Shopping-Shape Carriers, Re-Measured: 6 of 7 Vendors That Hold Two Shopping Shapes Also Hold a Top-10 Decision-Question Slot. 3 of 25 Single-Shape Vendors Do.

One release after the shopping-shape cross-section, the seven vendor–category pairs whose top-10 AI citation slot carries across two shopping shapes are the same seven. Engine count, confidence grade and overall citation volume do not separate them from the 25 single-shape pairs. Standing in the same category's how-buyers-choose, is-it-worth-it and problem-first answers does: 6 of 7 carriers hold a decision-shape top-10, against 3 of 25.

Para Labs Research··
ai-visibilitybrand-visibilitycommercemeasurementcase-study
Para Labs research image for Shopping-Shape Carriers, Re-Measured: 6 of 7 Vendors That Hold Two Shopping Shapes Also Hold a Top-10 Decision-Question Slot. 3 of 25 Single-Shape Vendors Do.

The mechanic measured here is shape carry-over: whether a vendor domain that holds a top-10 AI citation slot in one shopping question shape also holds one in another, and what distinguishes the vendors that carry from the ones that do not. The shopping-shape cross-section published on the 2026-09-18 release found 33 vendor–category pairs in the top ten of a published shopping segment, 26 of them holding exactly one shape and 7 holding two, and named the next experiment: whether the seven carriers share a source-layer property the 26 do not.

They do, and it is not the property a brand team would guess. On the 2026-09-19 release the carriers are the same seven. Six of the seven also hold a top-10 slot in at least one of the same category's three decision shapes (how buyers choose, is it worth it, problem-first research). Three of the 25 single-shape vendors do. None of the 25 holds two. The variables a brand team would reach for first — how many engines cite the domain, its confidence grade, its citation volume across the whole index — do not separate the groups at all. The largest domain in the set is single-shape.

This is a cross-section read on two consecutive daily releases, not an experiment. It reports what six engines cited over a fixed 126-day window. It does not claim that decision-shape citation causes shopping-shape carry-over, and it names the read that would falsify it.

The measurement

The source is the public view of the Machine Relations Index, contract machine_relations_index_public_view_v2.0, methodology mri_score_v2.0, release mri_score_v2.0+2026-09-19+0cad03121f60. The machine-readable release carries every number below.

Field Value
Window 2026-05-10 to 2026-09-19, 126 days observed
Answer runs 15,883
Citation events 125,115
Cited source domains 22,213
Eligible monitored queries 912
Engines ChatGPT, Claude, Gemini, Google AI Mode, Google AI Overviews, Perplexity
Evidence floor per segment 10 observed runs across 7 distinct run dates

A segment is one subject category paired with one buyer question shape. Six shapes are published: three shopping and selection shapes (best_x, top_list, x_vs_y) and three decision shapes (how_choose, is_x_worth, problem_first), plus a news shape that is published in only three of the categories used here and is left out of every count below. Thirteen categories publish all three shopping shapes, and the same thirteen publish all three decision shapes, so every vendor in this cross-section can be read across six segments of its own category.

Every rank below is the release's own published rank inside a segment, read with its denominator, never re-derived by sorting. Every segment used here has exactly ten domains at rank 10 or better, so each shape has 130 top-10 slots across the thirteen categories.

One release later, the same seven carry

For each of the 39 shopping segments the top ten domains were filtered to the vendor_owned class. The unit is the vendor–category pair, because the question is whether one brand holds across shapes inside one category; three domains reach a top ten in two categories each.

Read 2026-09-18 release 2026-09-19 release
Vendor-owned top-10 slots, best_x 9 of 130 9 of 130
Vendor-owned top-10 slots, top_list 12 of 130 12 of 130
Vendor-owned top-10 slots, x_vs_y 19 of 130 18 of 130
Vendor–category pairs 33 over 30 domains 32 over 29 domains
Pairs holding exactly one shape 26 25
Pairs holding two shapes 7 7
Pairs holding all three 0 0

The 2026-09-18 column is the published cross-section; the 2026-09-19 column is this read. One x_vs_y slot changed hands between releases and the count of single-shape pairs moved by one. The seven carriers did not move: sentinelone.com, paloaltonetworks.com and cynet.com in cybersecurity, experian.com in consumer finance, stripe.com in fintech, semrush.com and therankmasters.com in AI visibility.

What does not separate the seven from the 25

Four domain-level properties were read for every pair from the domain's own index profile. None of them orders the two groups.

