Ask an AI to recommend a brand twice — will you get the same answer, or does the list scramble every time?

This explores whether a brand that shows up in one AI answer will show up again the next time someone asks, and what the corpus says about what stays fixed and what keeps changing. The most direct evidence is striking. SparkToro ran nearly 3,000 responses and found that AI recommendation lists almost never repeat: fewer than 1 in 100 runs produced the same set of brands, and about 1 in 1,000 put them in the same order How consistent are AI brand recommendation lists across repeated prompts?. Two things cause it. Models sample their answers with some randomness, and real people phrase the 'same' question in many different ways. So a screenshot of one answer tells you almost nothing about a brand's standing. It is one draw from a distribution.

The surprising part is that this volatility sits on top of a very stable layer. Ranqo looked at more than 100 brands across five AI engines and found citation rates sorted cleanly by how well known a brand already is: about 73% for global names, 44% for mid-market brands and 11% for niche ones. The pattern held even when the query didn't name any brand Do AI search engines favor well-known brands over others?. Put the two studies side by side and a likely picture emerges, though neither study tests it directly. The exact list reshuffles every time, but the odds that any particular brand appears are anchored by its existing fame. Big brands keep reappearing in different arrangements. Smaller brands flicker in and out.

Other notes suggest some of the instability isn't random noise. It's systematic bias that moves with how the question is framed. In strategy advice, six LLMs gave the same recommendation regardless of context, yet just changing the order of the options shifted their answers by 19%, more than changing the industry did (11%) Do LLMs consistently favor the same strategic choices regardless of context?. Shopping agents show the same thing. They favor products based on where those products sit in a list, even in text-only interfaces and even when told to ignore position, and each model has its own pattern that comes back after updates Do AI shopping agents show persistent position bias in text interfaces?. So some of the 'volatility' in brand mentions may be predictable once you know the model and the prompt structure.

A useful lens comes from a different field. Research on human annotation separates genuine preferences from 'constructed' ones that people make up on the spot, and the test is whether the answer stays consistent when the question is asked in different ways Do all annotation responses measure the same underlying thing?. Applied to AI brand visibility, a brand that survives rephrasing, reordering and repeated sampling holds a real place in the model's sense of the category. A brand that shows up only occasionally is closer to an improvised answer.

The corpus has limits here. Only one study measures repeat consistency head-on, and none tracks a single brand's mention rate over weeks or across model versions. The practical takeaway still holds: measure AI brand presence as a frequency across many varied prompts, never as a single result. Treat order and position as their own biased signals, not as rankings.