Small categories get just as many distinct businesses named as huge ones, once you control for run count

Why this question matters

If you run a small, specific category, such as a regional tool-rental niche or a narrow B2B software line, you might assume an AI engine answering questions in your space would default to naming one or two obvious players simply because there is less to talk about. Conversely, you might assume a sprawling category, like general project-management software, would force the engine to spread its attention across dozens of names just to cover the territory. Either story would matter to a business owner deciding whether being named at all is a matter of luck, category size, or something else.

This study tests whether category size predicts how many distinct businesses an engine names. The answer has direct consequences for two audiences. First, operators in small categories: if small categories reliably got fewer distinct names, then being one of a handful of mentioned businesses would be a modest achievement, easy to reach and not very informative about quality. If instead small categories get just as many distinct names as large ones, then appearing in that list means the engine had many candidates to choose from and picked yours anyway. Second, operators in large categories: if size drove naming breadth, a crowded market would mean any single mention is diluted among dozens of competitors, and a business might reasonably give up on chasing visibility there. The data below says neither story is true across the 193 groups examined.

The practical stakes are about expectation-setting. A team that believes 'we're in a small niche, so we'll naturally be one of the few names mentioned' is making a bet on category size doing work it does not do. A team in a large category that believes 'there are too many competitors for us to be named' is making the same mistake in the other direction. What actually determines whether an engine names five businesses or fifteen is not covered by this study, but this study does establish that it is not simply the size of the category, measured here by the volume of rows, that is, indexed content or query volume, associated with the category. That rules out one convenient explanation and leaves the harder question of what does drive naming breadth for a separate inquiry.

How the measurement works

The unit of analysis is a single answer set: one engine's response to one prompt about one category, reduced to a count of distinct businesses named. Every one of the 752,124 answer sets in this study was classed into exactly one of two buckets: 'names 6 or more' or 'names 5 or fewer.' The cut at six was fixed before looking at results and applied uniformly, so a category that got exactly six names on one occasion and five on another crosses the line, but the boundary itself does not move to flatter any group.

Categories, not individual businesses, are the thing being sized. Each category was assigned a row count reflecting its underlying content or query volume, and then all 193 category-groups were sorted by that count and cut into four segments of roughly equal population: 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. This quartering by volume, rather than by an arbitrary threshold, means each segment represents a real quarter of the observed population rather than a hand-picked slice.

Two design choices are worth naming. First, the study did not exclude any rows: 0 of 752,124 rows were dropped, so the segments are not shaped by a hidden filter. Second, 3 corrections were applied before analysis, addressing known data issues such as duplicate answer sets or mis-parsed business names, without touching the six-name threshold itself. The average category-group in this dataset spans 3,897.02 answer-set units, and no single group dominates the total: the largest group accounts for only 1.5% of all rows, which matters for the next section because it means the headline split is not just one enormous category's behavior wearing a population-wide disguise.

The headline result

Across all 752,124 answer sets, 684,307 (91%) named six or more distinct businesses, and 67,817 (9%) named five or fewer. That is roughly a nine-to-one split in favor of naming a broad set of businesses, and it holds regardless of category size.

The smallest quarter of categories names six or more businesses in 93.2% of answer sets, barely below the largest quarter's 90%. The gap between the most and least 'broad-naming' segment is a few percentage points, not the wide divergence you would expect if category size were doing meaningful work.

What this does mean: category size, as measured by content or query volume, is not a good predictor of how many distinct businesses an engine will surface. A business in a niche category should not expect to be one of only two or three names mentioned simply because the category is small.

What this does not mean: it does not mean every category behaves identically, nor does it mean the six-name threshold is the only meaningful way to look at naming breadth, nor does it mean the businesses actually named are the same ones across repeated queries. It also does not identify what does drive the roughly one-in-eleven cases where an engine narrows to five or fewer names. This study establishes an absence of a relationship, not the presence of a different one.

The shape of the result: even, not concentrated

A finding that 'holds on average' can still be hiding a lot of variation: perhaps a handful of extreme categories drag the small-quarter and large-quarter averages toward each other while most categories in each segment look quite different from their segment's average. The segment table above already argues against this, since all four segments land within a narrow band of each other. But the segment view still only breaks the 193 category-groups into four buckets. The next section checks the same claim at the level of every individual group.

Before that, it is worth stating plainly what 'even' would look like versus what 'concentrated' would look like. If the result were concentrated, we would expect segments, or better, individual groups, to split sharply: some clustering near ninety-eight or ninety-nine of every hundred answer sets naming six or more, others clustering near ten or twenty of every hundred, with the 91% population figure sitting somewhere in the unhelpful middle as a blend of two very different populations. If the result is even, individual groups should mostly hover near the population figure, with only modest, expected variation.

The segment quarters already lean toward 'even': every quarter's six-or-more share sits between 93.2% and 90%, a tight range. That is a meaningfully different picture from, say, a scenario where small categories split fifty-fifty while large categories ran at ninety-nine to one. The consistency across segments built from wildly different row-count populations, 88,655 rows versus 261,745 rows, is itself evidence that whatever drives naming breadth is not tied to the volume metric used to build the segments.

This matters for how the finding should be used. A pattern that is genuinely even across the population supports a general statement like 'category size does not predict naming breadth.' A pattern that turns out to be concentrated in a few groups would instead support a narrower statement like 'a few unusual categories do not fit the size story, but most do,' which is a materially weaker and more conditional claim. The group-level spread, examined next, is what separates these two possibilities.

