How consistent is the naming pattern across all four size segments?

If category size mattered even a little to how many businesses an AI engine names in its answers, the effect should show up as a gradient. Moving from the smallest quarter of categories to the largest, the share of answers naming six or more businesses should climb, even if only slightly, at each step. This study is built to check for exactly that kind of gradient, because it does not just compare small against large, it splits the full sample into four ordered size segments and reports the naming split for each one separately.

The segments, built from 193 groups across 752,124 rows with none excluded, are ordered by row count: a smallest quarter of 88,655 rows, a second quarter of 179,165 rows, a third quarter of 222,559 rows, and a largest quarter of 261,745 rows. Within each, every answer set was classed as naming six or more distinct businesses or five or fewer.

Here is the full sequence, smallest to largest: 93.2% of the smallest quarter named six or more businesses, 91.3% of the second quarter, 90.9% of the third quarter, and 90% of the largest quarter. There is no climbing gradient. The middle two segments are effectively tied with each other, and the smallest and largest segments bracket them within a narrow band rather than anchoring two ends of a slope. If anything, the smallest quarter's share sits close to the middle of the range, not at the bottom, which is the opposite of what a size-driven gradient would predict.

This matters because a gradient, even a small one, would be evidence that category size is doing some causal work, that engines really do scale up their naming behavior as the pool of real candidates grows. The absence of a gradient across four independently measured segments is stronger evidence than a single two-group comparison would be, because a two-group comparison can hide a real but weak trend inside sampling noise. Four ordered points that fail to trend in either direction make the flat story harder to explain away as an accident of where the cutoffs were drawn.

What would make this wrong? If the size segments were built on a proxy that correlates poorly with actual market size, a real gradient in the world could be invisible here. It is also possible that a gradient exists but is small enough that it would only show up with finer segmentation, ten segments instead of four, or with a size measure that spans several more orders of magnitude than this sample does. Both are testable extensions, and neither is addressed by this study as run.

What the study does establish is that within this segmentation, the naming behavior is remarkably stable, four points across a size range from 88,655 to 261,745 rows all landing within roughly three percentage points of the overall 91% rate for six-or-more answers. A company deciding whether to worry about its category's size as a factor in AI visibility should treat that worry as low priority relative to other factors this study did not measure, such as how the business is described online.