AI Shopping Answers Draw on 233 Domains and Repeat 13: Source-Pool Depth Across 43 Segments

The mechanic measured here is source-pool depth: how many distinct domains an AI answer engine draws on across a fixed set of shopping questions, and how that pool divides between domains the engines return to and domains they touch once. Measured against the Machine Relations Index release machine_relations_index_public_view_v2.0, methodology v2.0, the September 25, 2026 Machine Relations Index release, generated 2026-09-25, the 43 published shopping-shape segments cite a median 233 distinct domains per segment. In the median segment 13 of those domains are cited in at least 10 percent of observed answers, and 44.4 percent of the pool is cited exactly once.

For a brand's leaders, those two numbers answer different questions. The pool figure says how many domains an engine is willing to reach for in the category. The depth figure says how many it actually returns to — and that is the set a shopper's answer is assembled from on any given day.

This is a cross-section, not an experiment. It reports what six engines cited over a fixed 132-day window. It makes no claim that any page, placement, schema change or campaign caused a citation rate, and it does not compare a before to an after. Para Labs publishes it as a measured baseline a later experiment can move against.

The measurement

Field	Value
Release	The September 25, 2026 Machine Relations Index release
Window	2026-05-10 to 2026-09-25, 132 days observed
Shapes included	best_x best-tools, top_list ranked roundup, x_vs_y head-to-head comparison
Segments	43 published category-and-shape segments across 15 categories
Observed answer runs	4,884
Domain citations counted	35,308
Distinct segment-and-domain pairs	10,486
Engines	ChatGPT, Claude, Gemini, Google AI Mode, Google AI Overviews, Perplexity
Evidence floor per segment	10 observed runs across 7 distinct run dates

Two parts of the release are deliberately left out. The news_topic shape and the legacy-unmapped category run on a different observation sample — 53 distinct run dates against the 7 that a shopping segment carries — so mixing them into a per-segment median would compare a high-frequency monitoring series against a periodic one. Every figure below is from the 43 shopping segments only.

The machine-readable release carries every row, the release manifest carries the window and engine roster, and each segment is readable on its own page — for example Consumer Products best-tools answers, AI Visibility roundups and Family Software comparisons.

Wide pool, shallow contest

Medians below are medians of the per-segment values, not ratios of the medians.

Shape	Segments	Median domains cited	Median citations per answer	Median cited in 10% or more of answers	Median cited in 5% or more	Median share cited exactly once
best_x	15	247	7.72	13	36	47.9%
top_list	14	267	8.09	13	37.5	43.6%
x_vs_y	14	182	6.03	11.5	30	44.0%
All 43	43	233	7.17	13	34	44.4%

Read the third and fifth columns together. The pool is 233 domains wide in the median segment and the set of domains cited in at least one answer in ten is 13 — a median 5.5 percent of the pool, and in no segment more than 13.9 percent. Loosen the bar to one answer in twenty and the set reaches a median 34 domains, still a median 15.6 percent of the pool. In 34 of the 43 segments, 15 or fewer domains clear the one-in-ten bar.

The other end of the same distribution is the part a dashboard reports as a win. In the median segment, 44.4 percent of all cited domains were cited exactly once across roughly a hundred observed answers. In 34 of 43 segments that share is 40 percent or higher; at the extreme, 371 of the 558 domains cited in AI Visibility roundups appeared once. A single citation is the most common outcome in the dataset, and it is not evidence of a durable position — it is the shape of the tail.

That distinction matters commercially. A brand that appears once in a hundred observed answers and a brand cited in a quarter of them both register as "cited" in a monitoring tool. One of them is in the set the answer is built from and one is in the pool the engine occasionally reaches into. The gap between a median 13 and a median 233 is the size of that misreading.

Every published shopping segment

Answer runs is the count of observed answers in the segment during the window. Domains cited is the count of distinct domains cited in at least one of them. Citations is the count of segment-and-domain citation observations. The top-slot figure is the most-cited domain and the share of observed answers citing it.

