How consistent is the memory-versus-search split across different product categories

A single pooled figure, like "16% of brand mentions appear both with and without web search," can hide a lot of variation underneath. It's possible that figure is an average of some categories where overlap is near total and others where it's near zero. This study checked that by recalculating the same three-way split (memory-only, search-only, both) separately within each of 82 categories, with 81 of those categories having enough mentions to compute a stable share.

The result: the pattern is remarkably consistent in direction, but genuinely variable in size. Across those 81 categories, the share of mentions falling into the "named with and without web search" class ranges from a low of 2.2% to a high of 36.8%, with a median of 14.7%. That's a wide spread, over sixteen times, between the lowest and highest category. In other words, there exist individual categories where overlap between memory and search answers is close to negligible, and others where more than a third of mentions survive both conditions, well above the overall pooled figure of 16%.

The other two classes show similar within-category spread. "Named only from memory" ranges from 25.5% to 47.5% across categories (median 33.4%), and "named only with web search" ranges from 34.6% to 69.7% (median 47.9%). Notice that even at its narrowest point, the search-only class's minimum (34.6%) is still higher than the overlap class's median (14.7%). That's a useful sanity check: it suggests search-only mentions are not just large on average because of a handful of extreme categories, they tend to be the largest or near-largest class in most categories individually, not only in the pooled total.

What might explain a category swinging toward higher overlap, up near the 36.8% end? The likeliest driver is how saturated or stable that category's brand landscape is. A category with a small number of long-established, heavily documented brands gives the model little room for search to surface anything new, memory and search converge because there isn't much else to find. A category that is newer, faster-moving, or has many small competitive entrants gives search much more room to pull in names the model wouldn't otherwise produce, pushing overlap down and the search-only share up. This study did not code categories by age or competitive density, so this is a plausible mechanism rather than a measured one. Confirming it would mean cross-referencing category-level overlap against some independent measure of market concentration or brand count, which falls outside this dataset.

The practical takeaway for a brand in any single category is to treat the pooled 16% figure as a rough midpoint, not a guarantee. A company competing in a category near the low end of this spread should expect its memory presence and its search presence to behave almost as two separate competitions, and plan visibility work for each independently rather than assuming one will carry the other.