Why Your Shopify Store Needs Better Product Data for AI Shopping

If a shopper asks ChatGPT for "a cotton kurta under ₹2,000 that ships to Pune in two days", the model has to answer from structured data it can read. Not from your hero banner. Not from your brand story. From product titles, attributes, price, availability, shipping terms and return policy, expressed in a form a machine can parse without guessing. Most Shopify catalogues we audit fail that test on the first three fields. The titles are marketing copy, half the metafields are blank, and the size chart lives in an image.

That's the whole problem. AI shopping surfaces — Google's AI Overviews and AI Mode, ChatGPT's shopping results, Perplexity, Bing Copilot, and whatever Meta ships next — don't browse. They retrieve, compare and summarise. A product that can't be compared on attributes doesn't make the shortlist, and there's no second page to be on.

What these systems actually read

Three sources, roughly in order of weight:

• Your merchant feed. Google Merchant Center, Meta catalogue, Microsoft Advertising. These are already structured, already attribute-rich, and already the backbone of Google's shopping data. Whatever an assistant knows about your price and stock, it probably knows from here.
• Schema.org markup on the product page. Product, Offer, AggregateRating, plus the shipping and returns properties most stores ignore. Shopify's Dawn-based themes emit a basic Product block. Basic is the word.
• The rendered page text. Crawlers read the HTML. If your fabric composition is in a tab that loads via JavaScript after a click, assume it was never read.

Note what isn't on that list: your app-driven "AI SEO" plugin that rewrites meta descriptions. Meta descriptions have never been a ranking factor and they are not what a retrieval model grounds an answer in.

The attribute gap is where stores lose

Here is the test we run on a new catalogue. Take ten products at random. For each, try to answer these without opening a PDF or reading an image: material, colour (as a standard name, not "Monsoon Haze"), size with a measurement, weight, country of origin, warranty length, care instructions, what's in the box. Count how many of the eighty cells you can fill from structured fields.

On a typical Indian D2C store running a theme set up two or three years ago, we get somewhere between twenty and thirty-five. Call it 30 out of 80, so 37.5% attribute coverage. Google's shopping taxonomy for apparel alone wants fifteen-plus attributes before it treats a listing as complete. An assistant asked to compare three kurtas on fabric weight picks the two that have the fabric weight.

Fixing this is unglamorous work. It's metafield definitions, a CSV, and someone who actually knows the product. We've done it for catalogues in the low thousands of SKUs and it takes weeks, not days, because the data mostly doesn't exist anywhere yet. The merchandiser knows the GSM of the fabric. It was never written down.

Shopify gave you the tooling in 2024, and most stores haven't used it

Standard metafields and the taxonomy that ships with Shopify admin are the single highest-return thing in this post. When you assign a product to a category in Shopify's taxonomy, you get category-specific attributes — sleeve length, neckline, heel height, screen size — as structured fields rather than free text. These flow to Google and Meta feeds automatically and they are legible to anything that reads your catalogue through those channels.

Three practical notes from doing this:

• Pick the deepest category that's accurate. "Apparel & Accessories" gets you nothing. "Apparel & Accessories > Clothing > Dresses" gets you the dress attribute set.
• Bulk-edit from the products list, not product by product. The bulk editor handles metafield columns.
• Custom metafields still matter for anything the taxonomy doesn't cover. FSSAI licence number on food SKUs, for instance, or BIS hallmark details on jewellery. Define them with proper types — a measurement is a measurement, not a single-line text field.

Product titles written for humans and machines at once

The conventional advice is to write titles for people. The conventional advice predates retrieval models. A title like "The Weekender" tells a shopper nothing and a model less.

The pattern that works: brand, product type, two or three distinguishing attributes, then the model or variant name if you need it. "Thriftizer" is not an example because we don't sell kurtas, so: "BrandName Men's Cotton Linen Kurta, Full Sleeve, Mandarin Collar — Indigo". It reads fine in a collection grid. It's also a complete answer to four likely filters.

Where this gets tense is with brands whose whole positioning is the poetic product name. Keep it. Put it at the end, and put the descriptive part first. Nobody has ever bounced because a title was too clear.

Descriptions: structure beats prose

Long unbroken paragraphs of brand voice are the worst-performing description format for AI retrieval, because there's nothing to extract. A model summarising three competing products wants discrete claims it can lift.

