Serious or minor: how we tell which AI claims matter

Open a report on a product and you might see fifteen or twenty findings. If every one of them looked equally urgent, you would either fix all of them or give up. So each finding carries a severity, and inside each finding type the report sorts by it. This post explains what those labels mean, how they are assigned, and where they are shaky.

Two labels that are easy to confuse

Every finding has a category and a severity, and they answer different questions.

The category says what the claim is about: age, ingredients, certification, care and so on. It is a label for finding and filtering, and the full list is in claim categories. The severity says how much it matters if the claim is wrong.

They are set separately, which is why the label on a finding reads category first, then severity, like “Safety · Serious” or “Care · Minor.” The two do not always move together. Most Safety and Age & choking claims are Serious, but a Care claim can be Serious too if it touches safety.

The three levels

These are the definitions our AI reviewers are given, and the same ones printed under what “serious” means.

Serious. Health, medical or efficacy claims. Safety or suitability, including age ranges and choking warnings. Certifications. Free-from claims, like “fragrance-free.”

Ingredients & specs. Ingredients, concentrations, sizes, counts, origin and dosage.

Minor. Texture and packaging.

That is the whole scale. The docs publish no numeric score and no threshold, so there is nothing more to it than this. A reviewer reads the claim, matches it to a definition, and assigns a level.

Why these claims are the serious ones

The common thread is what a shopper does with the information. Someone reading “safe for babies 6 months and up” is deciding whether to put a product on a baby. Someone reading “GOTS-certified” is relying on a test they cannot run themselves. Someone reading “fragrance-free” is often avoiding fragrance for a medical reason.

That last case surprises people. A free-from claim is about an ingredient, and ingredients sit in the middle tier. But “fragrance-free” is Serious anyway, because it is a free-from claim, and the person relying on it is often managing an allergy, eczema or a newborn’s skin. “Made in the USA,” by contrast, is Ingredients & specs. It matters, and it can be wrong, but a shopper who acts on a wrong origin is rarely harmed by it. We look at how these claims break in free-from claims in AI shopping.

The example claims in our docs come from Little Sprout Co., a fictional sample store, so none of them describe a real product. “Lightly weighted for calmer sleep” would be a Safety claim and Serious. “Tumble-dry safe” would be a Care claim, Minor unless it bears on safety. “Available in 6 colors” is an Options claim and has no safety angle at all.

Why we sort this way

Severity exists to set an order of work. Findings are grouped by type first, in a fixed order that is the order worth fixing in, and then by severity inside each type. A Serious contradiction, where your own listing says one thing and Shopify’s summary says the opposite, sits above a Minor one.

The same idea drove the numbers in our study of 192 products. There, “Serious” was used in its plainest sense: health, safety, certification and what a product does. Of the 192 top-ranked products we read, 70% had at least one claim of that kind that the brand never made, and 14% were described with a claim that directly contradicts the brand on health, safety or what is in it. Those figures are from one day, Sep 24, 2026, and from top results only.

Where the labels can be wrong

Severity is a judgment made by AI reviewers working from the definitions above, so it inherits their limits.

• The definitions are broad. A claim can sit between two levels. A reviewer picks one, and another reader might pick differently.
• Context is thin. The reviewer sees the claim and your copy, not your customer. A “gentle” claim on a hand soap and on a newborn wash look alike, and the difference is in who buys it.
• Minor is not harmless. Texture and packaging claims rarely cause harm, but a wrong one can still cause a return or a bad review.
• Reviewers can be wrong about the claim itself. That is why every finding quotes both texts in full, with the claim in question highlighted, so you can check the call yourself.

If a severity looks off to you, trust your own reading of the two quotes. The label is a way to order a list, not a ruling.

What to do with the order

Work down the report the way it is sorted. Start with contradictions marked Serious, since those are cases where AI agents may be telling shoppers the opposite of what you wrote about health, safety or what is in the product. Then move to Ingredients & specs, and leave Minor for last. How to write product descriptions for AI shopping agents covers the copy side of those edits. The suggested changes are always to your own listing, and AI getting your product wrong? covers how to trace where a claim came from before you rewrite anything.

If you want to read more of this kind of explanation, there are more explainers.

Common questions

Can I change a finding’s severity?

No. Severity comes from the definitions our reviewers use, not from a setting. If a finding looks mislabeled, the quoted text on both sides is the thing to go by.

Does a Minor finding mean I can ignore it?

Not necessarily. It means it is lower in the order of work. A Minor claim that is plainly wrong on your best-selling product may still be worth a five-minute edit.

Where do I see this on my own products?

The free scan checks 5 of your products, and every finding carries a category and a severity. Free is free for good, and paid plans cover the rest of your catalog. Details are on the pricing page.