Skip to main content
Case studyRetail and commerce6 min read

When AI assistants send your buyers to somebody else's storefront

A controlled evaluation across leading assistants found official store links missing from a third of high-intent shopping journeys — with resellers surfaced instead, for reasons that are entirely fixable.

Client

Enterprise retail brand

Practices

Advisory · Intelligence

High-intent journeys missing the official store
30-40%
Assistants evaluated in parallel
Multi-model
Where failures concentrated
SKU level

The channel nobody owns yet

Product discovery is moving into conversation. A buyer describes what they need, an assistant proposes options, and the shortlist is settled before any storefront is opened. For a brand, that changes where the competition happens: not on the product page, but in whether the product page is cited at all.

Our client suspected they were losing that step. Traffic patterns hinted at it, but nothing in their analytics could confirm it, because an answer that never links to you leaves no trace in your own data. The only way to find out was to go and ask.

How we measured it

We ran a controlled, multi-model evaluation: the same set of realistic buying journeys put to each of the leading assistants under matched conditions, scoring whether the brand's own product URL appeared at the moment a buyer was ready to act. Casual browsing queries were excluded deliberately — the question was not whether the brand gets mentioned, but whether it gets linked when someone is about to spend.

Running it as a comparison rather than a spot check is what makes the result usable. A single failed query is an anecdote; the same failure pattern across models, repeated, is a channel problem with a cause.

Dashboard summarising how often official store links appeared across assistant journeys

Journey-by-journey scoring across the evaluated assistants.

What came back

  • Official store links were absent from 30-40% of high-intent journeys — cases where intent to purchase was explicit, not exploratory.
  • Failures concentrated at SKU and variant level. The more precisely a buyer specified what they wanted, the less likely the brand's own page was to be the answer.
  • Resellers were preferred, and earned it. Third-party listings carried richer structured data and more predictable URL patterns, so they were simply the easier source to cite.
  • Context decayed across multi-turn refinement, so the further a conversation progressed toward a decision, the weaker the brand's presence in it became.

The assistant was not avoiding the brand. It was choosing the source it could read most reliably — and that source belonged to somebody else.

Why this is worth fixing

Being missing from this channel is not only lost traffic. It hands the reseller the margin, the customer relationship, and the framing of the product — and it does so at the exact moment a buyer has decided to act.

  • Authority. Being the source an assistant cites establishes the brand as the reference point for its own products, not a stocklist entry.
  • Conversion efficiency. A direct link at the decision moment converts better than a route through an intermediary, and costs less to serve.
  • Narrative control. Specifications, warranty, and positioning stay as the brand wrote them rather than as a marketplace summarised them.

What the client got

The deliverable was an evidence-backed picture of where the brand stands in AI-mediated commerce, with the gaps ranked by what each would return — and, importantly, no requirement to change anything before deciding. Leadership could see the position, the cost of holding it, and the sequence of work that would improve it, and then choose.

That order matters. Most of the remedies here are structured data and crawler policy: unglamorous, inexpensive, and hard to justify without evidence. Measuring first is what turns them from a technical preference into a decision with a number attached.

Share this case study