How much does the mix of list lengths vary from one query group to another?

This study classified 926,792 answer units into four classes by how many distinct businesses each answer named: names none, names 1 to 5, names 6 to 9, and names 10 or more. The units came from 193 query groups collected over the window 2026-04-27 to 2026-09-18, with an average of roughly 4,802.03 answer units per group. That per-group average is useful mainly as a denominator: it tells you these are not thin samples built from a handful of answers each, but groups with substantial numbers of answers behind each percentage.

The engine-level breakdown, covered elsewhere in this study, shows real differences between ChatGPT, Claude, and Gemini in how often each lands in a given list-length class. But those differences are modest compared to how much the class mix swings from one query group to the next. Take the largest class overall, names 6 to 9, which accounts for 51.8% of all answers pooled. Across the 188 groups counted, the share of answers in this class ranges from a low of 21.9% to a high of 85.1%, with a median of 53.1%. That is roughly a fourfold spread from the least to the most concentrated group, far wider than the gap between any two engines in this same class, where ChatGPT sits at 48.3%, Claude at 49.8%, and Gemini at 57.6%.

The same pattern holds for names 1 to 5, where the group-level share ranges from 1.5% to 28.8% with a median of 6%, again a wider spread than the engine-level differences in that class. It also holds for names 10 or more, ranging from 5.7% to 73.6% with a median of 38.3%.

One detail worth naming directly: no single query group dominates this dataset. The largest group accounts for 1.5% of all units, which means the wide group-to-group spread described above is not the product of one oversized group swamping the average. It reflects genuine variation across many distinct groups, each contributing a modest, comparable share of the total.

What does this mean for interpreting the engine-level findings elsewhere in this study? It means the engine you ask is not the only, or even the primary, thing determining whether you get a short list or a long one. The specific query group, meaning the type of question being asked, moves the class distribution by a larger margin than switching engines does. A query that structurally invites an exhaustive answer will push any engine toward more names, and a query that invites a narrow, specific answer will pull any engine toward fewer, regardless of which engine is doing the answering. This does not erase the engine-level differences documented elsewhere in this study, since those differences held up after aggregating across all 193 groups. But it does mean that a business trying to predict how many competitors will be named alongside it in a given AI answer should think first about what kind of question is being asked, and only second about which engine is answering it. The study applied 3 mitigations to guard against exactly this kind of group-driven distortion skewing the engine comparison, and the engine-level gaps reported elsewhere in this study held up under that check.