AI Already Chose the Product Before Your Customer Opened Your App

AI Already Chose the Product Before Your Customer Opened Your App

Your search bar is now a confirmation box. Shoppers aren’t discovering products on your site — they’ve already been told what to buy by a chatbot, and they’re just typing it in to find it. Twenty retailers are sitting on proof of this, and most of them don’t know it yet.

What happened

Iannelli and Ai linked real AI assistant conversations to real browsing sessions across twenty large online retailers and watched what happened next. Among shoppers who hadn’t visited a store in the prior week, product-page visits to the exact item the assistant named jumped from 0.02 to 2.07 per 100 answers — a +2.0 percentage-point lift — within 24 hours of the conversation. On days when a shopper landed on a product page, 44% were viewing a product from their own AI answer; for a matched answer given to a different person, that rate was 1.5% — a ~29x gap that’s hard to dismiss as noise. The pattern replicated across a separate five-month window. The kicker: 89% of these shoppers arrived with no referral link from the assistant, and 54% came through a web search — the same share as their normal browsing. But when a search immediately preceded the first visit, it named the assistant’s product 28% of the time (versus 1% in earlier weeks), as if the search box had become an address bar. The authors propose “product follow-through” as a new metric to evaluate shopping assistants — no referral data or provider logs required, computable per assistant. This matters for anyone thinking about AI visibility, share of voice in AI, or brand entity presence in LLMs — because if your product isn’t named, the funnel doesn’t start.

Cold read

This is a correlation study dressed in causal language. The paper links conversations to browsing but cannot fully rule out that shoppers who use AI assistants for product research are simply more purchase-intent shoppers to begin with — the assistant name-drops a product because the user already knew roughly what they wanted. The 44% figure is striking, but it’s conditional on days with a product-page visit at all; we don’t see the denominator of how many AI conversations resulted in zero store visit, which is almost certainly the majority case. The study covers twenty unnamed large retailers — results may not generalize to niche verticals, B2B, or categories where AI assistants hallucinate product specs badly enough to erode trust. “Product follow-through” is a clever metric, but it requires retailers to have both conversation data and browsing data linked at the user level — a data infrastructure most mid-market operators simply don’t have. And the finding that shoppers “keep browsing” once they arrive is mentioned without quantification, which is precisely the number a conversion-rate-obsessed founder would need.

What it means for you

  • Signal maturity: 3/5 — replicating pattern, but causal mechanism unconfirmed and data access is elite-tier only
  • Who gets hurt: Organic SEO teams and paid-search managers whose attribution models record a “search-driven” visit and congratulate themselves while the real decision already happened in a chat window
  • What breaks if this is true: Your entire top-of-funnel investment thesis — content, SEO, PDP optimization — is now downstream of whoever controls the AI answer; GEO and LLMO become table-stakes budget lines, not experiments
  • Why it might not land: Most retailers can’t link assistant sessions to on-site behavior; without that infrastructure, “product follow-through” is a metric you can admire but not act on
  • Watch for: Any major retail analytics platform (Adobe, GA4, Contentsquare) shipping a “pre-visit AI mention” attribution layer — that’s the signal this finding has crossed from academic to operational

Forecast as of 2026-11-18

By Q3 2027, at least two of the top-five e-commerce analytics vendors will ship a feature explicitly designed to measure AI-assistant-to-visit attribution — and at least one major retailer will publicly cite “share of AI product mentions” as a reported KPI alongside search impression share.


Source: Straight to the Product: What Online Retailers Miss When Shoppers Ask AI First — Michael Iannelli, Alan Ai. https://arxiv.org/abs/2611.20378v1

Similar Posts