Property 7 carrier pairs 25 single-shape pairs
Cited by all six engines in the window 7 of 7 19 of 25
Confidence grade 0 A, 4 B, 3 C 2 A, 8 B, 12 C, 3 collecting
Cited runs across the whole index, median 143 88
Cited runs across the whole index, range 62 to 216 25 to 304
Categories in which the domain has a published segment, median 2 3

Engine reach is saturated on both sides. Grade runs the wrong way: the only A-grade rows in the set are single-shape. Overall volume overlaps almost completely, and the domain at the top of the volume range is the clearest counter-case. ibm.com is cited in 304 runs, carries grade A, sits at rank 13 of 22,213 in the full index and has published segments in ten categories. In AI infrastructure it holds x_vs_y at rank 3 of 231 (24 of 108 runs) and reaches best_x at rank 146 of 204 (1 of 77) and top_list at rank 108 of 270 (2 of 108). In AI security and privacy it holds x_vs_y at rank 8 of 263 and nothing else in the top ten. The biggest domain in the set is a one-shape vendor twice over.

Breadth across categories runs the wrong way too. The carriers are narrower, not wider: a median of two categories with a published segment against three, and the widest domains in the set — ibm.com in ten categories, zapier.com in nine, guideflow.com in nine, hubspot.com and nvidia.com in seven — are all single-shape.

What does: standing in the category's decision questions

The property that separates the groups is inside the category, not across the index. For each pair, the domain's rank was read in the same category's three decision shapes.

Decision-shape standing in the same category 7 carrier pairs 25 single-shape pairs
Top-10 slot in at least one decision shape 6 of 7 3 of 25
Top-10 slot in two or more decision shapes 3 of 7 0 of 25
Cited in all three decision shapes 6 of 7 8 of 25
Cited in none of the three 0 of 7 5 of 25
Cited runs across the three decision shapes, median per pair 28 5
Cited runs across all six buyer shapes of the category, median per pair 72 30

Seven carrier pairs account for 239 decision-shape cited runs. Twenty-five single-shape pairs account for 223. The seven per-pair profiles, read from the segment leaderboards:

Vendor, category Shopping shapes held (rank, cited/observed) Third shopping shape how_choose is_x_worth problem_first
sentinelone.com, cybersecurity best_x #2 of 213 (27/75); x_vs_y #2 of 189 (27/107) top_list #12 of 264 (13/130) #6 of 325 (14/131) #5 of 227 (19/131) #15 of 263 (8/125)
paloaltonetworks.com, cybersecurity best_x #7 of 213 (15/75); x_vs_y #8 of 189 (15/107) top_list #44 of 264 (7/130) #131 of 325 (2/131) #1 of 227 (37/131) #7 of 263 (11/125)
experian.com, consumer finance best_x #9 of 236 (24/132); top_list #5 of 233 (31/132) x_vs_y not observed #5 of 191 (28/132) #10 of 262 (11/132) #3 of 226 (29/132)
stripe.com, fintech top_list #1 of 200 (25/102); x_vs_y #10 of 136 (11/102) best_x #119 of 224 (2/108) #14 of 170 (8/102) #4 of 206 (11/108) #15 of 171 (7/102)
semrush.com, AI visibility top_list #7 of 412 (9/71); x_vs_y #6 of 417 (20/132) best_x #13 of 351 (15/113) #39 of 448 (5/113) #44 of 467 (6/137) #5 of 420 (17/132)
cynet.com, cybersecurity best_x #3 of 213 (25/75); x_vs_y #5 of 189 (20/107) top_list #182 of 264 (1/130) #9 of 325 (11/131) not observed not observed
therankmasters.com, AI visibility best_x #3 of 351 (29/113); top_list #5 of 412 (11/71) x_vs_y #378 of 417 (1/132) #56 of 448 (4/113) #30 of 467 (9/137) #140 of 420 (2/132)

Two rows deserve their own sentence. experian.com holds a top ten in every decision shape of consumer finance and is the only carrier whose third shopping shape is not observed at all, so its carry-over runs through the decision questions and not through a general presence in the category. therankmasters.com is the carrier the property does not explain: two shopping shapes held, no decision-shape top ten, and a x_vs_y rank of 378 of 417. One of seven does not fit, and the article says so rather than dropping the row.

The three single-shape vendors that do hold a decision-shape top ten are the ones to watch. hubspot.com holds x_vs_y in AI visibility at rank 1 of 417 (35 of 132) and is_x_worth at rank 5 of 467 (19 of 137). zapier.com holds best_x in enterprise software at rank 9 of 167 (11 of 101) and problem_first at rank 2 of 258 (12 of 106). adyen.com holds top_list in fintech at rank 9 of 200 (12 of 102) and is_x_worth at rank 3 of 206 (11 of 108). If decision-shape standing and shopping-shape carry-over travel together, these three are the next carriers; if they are not, the property is coincidence at n=32.

