You may own "where to buy." That is not the same as owning "what to buy."
GapWatch measures what AI assistants tell shoppers about a brand — which questions it is named on, which it is absent from, and which websites those answers are built from. In consumer categories the two are rarely the same set of questions.
The finding
In a national retail category we measured across four assistants in two languages:
- Asked where to buy, the retailer was named in a clear majority of answers — 62.5%, the leading retailer in its category.
- Asked which product to buy — for a specific need, a specific use, a specific problem — it was named in 4.2%.
- A direct-to-consumer competitor the group does not own was named in 95.8% of those same answers.
Store questions and product questions are answered by different mechanisms and different sources. A retailer can lead the first and be effectively absent from the second — and the second is where the purchase is decided.
Three more patterns we see repeatedly
Direct-to-consumer brands outrank retailers, systematically
In the same measurement, every direct-to-consumer brand placed above every retailer — on where-to-buy as well as on which-to-buy. That is a property of the channel rather than of any one company's marketing, and it means a group's own DTC brands may be doing work its retail banner structurally cannot.
A second language is a different market, not a translation
In the French-language measurement of that same category, the leading brand was one that did not appear in the English results at all, and an entire competitive layer surfaced that was absent from English answers. Each language is measured separately.
And a launch has a before-picture that expires
Asked whether a chain re-entering a national market was operating yet, thirty-five of thirty-five answers were correct that it was not — and 91% volunteered the comeback unprompted, before a single store opened. That measurement exists only because it was taken beforehand. It cannot be reconstructed afterwards at any price.
What clients ask us
We have a portfolio. Do the assistants know which brands are ours?
Sometimes, and unevenly. In one group of twelve brands, the assistants correctly attributed six in every answer and were inconsistent on the rest. Whether that matters depends on whether the architecture is meant to be visible — but it should be a decision rather than an accident.
What about a campaign?
Campaign periods are frozen before the campaign runs, so the before-and-after cannot be adjusted once the result is known. We measure conversation volume, sentiment, share against named competitors, and which platforms carried it.
Can you tell us whether it drove sales?
No. We measure what is said and where it comes from. Correlating that with sales requires your sales data and a longer series than a single campaign provides, and we would rather say so than imply otherwise.
What an engagement looks like
A frozen baseline across the brands and competitors that matter. Monthly measurement in each language separately. The ranked source list per brand and per question type. Then the work on those sources.