilot
Keep the source facts beside each test so a reviewer can grade the answer. Include these six cases:
- Exact match. Ask about a known variant with complete data. Pass when the answer identifies the right item and uses the relevant facts.
- Near match. Ask for an option that resembles an existing one but is not offered. Pass when the answer makes the difference clear rather than treating the alternatives as identical.
- Missing fact. Ask a compatibility question the supplied material cannot answer. Pass when the assistant explains the limit and offers a useful next step without inventing compatibility.
- Conflicting sources. Use an approved test environment with a deliberate disagreement between two sources. Pass when the team can identify the conflict and confirm which source should govern before customer use.
- Changed fact. Update one approved test value, then check when the answer reflects it. Record the actual delay; do not turn a single observation into a universal freshness promise.
- Destination check. Follow the recommended product or cart path. Pass when the product and selected options remain consistent. A click or cart add is not a completed purchase.
Save the question, expected behavior, observed answer, source version and result. Keep factual correctness separate from whether the explanation sounds helpful. Fix missing data, mapping or access first when those caused the failure, then rerun the affected cases.
Where Brambles.ai fits
Brambles.ai powers AI chat for product discovery. Its Shopify App Store listing describes catalog-based recommendations, uploaded FAQs and knowledge sources, and add-to-cart actions. Use those capabilities as a starting point for an evaluation with your catalog.
Confirm the exact variants, custom fields, external sources and update behavior with the implementation team. The source map and tests above are evaluation guidance; they do not establish automatic support for every PIM, CSV format or options app. Start with the Shopify integration and the Shopify product-discovery test guide.
Measure readiness separately from commercial results
Before a pilot, report how many reviewed questions had accessible sources and how many answers passed each test. Keep failures caused by missing data separate from failures despite complete data. Those figures describe your test set, not the whole catalog or all future shoppers.
During the pilot, follow the customer path with the events your store actually records: eligible visits, chat use, useful product destinations, selected variants, cart actions and verified orders. Keep each denominator visible. A more accurate answer is useful evidence, but it does not establish higher revenue or fewer returns. The chat revenue measurement guide explains the next measurement step.
FAQ
Should I rewrite every product description first? Start with the questions and missing facts that matter to a buying decision. Clear specifications and consistent variant data may be more useful than longer descriptions. Test the affected questions after each correction.
Can I use a spreadsheet or CSV as the source? Potentially, if the selected implementation supports that format and can map records to the right products. Verify access, identifiers, update handling and correction behavior before relying on it. A supported upload format does not prove a working ongoing sync.
What should the assistant do when it does not know? State the specific limit, ask a useful clarifying question when that could resolve it, or point to an appropriate product page or support route. It should not fill a missing fact with a confident guess.
Does better product data guarantee more sales? No. It can improve the information available for a recommendation. Whether shoppers use the experience and complete purchases needs a separate pilot and outcome measurement.
Related posts
View all
How to Test AI Chat for Final-Sale Variants on Shopify
Use one product and two colors to check whether AI chat gives the right return terms before a shopper adds the item to their cart.

How to Test AI Chat for Fixed Bundles on Shopify
Compare a bundle added from AI chat with the same bundle added from its Shopify product page. Check the options, quantities, and cart grouping.

How to Test AI Chat Across Shopify Markets
Use two Shopify markets and two products to see whether AI chat recommends items shoppers can buy after they change countries.
Explore Brambles.ai
Learn more about our AI-powered agentic commerce platform, agentic shopping, and shopping assistance solutions.
Explore More Insights
Discover more articles on AI, automation, and business innovation
View All Articles