Skip to content

[study]

Free-from claims in AI shopping: where summaries go wrong

Fragrance-free, preservative-free, tallow-free. Why AI product summaries get free-from claims wrong, with real examples, and how to write yours so they hold.

The Hearsay team Oct 12, 2026 4 min read

White Hearsay wordmark and headline “Free-from claims in AI shopping: where summaries go wrong” on black, with a large pink bracket pair around “study”.

“Fragrance-free.” “Preservative-free.” “Free from parabens, sulfates and synthetic dyes.” For a lot of shoppers, especially parents and people with sensitive skin, these are the claims that decide a purchase. They’re also the claims we most often found AI product summaries getting backwards.

What we found

On Sep 24, 2026 we read what Shopify’s catalog tells AI agents about 192 top-ranked products from 20 everyday searches, and checked all 3,546 claims against each brand’s own description, tags and product page. 14% of the products were described to AI agents with a claim that directly contradicts the brand on health, safety or what’s in it. For baby products it was 1 in 5 (how we checked).

Several of the clearest ones were free-from claims. Here are four, with the brand’s own words next to what AI agents were told. We don’t name brands. The products are real, and each example was checked by hand.

A natural deodorant. AI agents were told: “Essential oil-free and tallow-free formula.” The brand’s own copy: “Tallow from grass-fed cows, naturally-mined baking soda … and coconut oil.” Tallow is the first ingredient.

A baby shampoo. AI agents were told it’s free from “synthetic fragrances … and preservatives.” The brand’s own ingredient notes: “Sodium Benzoate: Food–grade preservative. Sodium Dehydroacetate: Organic salt preservative.”

Two “fragrance-free” products. In both, the product’s own ingredient list includes fragrance or essential oils.

A related case isn’t strictly free-from, but it fails the same way. An eye cream was described as “non-comedogenic.” The brand’s FAQ says it “has not been tested for noncomedogenicity.”

The full study, with its limits, is in We checked 192 products Shopify describes to AI agents.

Why free-from claims break

The catalog’s product description, top features and selling points are fields Shopify marks as “generated or enriched by Shopify” and says “might vary in accuracy depending on available product data” (Global Catalog docs). Free-from claims are where that variation does the most damage, for a few reasons.

They’re negations

“Contains tallow” and “tallow-free” are one word apart. A model condensing a long ingredient paragraph into a short feature line can flip it. The deodorant above is exactly that: an ingredient became its own absence.

They’re category defaults

Within a category, many products share the same free-from claims. Baby washes are often fragrance-free. Natural deodorants are often aluminum-free. A summary that leans on what’s typical for the category can attach the common claim to the product that doesn’t make it.

The evidence lives somewhere else

Brands usually list ingredients in a separate panel, tab or image, not in the description. The main description says “gentle” and “clean,” and the preservative is named three clicks away. A summary built mostly from the description has nothing that stops it saying “preservative-free.”

Vague words invite specific claims

“Clean,” “pure,” “no nasties” and “non-toxic” don’t say what’s absent. A summarizer that turns them into a list has to guess what they mean.

Why this category matters most

In our method, free-from claims are classed as Serious, alongside health, safety, age ranges and certifications. That’s because a shopper avoiding an ingredient is often doing it for a medical reason: an allergy, eczema, a pregnancy, a newborn. “Fragrance-free” is Serious even though it’s about an ingredient (What “serious” means).

A wrong color name costs a return. A wrong “fragrance-free” can cost a reaction.

How to write free-from claims that hold

You can’t edit what Shopify’s catalog writes. You can make your own copy much harder to misread:

  • State the free-from claims you mean, in the description, in words. Not only in a badge image or an icon row.
  • Be literal. If a product has essential oils, say “no synthetic fragrance; scented with lavender essential oil,” not “fragrance-free.” If it has preservatives, name them and say why they’re there.
  • Say what it contains, near the top. One plain sentence with the key ingredients, early in the description, gives a summary a positive fact to hold on to. That’s harder to flip than an absence.
  • Replace vague words with specific ones. Swap “clean formula” for the list of what’s left out.
  • Keep one version. The description, ingredient tab, metafields and images should agree. Shopify asks merchants to keep required disclosures “in the first 6,000 characters” of the description for AI channels (Selling on ChatGPT).

We go through more of these rules, with a before-and-after, in How to write product descriptions for AI shopping agents.

If a summary already says it

If the catalog says your product is free from something it contains, fix your own copy first, as above. If your copy is already clear, collect both versions and raise it with Shopify Support. AI getting your product wrong? walks through the whole process.

This is the check Hearsay runs. It reads what Shopify’s catalog tells AI agents about each product and compares every claim, free-from claims included, against your own copy. Contradictions come first, sorted by severity. Checking free-from claims automatically is hard, which is why every finding quotes both texts in full so you can see for yourself. The free scan checks 5 of your products now.

You didn’t say that.

See what AI agents are told about 5 of your products. Free, and under a minute.