What Shopify’s AI adds to your listings: inferred fields

When an AI shopping agent reads your product from Shopify Catalog, part of what it gets is your copy. Part of it is Shopify’s. Shopify’s developer docs are open about this, but the docs are written for people building agents, so most merchants have never seen it.

What “inferred” means

In the reference for the Global Catalog, the API that agents search across all Shopify stores, some fields carry an Inferred label. Shopify’s definition:

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.

That’s from Shopify’s Global Catalog docs. Shopify’s public announcements describe the same thing in general terms: the catalog “uses specialized LLMs to categorize, enrich, and standardize product data” (Jan 11, 2026), and it “infers categories, extracts attributes, consolidates variants, and clusters identical items” (Dec 10, 2025).

The six inferred product fields

Shopify’s docs list these as inferred, with these descriptions:

Field	Shopify’s description
description	Product description generated or enriched by Shopify.
options	Product options normalized for catalog discovery and variant selection.
metadata.attributes	Product attributes such as material, style, and occasion.
metadata.tech_specs	Technical specifications.
metadata.top_features	Top product features.
metadata.unique_selling_points	Unique selling propositions.

A seventh, variants[].condition, labels items as new or secondhand.

Two of these matter most for accuracy. The description is the paragraph an agent is most likely to lean on when it summarizes your product. Top features and selling points are short, confident lines, like “fragrance-free” or “suitable from birth”, that are easy for an agent to repeat as facts.

Why an inferred field can drift from your copy

A model writing a summary compresses. It picks what seems important and rewords it. Most of the time the result is a fair paraphrase. When it isn’t, the cause usually looks like one of these:

• Category defaults. Products in a category often share claims, like “fragrance-free” for baby wash. A summary can attach the common claim to a product that doesn’t make it.
• Lost qualifiers. “Ages 4 and up” and “for ages 2 and up” are one word apart in a summary and very different on a label.
• Negation flips. A description that says what a product contains can come back as a list of what it’s free from.
• Facts from outside the description. Ingredient lists, FAQs and warnings often live in tabs or metafields, not the main description. A summary built without them can say something those sections contradict.

These aren’t guesses. On Sep 24, 2026 we read the catalog’s version of 192 top-ranked products across 20 everyday searches and checked every claim, 3,546 in all, against each brand’s own copy and product page. 89% of the products had at least one claim the brand never made, and 14% had one that directly contradicts the brand on health, safety or what’s in it (how we checked).

Some of what AI agents were told, with the brand’s own words beside it:

• A natural deodorant whose first ingredient is tallow was described as an “essential oil-free and tallow-free formula.”
• An eye cream was described as “non-comedogenic.” Its own FAQ says it “has not been tested for noncomedogenicity.”
• A baby shampoo was described as free from “preservatives.” Its own ingredient notes name two preservatives.

More examples, and the limits of what one day’s snapshot can tell you, are in We checked 192 products Shopify describes to AI agents.

Can you edit an inferred field?

Not directly. We haven’t found any setting in the Shopify admin, or any API, that lets a merchant write the inferred description or top features. Nobody outside Shopify can edit them, including us.

What you can change is the input:

• Your own product copy. Shopify’s advice is “clear, accurate, and detailed descriptions” (Optimizing your products for AI platforms). If a claim matters, say it plainly in the description, not only in an image or a tab.
• Catalog Mapping. Under Sales channels > Agentic, you can point the catalog’s title, description and category at a different source, such as a metafield. Shopify says these settings “are inputs that can influence Shopify Catalog” and are “processed with a delay” (Mapping your product data sources).
• Structured fields. Type, vendor, tags, barcode and variant options are all listed by Shopify as fields AI platforms consider. Filled-in structure leaves less for a model to guess.

None of these guarantees a particular wording. Shopify says so itself: mapping “doesn’t guarantee product placement, ranking, or exact wording in AI search results” (Shopify Catalog).

How to see what your inferred fields say

There are three routes:

1. The search preview in your admin. Sales channels > Agentic has a search tool that shows “the raw outputs of Shopify Catalog search” for a query, with listing-quality indicators (Managing agentic storefronts). It’s built for ranking. It won’t line up an inferred claim against your own copy.
2. The Catalog API. A developer can query Shopify’s catalog endpoints and read the inferred fields as JSON. That shows the raw text but doesn’t check it.
3. Hearsay. This is the job we built it for. It reads the inferred description, top features, tech specs and selling points for each product, splits them into single claims, and checks each one against your description, tags, product details and product page. Each claim comes back as a contradiction, something you never said, or a match, with both texts quoted in full. The claim categories doc shows how findings are labeled.

The free scan does this for 5 of your products without installing anything. The app rechecks daily and emails you when a claim changes. Pricing starts free.