Why this is the finding a brand team should have wanted

The obvious reading of the first cross-section was that shopping shapes are three different contests and a vendor wins the one whose question text its pages happen to match. That reading survives this read: the shapes are still separate, and no vendor holds all three. What this read adds is that the vendors who hold two are not bigger, more widely cited or better graded. They are cited when the same category's buyers ask how to choose, whether it is worth it, and what to do about a problem — questions that have no product list in the answer.

That fits what the engines say about themselves and what practitioners observe, without proving any mechanism. Google describes its AI Mode shopping experience as Gemini reasoning over a Shopping Graph of more than 50 billion product listings, and its product structured data documentation separates merchant listings from product snippets, which is a product-record layer that the decision questions do not touch. A 200-query teardown of chatgpt.com/shopping found comparison tables rendered 13 times more often than in base ChatGPT for the same queries, so the shopping surface and the conversational surface are literally different retrieval pipelines. Shopify's guide to Perplexity Shopping says the engine relies primarily on structured product data rather than visual inference, and a catalog vendor's read of Perplexity separates feed-driven product cards from organic text citations, where content quality and semantic match decide. Practitioner writing converges on the same split: comparison content wins because engines answer category-level questions, a pilot on question shape reports 86% of citations moving to comparison pages when the phrasing changes from "best X" to "X alternatives", and guides to product-page recommendation across ChatGPT, Gemini and Perplexity and to how the three engines choose products each describe engine-specific selection with a shared requirement of consistent product data. A comparison-page study argues engines treat a vendor's own comparison page as self-asserting, which would predict exactly the pattern here: the vendor's shopping-shape slot is held by pages that answer, and the decision shapes are where answering is the whole test.

None of that is evidence for this cross-section; it is context for why the shopping shapes and the decision shapes could be read by different parts of the same engine. The index measures what was cited. It does not see feeds, schema or pipelines.

What a brand-side operator can take from this

  • Read your own six segments before deciding which shape to chase. Every domain in the index has a profile at machinerelations.ai/index/domains/<domain> listing each published segment with rank and denominator. A vendor at rank 3 in best_x and unobserved in how_choose and is_x_worth is a single-shape holder by this read; the pattern in the seven says the decision questions are the ones to be cited in next.
  • Carry-over is rare and does not come with size. Seven of 32 vendor pairs carry, and the biggest, broadest domain in the set does not. A plan that assumes a strong best-tools position spreads to comparisons on its own has no support in two releases.
  • A decision-shape citation is a testable target. Which sources engines cite when buyers ask how to choose or whether a category is worth it is published per category, with the same evidence floor. A brand can freeze its rank in those three segments on one release and read the next.
  • Treat this as a baseline for an experiment, not as a lever. The cross-section says the two properties travel together in 6 of 7 cases. It cannot say which came first or whether either is under a brand's control.

Limits

The unit is 32 pairs, and the carrier group is seven. A property that fits six of seven at that size is a hypothesis with a good first read, not a result. Ranks are read from published segments only; a rate in a collecting segment is not a rank and was not used. The index observes six engines on a fixed English-language commercial basket and cannot see feed-driven product surfaces separately from text answers, so "shopping shape" here means the question text, not a shopping product. The 2026-09-18 column is the published cross-section, not a re-run on the earlier release file. The decision-shape counts exclude the news shape because it is published in only three of the thirteen categories. Nothing here measures whether any page, feed, schema change or campaign moved a rank.

Sources and method

FAQ

What is a shopping shape, and what is a decision shape? The index groups its monitored buyer questions into six shapes. best_x, top_list and x_vs_y ask for a product or a set of products and are the shopping shapes. how_choose, is_x_worth and problem_first ask how to decide, whether to spend, or what to do about a problem, and are the decision shapes. A category paired with a shape is a segment.

Does holding two shopping shapes make a vendor "carry"? Only in the sense used here: a top-10 slot in two of the three published shopping segments of one category on the same release. It is a threshold for counting, not a grade.

Is decision-shape citation the cause of shopping-shape carry-over? The cross-section cannot say. The two travel together in six of seven carrier pairs and in three of 25 single-shape pairs. The three single-shape vendors that already hold a decision-shape top ten are named above so the next read can test the direction.

Why does the biggest domain in the set not carry? ibm.com is cited in 304 runs across ten categories, but inside AI infrastructure it is cited in seven decision-shape runs and holds one shopping shape. Volume across the index and standing inside one category are different measurements, and only the second one tracks carry-over here.

Can I reproduce these numbers? Yes. The release file is public and versioned by its release id. Filter strata to published rows, take the thirteen categories where all three shopping shapes are published, read each domain's category_signals for those categories, keep vendor-owned domains at rank 10 or better, and count shapes per domain-and-category pair. Then read the same domains' ranks in the three decision shapes of the same category.