AI Commerce Research

What AI actually sees when it reads your product page.

Every claim on this page comes from Squiggle's Product Understanding Lab — research read directly from real, publicly published product listings, one category at a time. Explore the six categories we've published below, then see how we turn research into evidence without ever naming the stores we studied.

Buyer's question: "Can you recommend brake pads for a 2018 Volkswagen Golf?"

Before — thin product listing

Product title

Brake Pads Front

Description

High quality brake pads. Easy to fit. Made for most vehicles.

AI shopping assistant says

"I can see this store sells brake pads, but I can't find compatibility information for your vehicle in the listing. You may want to check with the manufacturer before purchasing."

After — complete product listing

Product title

EBC Greenstuff Sport Brake Pads — Front — VW Golf Mk7 (2013–2019)

Description

Greenstuff compound engineered for street driving. Low dust, strong initial bite. Fits Golf Mk7 1.4T, 2.0T, and 1.6 TDI. Direct OEM replacement.

AI shopping assistant says

"Yes — the EBC Greenstuff Front Pads are a strong match for your 2018 Golf. Listed as compatible with Golf Mk7 models, with low dust and strong initial bite. In stock."

The gap: Missing vehicle compatibility and fitment data.

Every category asks a different question. The mechanism never changes — AI can only recommend a product when it can confidently answer that question.

The same product. The same store.
The only difference is whether AI has enough information to recommend it.

How this research happens

How the Product Understanding Lab works

The Lab reads real, publicly published product listings — one category at a time — and asks a single question: could an AI shopping assistant confidently answer the question a real buyer would ask about this product?

We're not looking for one store's mistake. We're looking for a pattern that repeats — one that shows up across multiple brands, including brands that do almost everything else right. A pattern only becomes a registered finding once we've checked it against more than one brand and it holds.

From research to claim

How findings become public claims

Not everything the Lab observes becomes something we publish. Before an internal finding becomes a public claim — like the examples above, or the category write-ups in our research library — it has to clear three questions: is it accurate, is it genuinely useful to a merchant regardless of what platform or tools they use, and can it be described without identifying any individual store?

Only findings that clear all three get published. Everything else stays an internal working hypothesis until it does.

Research governance

What we look at, and what we never touch

We only ever read what's already public — the same product pages any shopper, or any AI assistant, can see. We never access a store's account, sales data, or private information to do this research. There's nothing to opt into and nothing to install; it's simply public information, read carefully.

Why we never name names

Why company identities remain private

We never name an individual store's results in anything we publish without that store's consent. Findings are always described in aggregate — as a pattern across a category — never as a scorecard on any one business.

This matters for two reasons. First, it protects the brands we study during research, most of which do the majority of things well and none of which asked to be publicly graded. Second, it's the same protection your own store gets: if a pattern from your own audit ever informs future research, it will never be attributed to you without your permission.

Where each category stands

Current research coverage

Live means findings are published and a detector runs in your audit today. Published means the Lab's findings are checked against real catalogues and in the Research Library, with broader engine support still being built. In preparation means the category is entering Product Understanding research next — the site is ready for it, but we show no findings until they're checked. Under investigation means the category is on our research roadmap but not yet actively studied. Your Store Audit today reflects Automotive fitment plus catalogue-wide checks (duplicate content, missing descriptions, technical data gaps) across every category.

Read the published findings

Our research library carries the full category write-ups — what we found, why it happens, and exactly what to fix.