Category	Shape	Answer runs	Domains cited	Citations	Cited in 10%+	Cited exactly once	Top-slot domain
AI Infrastructure	best_x	107	263	872	15	130 of 263, 49.4%	medium.com, 42.1%
AI Security & Privacy	best_x	136	275	1111	15	124 of 275, 45.1%	arxiv.org, 38.2%
AI Visibility & GEO	best_x	113	351	1115	20	189 of 351, 53.8%	youtube.com, 26.5%
Consumer Finance	best_x	132	236	1027	18	113 of 236, 47.9%	forbes.com, 39.4%
Consumer Health	best_x	132	298	1090	15	156 of 298, 52.3%	healthline.com, 43.2%
Consumer Products	best_x	131	318	1043	12	157 of 318, 49.4%	forbes.com, 29.8%
Cybersecurity	best_x	105	247	833	13	136 of 247, 55.1%	huntress.com, 53.3%
Deep Tech & Hardware	best_x	121	328	928	9	158 of 328, 48.2%	reddit.com, 20.7%
Education & Training	best_x	106	263	818	13	130 of 263, 49.4%	reddit.com, 34.0%
Emergent Prosumer	best_x	100	219	641	8	92 of 219, 42.0%	reddit.com, 21.0%
Enterprise Software	best_x	101	167	616	9	48 of 167, 28.7%	erpresearch.com, 33.7%
Family Software	best_x	114	118	637	13	37 of 118, 31.4%	techradar.com, 34.2%
Fintech	best_x	108	224	794	16	97 of 224, 43.3%	airwallex.com, 38.9%
Healthcare Services	best_x	107	199	660	13	83 of 199, 41.7%	omnimd.com, 32.7%
HR & Talent	best_x	107	186	767	17	69 of 186, 37.1%	pin.com, 29.9%
AI Infrastructure	top_list	108	270	930	23	138 of 270, 51.1%	spheron.network, 37.0%
AI Security & Privacy	top_list	106	313	987	18	171 of 313, 54.6%	ovaledge.com, 39.6%
AI Visibility & GEO	top_list	101	558	1102	10	371 of 558, 66.5%	youtube.com, 26.7%
Consumer Finance	top_list	132	233	1096	15	101 of 233, 43.3%	nerdwallet.com, 38.6%
Consumer Health	top_list	131	338	1202	15	143 of 338, 42.3%	forbes.com, 30.5%
Consumer Products	top_list	131	380	1186	14	195 of 380, 51.3%	chewy.com, 23.7%
Cybersecurity	top_list	130	264	1098	13	100 of 264, 37.9%	hoxhunt.com, 28.5%
Deep Tech & Hardware	top_list	106	279	834	12	116 of 279, 41.6%	unmannedsystemstechnology.com, 28.3%
Education & Training	top_list	100	295	746	9	149 of 295, 50.5%	reddit.com, 19.0%
Emergent Prosumer	top_list	101	211	659	13	82 of 211, 38.9%	reddit.com, 19.8%
Enterprise Software	top_list	114	264	762	6	125 of 264, 47.3%	procuredesk.com, 16.7%
Family Software	top_list	108	195	703	13	79 of 195, 40.5%	apple.com, 37.0%
Fintech	top_list	102	200	627	11	85 of 200, 42.5%	stripe.com, 24.5%
Healthcare Services	top_list	102	223	661	9	98 of 223, 43.9%	omnimd.com, 24.5%
AI Infrastructure	x_vs_y	108	231	678	13	136 of 231, 58.9%	youtube.com, 33.3%
AI Security & Privacy	x_vs_y	114	263	762	11	140 of 263, 53.2%	arxiv.org, 47.4%
AI Visibility & GEO	x_vs_y	132	417	1021	10	251 of 417, 60.2%	hubspot.com, 26.5%
Consumer Finance	x_vs_y	138	170	914	16	78 of 170, 45.9%	acorns.com, 38.4%
Consumer Health	x_vs_y	131	217	943	18	107 of 217, 49.3%	healthline.com, 32.8%
Consumer Products	x_vs_y	131	268	931	10	119 of 268, 44.4%	youtube.com, 29.8%
Cybersecurity	x_vs_y	107	189	661	12	85 of 189, 45.0%	exabeam.com, 26.2%
Deep Tech & Hardware	x_vs_y	105	171	616	13	72 of 171, 42.1%	reddit.com, 22.9%
Education & Training	x_vs_y	100	126	542	16	55 of 126, 43.7%	uopeople.edu, 28.0%
Emergent Prosumer	x_vs_y	95	118	448	11	48 of 118, 40.7%	youtube.com, 37.9%
Enterprise Software	x_vs_y	114	186	639	7	72 of 186, 38.7%	erpfocus.com, 20.2%
Family Software	x_vs_y	108	101	530	14	28 of 101, 27.7%	useboomerang.com, 26.9%
Fintech	x_vs_y	102	136	534	10	44 of 136, 32.4%	wise.com, 18.6%
Healthcare Services	x_vs_y	107	178	544	5	62 of 178, 34.8%	adsc.com, 17.8%