What we ship on product pages now:

1. Two to three sentences of actual prose, up top, that says what the thing is and who it's for.
2. A spec list as real HTML list items with label-and-value pairs.
3. A short "good for / not good for" block. This one is underrated. A model asked "is this suitable for sensitive skin" can answer directly from it, and honest exclusions cut returns.
4. Size and fit as text, with the chart as an HTML table. Not an image. We see size charts as JPEGs on maybe half the apparel stores we look at, and every one of them is invisible to every machine that matters.

FAQs on the product page, marked up with FAQPage schema, do real work here too — specifically the questions your support inbox gets ten times a week. Does it shrink. Is it COD-eligible. How long does delivery take to a tier-2 city.

Schema markup your theme probably isn't emitting

Check your product page in Google's Rich Results Test. Then check which of these are present:

• shippingDetails with rates and delivery time estimates
• hasMerchantReturnPolicy with the window in days
• gtin, mpn and brand — a missing GTIN is the most common reason a product gets deprioritised in shopping surfaces
• aggregateRating pulling from your actual review app, not hardcoded
• priceValidUntil on sale pricing
• Variant-level Offer entries rather than one offer for the parent product

Most themes handle the first level and stop. The shipping and returns properties matter disproportionately because they answer the two questions that kill a purchase decision, and an assistant that can state your return window will recommend you over a competitor it can't state one for. Adding them is a theme-level JSON-LD change, usually a day or two of developer time. If you're weighing that against another app subscription, the markup wins. Our notes on what Shopify development actually costs in India cover how this kind of work tends to get scoped.

Inventory and price accuracy is now a visibility issue

A feed that says in-stock when the product is out of stock used to cost you a disappointed click. Now it costs you trust at the platform level. Google suspends Merchant Center accounts over price and availability mismatches. Assistants that get burned on availability stop surfacing the source.

For Indian stores specifically, three recurring breakages:

• Multi-location inventory. If you fulfil from a Bengaluru warehouse and a Delhi 3PL, and your feed publishes a single availability value, it will be wrong for one of them regularly. Shopify Markets handles this better than a manual feed, but it still needs setting up deliberately.
• GST-inclusive versus exclusive pricing. Your feed must publish the price the customer pays at checkout. If your store shows exclusive prices to B2B segments, the feed needs to be scoped to the consumer-facing price.
• Metal-rate products. Jewellery priced off a live gold rate goes stale the moment the spot price moves, and a feed updated once a day will be wrong by the afternoon. If that's your category, the pricing has to update automatically rather than through a weekly CSV — that's exactly the problem our Gold & Silver Live Rate app was built for, with the rate breakup shown on the product page so the number is explainable as well as current.

Your own site search is the same problem, wearing different clothes

Here's the part that pays for the work regardless of what happens with AI search. Every attribute you add for machines is an attribute your own filters can use. A store with 1,200 SKUs and three filters is sending shoppers into a wall. The same store with proper material, fit, occasion and colour metafields can filter meaningfully, and on-site search starts returning the right thing for "sleeveless cotton top" instead of everything containing the word cotton.

We've seen filter-use rates on collection pages go from under 10% of sessions to above 30% purely from attribute coverage, with no UI change. The filters were always there. There was nothing to filter on. If your catalogue is past a few hundred SKUs, FilterPro will put the structured data to work on the storefront side; the data has to exist first, though, and no app invents it for you.

What to do in the next fortnight

Order of operations, because doing this out of sequence wastes effort:

1. Assign every product to the deepest accurate Shopify taxonomy category. Bulk edit. One afternoon for most catalogues.
2. Fill the category attributes for your top 50 products by revenue. Not all of them. The top 50.
3. Open Merchant Center diagnostics and clear every disapproval and warning. GTINs first.
4. Convert size charts and spec sheets out of images into HTML tables.
5. Add shipping and return policy properties to your product JSON-LD.
6. Re-run the ten-product, eight-attribute test. If you're not above 70% coverage, keep going before you spend anything on ads.

The stores that will show up in AI shopping answers in a year are the ones whose catalogue data is boringly complete today. There's no clever shortcut and no app that does it for you, because the missing information lives in your merchandiser's head.

If you want a second pair of eyes on where your catalogue stands, our free store audit includes a structured-data pass across product pages and feeds, and we'll tell you which of the six steps above is actually worth your time first.