Does ChatGPT's brand instability depend heavily on which category you ask about?

The headline figure in this study, that ChatGPT names a different brand first in about three out of four repeated asks of the same buyer question, is an average across many categories of product or service. Averages like this can hide a lot. A category where the top brand is a clear runaway winner might show very little switching, while a category crowded with similar competitors might switch almost every time. This page looks at that spread.

The study measured instability separately within each of 143 categories (grouped from the full set of 172 groups used elsewhere in the study). In each category, the instability rate is the share of paired answers where the brand named first in the earlier answer differed from the brand named first in the later one.

Across these 143 categories, the median instability rate was 77.3%, close to the overall figure of 74.6% cited for the full dataset. But the range around that median is wide. The least volatile category in the study showed a different-brand rate of only 34.3%, meaning the same brand stayed on top the large majority of the time there. The most volatile category showed a different-brand rate of 92.8%, meaning the top brand changed on nearly every single repeated ask.

That is a difference of roughly sixty percentage points between the steadiest and the most chaotic category, which is large enough that the single headline number, three in four, materially misrepresents what a company in either tail of this distribution should expect. A brand competing in a category near the low end of this range faces a meaningfully more defensible first-mention position than the headline suggests. A brand in a category near the high end faces a nearly coin-flip, or worse, situation where being named first today says almost nothing about being named first next time.

Why would categories differ this much? A plausible mechanism is the number and closeness of credible competitors. In a category where one brand is dominant, by market share, by review volume, by sheer frequency of being written about, ChatGPT's retrieval is likely to surface that same brand repeatedly regardless of exactly which sources get pulled on a given ask, so the named leader stays put. In a category crowded with several brands of similar standing, small differences in which sources get retrieved, or how recently they were published, can tip the first-named spot between brands that are otherwise closely matched. This study did not measure market concentration directly, so this is an inference from the pattern, not a measured cause. A direct test would compare each category's instability rate against an independent measure of competitive concentration, which this dataset does not include.

For the same reason, the mirror statistic is worth stating plainly: the same-brand-still-leads rate also varies by category, from a low of 7.2% up to a high of 65.7%, with a median of 22.7%. These are just the complements of the different-brand figures above, but seeing them stated as stability rather than instability makes the practical question clearer: before treating a first-place mention as fragile or durable, a company needs to know roughly where its own category sits on this spread, not just the dataset-wide average.