The top-slot share ranges from 16.7 percent to 53.3 percent of observed answers, with a median of 29.8 percent. Exactly one domain across the 43 segments is cited in a majority of its segment's answers: Huntress, in Cybersecurity best-tools answers, at 53.3 percent. In every other segment the most-cited source appears in a minority of observed answers, and at the median it appears in roughly three answers in ten.

Pool width is not the same thing as concentration

Two adjacent measurements answer different questions and should not be read as this one. Concentration asks how few domains it takes to reach half the citations in a category, published as category citation concentration and source distribution. Rank movement asks whether a given domain's position holds still over time, covered in AI visibility ranking instability. This measurement asks a third thing: across a segment's whole observed answer set, how many domains were reachable at all, and how many were reached more than once.

A brand can be inside a concentrated head, outside it, or — most commonly — in the single-citation tail, and the three measurements will disagree about how well it is doing. The one that maps to a shopper's experience is the depth figure, because a shopper sees one answer. One category in this table is worth its own read: in Emergent Prosumer, reddit.com holds the rank-1 cited slot in four of the six published question shapes, a consistency this pool-depth table cannot show because it reports only the single top-slot rate per shape, not whether the same domain holds it across shapes.

Limits

The release reports how many observed answers cited a domain, not which ones. Overlap between two domains' citations is therefore not computable from the public artifact, so this measurement says nothing about whether the 13 frequently cited domains in a segment appear together or in rotation.

Domains are counted as registrable domains. The ingest step that writes the citation rows merges subdomains into the registrable domain, so two subdomains each cited once in a segment are recorded as one domain cited twice. That biases the single-citation share downward: the true share of sources cited exactly once is at least the figure reported here, and the pool represents more distinct hosts than the domain count shows.

Segment status matters. Of 157 category-and-shape segments in the taxonomy, 95 are published and the rest are still collecting evidence against the floor of 10 observed runs across 7 distinct run dates. A category absent from the table above is not a category where nothing is cited; it is a category where the observation count has not yet reached the publication floor.

One field in the public artifact is worth reading carefully before recomputing these figures. The per-segment field runs_cited_domains counts distinct domains rather than runs: in AI Infrastructure best-tools answers it reads 263, the domain count, while the citation-observation count for that segment is 872. Sum runs_cited across the domain records to get the event count.

Methodology and sources

Citation rates, answer-run counts and per-segment domain records were read from the public release artifact at https://machinerelations.ai/data/machine-relations-index.json, checksum 04fcb7fb8f290732063c585fa3d37f8aadd2d1bb63ed9082a4f502d6560c877d, with the window and engine roster read from https://machinerelations.ai/data/mri-release-manifest.json. Per-segment figures were computed by grouping every domain record's methodology v2.strata entries by category and question shape, restricted to segments with status: published, excluding the news_topic shape and the legacy-unmapped category. Medians are medians of per-segment values. The definition of a citation and of the source-role labels used above is published in the Machine Relations Index methodology. Named top-slot domains are linked at their own sites for verification: arxiv.org, youtube.com, forbes.com, healthline.com, nerdwallet.com, reddit.com, chewy.com, techradar.com, stripe.com, hubspot.com, apple.com and huntress.com.

A companion cross-section on the same release, measuring six named supplement brands across the same three shapes, is published as a consumer-health shape gap.