How often does ChatGPT name a different brand first when you ask the same question again?

If you ask ChatGPT a buyer question, like which product or company to consider for some purchase, and then ask the same question again later, does it name the same brand first both times? Usually not.

This study paired up answers to the same buyer question, one earlier and one later, and checked whether the brand named first in the first answer was still named first in the second. Across 172,624 such pairs, spanning 172 categories and a measurement window running from 2026-04-27 to 2026-10-16, the brand named first changed in 74.6% of pairs. The same brand held the top spot in only 25.4% of pairs.

Each pair is one unit of measurement here, not each question. A question asked many times over the window contributes many pairs, not one. That matters for interpreting the headline number: it describes how unstable any given pairing of answers tends to be, not how many distinct questions were tested. Across the 172 categories in the study, the typical group contributed roughly 1,003.63 pairs, though no single category dominates the total, the largest accounts for only 1.7% of all pairs.

The comparisons were built to isolate one thing: whether the leader changes when nothing about the setup changes. Every pair held the underlying model, the web search setting, and the prompt wording fixed. That is deliberate. ChatGPT is updated periodically, and web search results shift constantly, so if either of those were allowed to vary, a change in the named brand could simply reflect a new model version or a different web page being pulled in that day, rather than any real instability in how the assistant ranks brands. By holding model, search setting, and prompt version constant, the study can say the instability is not an artifact of the tool itself changing underneath the question. No rows were excluded from this analysis (0 excluded of 172,624), and the method applied three mitigations (see 3) to guard against counting noise as signal, such as filtering out near-duplicate categories or normalizing brand name variants before comparing.

Why would the named brand change at all, if nothing about the question or the tool's settings changed? The likeliest mechanism is that ChatGPT's answer to this kind of open-ended buyer question depends on information retrieved at the moment of asking, recent articles, updated product pages, shifting review aggregates, rather than on a fixed, memorized ranking. If the underlying evidence the model draws on shifts even slightly between two asks, the brand that edges out first place can shift too, especially when several brands are close contenders rather than one clear leader.

An alternative explanation worth naming: maybe the model's ranking is stable, but the brands themselves are genuinely close enough in quality or relevance that small, meaningless differences in retrieval tip the order each time, even though the underlying merit hasn't moved. Distinguishing these two stories, a genuinely volatile ranking process versus a stable-but-tightly-clustered one, would require looking at how far down the list the former top brand falls on the second ask. If it usually drops only one or two spots, that favors the close-contenders story. If it drops out of the top ranks entirely, that favors real volatility. This study measured only whether the same brand stayed first, not how far a dethroned brand fell, so it cannot settle that question on its own.

What this means for a company watching its own brand mentions: being named first once is not a stable asset on its own. A single favorable ChatGPT answer, screenshotted and circulated internally as proof of market position, is a snapshot of one moment's retrieval, not a settled ranking. The practical response is to treat first-mention tracking as an ongoing measurement rather than a one-time check, because the next ask, even an hour later with nothing else changed, has roughly a three-in-four chance of naming someone else first.