What does Shopify’s AI tell shoppers about your products? I checked 290 against the merchants’ own pages

When an AI agent looks up a Shopify product through Shopify’s catalog, it gets more than the merchant wrote. Shopify adds AI-written “top features”, “tech specs” and “selling points” to almost every product: 292 of the 300 I pulled on 20 November. I checked 290 of them against the merchant’s own product page. For 15 products, about 1 in 19, at least one of those lines made a concrete claim the page doesn’t make: dishwasher safe, OEKO-TEX certified, free from parabens, hypoallergenic, Teflon and BPA free. The record the agent receives doesn’t mark which lines Shopify wrote. If you sell on Shopify, it’s worth reading what your own products say there.
Where these lines come from
Shopify Catalog is the product database behind Shop, ChatGPT and Copilot’s Shopify listings, and Shop MCP, which any AI agent can query for free. I covered that last part in my post on Shop MCP. Each product comes back with a metadata block holding top_features, tech_specs and unique_selling_points.
Shopify’s developer documentation is clear about who writes them. It tags each of those fields Inferred and says: “Some Global Catalog response fields are generated or enriched by Shopify, and are marked Inferred in the reference. These fields might not always be present or might vary in accuracy depending on available product data. Treat them as discovery and merchandising signals, not as merchant-authored source text.” The Global Catalog extension page lists the description too: “Product description generated or enriched by Shopify.”
That warning is written for developers. The label lives in the documentation, not in the data: across all 30 searches, the word “inferred” didn’t appear once in what came back. An agent that hasn’t read the docs gets one product record and has no way to tell your words from Shopify’s.
How I checked
I ran 30 everyday shopping searches through Shop MCP with a US shopper context, from “cast iron skillet” to “baby onesie” and “vitamin c serum”, 10 results each. That gave 300 different products. For 290 of them, from 221 stores, I could load both the merchant’s own product data and the live product page.
I didn’t use Shopify’s description as the merchant’s text, because Shopify says it may have enriched it. I used the merchant’s product description from the store itself, plus the full text of the product page.
Then I picked 25 properties a shopper might rely on and that a page either states or doesn’t: dishwasher safe, oven safe, waterproof, vegan, cruelty-free, fragrance-free, paraben-free, BPA-free, hypoallergenic, OEKO-TEX and similar certifications, recycled materials, warranty, handmade, and so on. 255 AI-written lines made one of those claims. For 226 of them, the same idea appeared somewhere on the merchant’s page. I counted any mention anywhere on the page, menus included, so that step is generous to Shopify.
That left 29. One was a false alarm from my keyword list (“Guarantees high-quality craftsmanship” isn’t a warranty). I loaded the other 28 pages in a real browser, so text hidden in tabs or loaded by scripts counted too, and read each one.
What I found
13 of the 28 were fine once a person read the page. A cast iron skillet’s AI line says “Oven safe: Expands cooking options beyond the stovetop”, and the merchant says it’s “Compatible with any stovetop (including induction), oven, grill, or campfire.” Same fact, different words.
3 were stronger than the page. A t-shirt’s line says “Hypoallergenic fabric is gentle on sensitive skin and reduces irritation.” The merchant says the fabric “stays comfortable even for those who usually find shirts itchy or irritating”, which is a comfort promise, not hypoallergenic. A vitamin C serum’s line says “Fragrance-free and unscented formula.” The merchant says it’s free of “artificial fragrances” and that the “scent … of each production batch can vary slightly.” A rain jacket became “waterproof construction” when the page describes water-resistant pockets and protection from wind and rain.
12 aren’t on the page at all. Some examples, quoted exactly:
- A ceramic coffee dripper: “Dishwasher safe (select models): Simplifies cleaning and maintenance.” No mention of dishwashers anywhere on the page.
- A 5-pack of baby bodysuits: “Oeko-Tex certified for safety and peace of mind.” The same brand’s 8-pack page states the certification. This one doesn’t.
- A vitamin C serum: “free from parabens, sulfates, phthalates, and artificial fragrances.” The page says vegan and cruelty-free, nothing about the rest.
- A sunscreen serum: “Vegan, fragrance-free, and reef-safe formula.”
- A stovetop kettle: “100% Teflon and BPA free.”
- A travel mug: “Handmade design option: Available in artisan block-printed versions.”
I’m not naming the stores, and I’m not saying these claims are false. Some are probably true. My guess is that some come from what the model knows about the brand or a sibling product, like the bodysuits, but that’s inference: Shopify doesn’t say how the fields are written. What I can say is narrower. The merchant didn’t make these claims on the page I checked, and the agent can’t tell they came from somewhere else.
15 of 290 is a floor, not an estimate of the whole problem. I only checked 25 properties. I didn’t check numbers in the tech specs, which is where a unit conversion or a rounded capacity could slip in. And my automatic step counted a match if the word appeared anywhere, including a menu.
Why it matters to you
Claims like these are the ones shoppers buy on. Someone with a skin allergy, a parent choosing baby clothes, a person who only buys cruelty-free: an AI answer that repeats “OEKO-TEX certified” or “paraben-free” is the reason they pick your product. If it isn’t true, the return, the complaint or the review lands on you, not on Shopify.
Shopify’s merchant help says nothing about these AI-written fields in the four Catalog pages I read. What it does say: “You can’t opt out of Shopify Catalog itself,” and the optimization page asks you to “provide accurate, detailed, and descriptive product information.” You can map which of your fields Catalog reads, but Shopify notes that this “doesn’t guarantee product placement, ranking, or exact wording in AI search results.” I found no documented way to see or correct the AI-written lines from the admin.
So the practical lever is your own page. The 13 claims that checked out had one thing in common: the merchant had already written the fact down. Where the page was silent, the AI filled the gap. If your product is dishwasher safe, certified, fragrance-free or covered by a guarantee, say so plainly in the product description. If it isn’t, saying “hand wash only” or “not certified” gives the enrichment something true to work from. That last part is my inference, not something Shopify documents. Writing the facts down is good for human shoppers either way. It’s the same pattern I found in what AI agents hear about your returns policy: what you write is what gets repeated.
Read what Shopify says about your products
This asks Shop MCP the same question my study did. Replace the search words with one of your product names and run it in a terminal:
curl -s https://mcp.shop.com/ -H 'Content-Type: application/json' -H 'Accept: application/json, text/event-stream' -H 'MCP-Protocol-Version: 2026-07-28' -H 'Mcp-Method: tools/call' -H 'Mcp-Name: search_catalog' -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_catalog","arguments":{"catalog":{"query":"YOUR PRODUCT NAME"}},"_meta":{"io.modelcontextprotocol/protocolVersion":"2026-07-28","io.modelcontextprotocol/clientInfo":{"name":"self-check","version":"1"},"io.modelcontextprotocol/clientCapabilities":{}}}}' | grep -oE '(top_features|tech_specs|unique_selling_points)[^]]*'
Each line it prints is something an AI agent may tell a shopper about that product. Read them next to your product page. Anything you didn’t say is worth adding to your page if it’s true, and worth a note to Shopify Support if it isn’t.