What the spread across groups shows

Segment averages can mask disagreement between the individual groups that make them up. To check this, every one of the 188 category-groups with enough data was scored individually for its share of answer sets naming six or more businesses.

The median group names six or more businesses in 91.8% of its answer sets, close to the population figure of 91%. That closeness is what 'the headline is a property of the whole population, not one group' looks like in practice: half of all groups sit above this median and half below, and the median itself nearly matches the aggregate.

But the range is real and worth stating honestly: the lowest group in the 'six or more' class sits at 65.3%, while the highest reaches 98.3%. That is a wide spread for any single group to occupy, from a category where roughly two-thirds of answer sets name six or more businesses to one where nearly all of them do. So while the median and the segment averages support 'no size effect,' individual categories still vary a great deal for reasons this study did not measure, plausibly things like how many businesses genuinely operate in that space, how distinctive their names are, or how the prompts were phrased. Category size explains little of this spread; something else explains most of it, and identifying that something else is outside what this dataset can answer.

Reading the six-name threshold correctly

No calibration or confidence-scoring metric is reported for this study; the classification of each answer set into 'names 6 or more' or 'names 5 or fewer' is a direct count, not a model prediction that could be miscalibrated. This section instead addresses the related question a careful reader should ask: how much should the six-name cut line itself be trusted as meaningful, given that it is a fixed boundary rather than a naturally occurring gap in the data.

The honest answer is that six was chosen as a round, pre-specified cutoff, not derived from a natural break in the distribution of names-per-answer-set. This means two things. First, the specific number six carries no special empirical status. If the boundary had been drawn at five or seven, the two-class split would shift somewhat, and a reader should not treat 'six' as a magic number the businesses in this space converge on. Second, and more reassuring, is that the near-identical segment splits shown earlier were not manufactured by threshold placement. Because the same fixed threshold was applied uniformly across every segment and every group, a threshold artifact would need to interact with category size in some coincidental way to produce the even pattern observed. There is no obvious mechanism for that, which supports treating the flatness across segments as real rather than an artifact of where the line was drawn.

What would change this conclusion: if re-running the analysis at a different threshold, say four or eight, produced a segment table where small and large categories suddenly diverged, that would indicate the six-name cut happened to sit at a point where category-size effects are invisible but exist elsewhere in the distribution. This study did not test alternate thresholds, so a reader who wants that reassurance should treat the six-name finding as specific to that boundary until shown otherwise, even though there is no positive reason here to expect a different threshold would tell a different story.

What someone acting on this finding should actually do

For an operator in a small category: stop assuming that a small market automatically means you are one of a short list of names an engine will surface. The data shows small categories get broad naming, six or more distinct businesses, in 93.2% of answer sets, nearly matching large categories. If you are not currently appearing in engine answers about your category, category size is not a plausible excuse. The more useful question is why the engine is choosing other names over yours from among a large pool of candidates, since that pool exists whether the category is small or large.

For an operator in a large category: do not conclude that a crowded market makes individual mentions worthless. Because broad naming is the norm everywhere in this dataset, being named alongside others is the baseline condition, not a sign of unusual dilution tied to your category's size specifically. Whatever is limiting your visibility is more likely tied to the factors this study could not measure, distinctiveness of name, depth of independent coverage, or specificity of the prompts used, than to sheer category size.

For anyone building a monitoring or benchmarking practice around this: segment your own tracking by category size only if you have a specific reason tied to something other than naming breadth. This study suggests that size-based segmentation will not explain differences in how many businesses get named, so budget and analyst time are better spent segmenting by other variables, prompt phrasing, category topic, or time window, when trying to explain why some categories show narrower naming than others. Use the group-level spread reported here, not the segment averages, as the realistic range to expect: a given category could plausibly land anywhere from roughly two-thirds to nearly all of its answer sets naming six or more, and that range is the planning assumption, not the 91% population figure alone.

How to read the published dataset yourself

The dataset underlying this study is organized around three levels: the full population (752,124 answer sets across 752,124 rows and 193 category-groups), four size segments, and individual groups within each segment. A reader working with the raw tables should keep straight which level a given number describes before quoting it.

Start with the class definitions. There are 2 classes in this study: 'names 6 or more' and 'names 5 or fewer.' Every answer set belongs to exactly one. When you see a percentage attached to a class, check whether it is scoped to the full population, a segment, or a single group, since these numbers look similar but answer different questions. A group-level figure of, say, 65.3% describes one category's behavior and should never be reported as if it were the population rate of 91%.

Check the denominators before trusting a percentage. The smallest quarter is built from 88,655 rows, smaller than the other three segments, so any single-group figure drawn from within that quarter rests on a thinner base than one drawn from the largest quarter's 261,745 rows. The dataset documents 188 groups as having enough data to be counted in the spread analysis; groups excluded from that count should not be treated as having zero variation, they simply lacked sufficient volume to score reliably.

Look for the mitigations note. 3 corrections were applied before analysis, and 0 of 752,124 rows were excluded outright. A dataset with zero exclusions and a small number of documented corrections is generally easier to audit than one with heavy filtering, since there is less room for a filtering choice to have shaped the result. If you rerun this analysis on a different measurement window than 2026-04-27 to 2026-08-16, expect the segment cut points to shift, since they are built from each window's own row-count distribution, not from fixed category-size thresholds carried over from this study.