Methodology
Every recommendation in your audit rests on research, not opinion. This is how that research is done: how we read real product listings, how a finding earns its place, how we protect the stores we study, and why the guidance keeps getting sharper over time.
The Benchmark Programme
The Product Understanding Lab studies how real, publicly published product listings answer — or fail to answer — the questions a buyer asks before purchasing. We work one category at a time, deliberately: skincare buyers ask different deciding questions than automotive buyers or food buyers, and a pattern only becomes useful once we understand the category well enough to know which questions actually decide the sale.
We are never looking for one store’s mistake. We are looking for a pattern that repeats — one that shows up across multiple brands, including brands that do almost everything else well. A pattern only becomes a registered finding once it has been checked against more than one catalogue and it holds.
The Validation Programme
Not everything the Lab observes becomes advice we give. Before an internal observation becomes something we put in an audit or publish, it has to clear three questions: is it accurate, is it genuinely useful to a merchant regardless of platform or tooling, and can it be described without identifying any individual store? Only findings that clear all three become recommendations. Everything else stays a working hypothesis until it does.
The organising unit is the buyer question. For each category we maintain the deciding questions buyers actually ask, and we validate which ones most affect whether an AI assistant will confidently recommend a product. That is why your audit speaks in buyer questions rather than generic content scores — the buyer question is the thing validation has shown actually matters.
Recommendation Confidence
The Lab’s findings all express themselves through a single idea: how confidently an AI assistant can recommend a product. Retrieval gets a product considered; answering the deciding buyer question is what earns the recommendation. It is worth understanding on its own.
Read how Recommendation Confidence works →Research integrity
We only ever read what is 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 is nothing to opt into and nothing to install.
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. And we describe patterns qualitatively: “we commonly see”, “less commonly”, rather than precise percentages the sample doesn’t support. If a pattern from your own audit ever informs future research, it will never be attributed to you without your permission.
Why recommendations evolve
Squiggle is designed so that every completed benchmark and every checked finding makes future audits smarter — not because templates changed, but because the research behind them got sharper. As a category is studied more deeply, the deciding questions we check become more precise, and the guidance in your report follows.
This is also why we frame recommendations carefully. AI shopping is new and moving; we describe what we observe and what it tends to mean, and we update as we learn more. You are getting our current, most carefully checked understanding — held honestly, and improved continually.
Research coverage
Automotive
Published findings, and fitment detection is live in the audit engine today.
Beauty & Skincare
Findings checked against real catalogues and published in the Research Library.
Food & Beverage
Findings checked against real catalogues and published in the Research Library.
Home & Garden
Findings checked against real catalogues and published in the Research Library.
Apparel & Fashion
Findings checked against real catalogues and published in the Research Library. Scoped to new/standard-sized retail — the fit answer is usually present on the page, but trapped in a size chart or fit-finder, not the readable text.
Sporting & Outdoor
Findings checked against real catalogues and published in the Research Library.
Baby & Kids
Emerging research into Baby & Kids listings — age range, size and safety information look like the questions that matter most. No findings shown until they’re checked.
Pet Supplies
Emerging research into Pet Supplies listings — animal size, breed suitability and usage instructions look like the deciding details. No findings shown until they’re checked.
Books & Media
Books are identified through product metadata (ISBN, edition, format); a dedicated reference set is being prepared under Product Understanding research.
Digital Products
Not yet studied in depth — digital and downloadable products raise different buyer questions to physical goods, and are on our roadmap for Product Understanding research.
Consumer Electronics
Under investigation — some compatibility signals have been probed, but there’s nothing to show here yet.
Jewellery & Watches
Under investigation — sizing, materials and authenticity look like likely deciding questions. Nothing to show here yet.
Furniture
Under investigation — dimensions, materials and assembly look like likely deciding questions, echoing what we’ve seen in Home & Garden. Nothing to show here yet.
Services
Under investigation — we’re exploring whether Squiggle’s product-understanding approach extends usefully to service listings. Nothing to show here yet.
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 no findings are shown until they’re checked. Under investigation means the category is on our research roadmap but not yet actively studied — the earliest, most conservative stage. Your audit today reflects fitment detection for Automotive plus catalogue-wide checks across every category.
Your free Store Audit applies this research to your own catalogue — in your inbox, in under 10 minutes.
Get my free Store Audit →