Recommendation Confidence
Being findable gets a product considered. Whether an AI assistant actually recommends it comes down to something else: can the assistant confidently answer the question a buyer asks before they buy? That is what we call Recommendation Confidence — and it is the single idea behind everything Squiggle measures.
The core idea
Retrieval gets you considered.
Answerability gets you recommended.
Two different things have to happen for an AI assistant to recommend your product. First it has to find your product at all. Then — and this is the part most merchants never see — it has to be able to answer the buyer’s deciding question from what your page says. A product can be perfectly findable and still never be recommended, simply because the page doesn’t answer the question that decides the sale.
Every AI recommendation moves through three stages. Squiggle focuses on the two that most merchants can influence directly.
When a shopper asks an AI assistant for a recommendation, it first gathers a set of products it might suggest, drawing on what it can read across the web and the retailers it has access to. Retrieval decides whether your product is even in the running. It is necessary — but on its own, it is not enough.
For each candidate, the assistant reads the product page the way a careful shopper would, looking for the specific facts that answer the buyer’s deciding question: is this right for me, will it fit, can I use it, what exactly is it. It can only reason about what the page actually says in words — a fact shown only in an image, or left off entirely, is a fact the assistant does not have.
The assistant recommends the product it can describe with confidence. If a page answers the deciding question clearly, the assistant can stand behind it. If the page leaves that question unanswered, the assistant does the safe thing: it hedges, or it recommends a clearer listing instead. Getting retrieved gets you considered. Answering the buyer’s question is what gets you recommended.
The four bands
Squiggle describes each product’s Recommendation Confidence qualitatively — never as a raw number — because that is how an assistant actually behaves toward it. These are the four bands your audit uses.
AI has no evidence to recommend the product with confidence, so it stays out of the answer entirely.
AI could not confidently answer most of the buyer’s questions, so it hesitates or hedges.
AI can partly answer the buyer’s questions, but not confidently enough to recommend without qualification.
AI can confidently answer the questions a buyer asks, so it recommends the product without hesitation.
What moves it
The fastest way to move a product’s Recommendation Confidence is rarely to write more. It is to answer the one question the page currently leaves open — the fact a buyer needs before they commit, and that an assistant therefore needs before it can commit on their behalf. A short page that answers the deciding question outperforms a long one that talks around it.
Improvement also has a natural ceiling. Once a page answers the questions that decide the purchase, adding further detail moves confidence very little — a plateau we see consistently. That is why Squiggle frames every recommendation as moving confidence toward a higher band, one meaningful step at a time, rather than promising a leap from nothing to certain.
A note on what this isn’t
Recommendation Confidence is not a search ranking, and improving it is not SEO. Squiggle makes no promises about traffic, rankings or positions. What we can do is show you, product by product, where an AI assistant currently lacks the evidence to recommend you — and which buyer question, answered, would change that. Everything we publish is framed as an observation, never a guarantee.
Your free Store Audit shows the Recommendation Confidence of your priority products — where each one stands now, and the one change that would move it most.