Machine Relations Index Report, September 18, 2026: community platforms lead 26 of 85 measured segments and the most-cited domain is absent from two engines

In the September 18, 2026 release of the Machine Relations Index (mri_score_v2.0+2026-09-18), community and social platforms hold or share first place in 26 of the 85 published category-and-question segments. Editorial publications hold or share first place in 12. The single most-cited domain across all six engines, reddit.com at a 12.75% citation rate over 2,012 answer runs, is cited by Gemini, Google AI Mode, Google AI Overviews and Perplexity, and by neither ChatGPT nor Claude in the 125-day window.

This is the first edition of the weekly Index Report. Each edition reads one daily release of the public Index and reports four things: who leads each measured segment, where the six engines differ, what the Index can and cannot yet measure, and what moved since the last edition. Every figure below is taken from the public JSON release at machinerelations.ai/data/machine-relations-index.json, generated 2026-09-18, window 2026-05-10 to 2026-09-18.

The release at a glance #

A segment is one subject category paired with one question shape. The six buyer shapes are best tools (best_x), how buyers choose (how_choose), is it worth it (is_x_worth), problem-first research (problem_first), top lists (top_list) and comparisons (x_vs_y). Seven categories also carry a news-shaped segment (news_topic) that has been sampled far longer, at 596 to 2,138 runs across 53 or 54 run dates. A citation rate is the share of a segment's observed answer runs in which at least one engine cited the domain.

The ten most-cited domains across the whole Index #

Rates here are measured over all 15,782 observed runs, so they are lower than any single segment's rate. Grade A means the domain clears the evidence floor with the highest volume tier of observations. Per-domain profiles for every ranked domain are published under machinerelations.ai/index/domains.

Who leads each measured category, shortlist shape against objection shape #

The Index's standing finding is that the question shape, not the outlet's authority, decides which class of source an engine cites. The cleanest way to see it is to place the leader of a category's shortlist question (best tools) beside the leader of the same category's objection question (is it worth it). Thirteen categories publish both.

Read across the rows. Editorial publications take the shortlist question in four categories (Medium, Forbes twice, TechRadar) and the objection question in none. Reddit takes the objection question in seven of thirteen categories. Cybersecurity is the clearest vendor-owned category on both shapes: Huntress leads best-tools at 58.7% of 75 runs, the highest single-segment rate in the release, and Palo Alto Networks leads is-it-worth-it. Each category's full per-shape leaderboard, with denominators, is published under machinerelations.ai/index/categories.

First place by source class, all 85 published segments #

Counting each segment's first-place domain by source role, and awarding a shared first place to every domain tied on rate:

The column sums to 89 because four segments have a shared first place. Reddit alone leads or shares first in 21 segments and YouTube in 10. The per-domain column is the sharper number: 27 community domains are cited 159.8 times each on average, against 10.9 for the 1,222 editorial publications. Two boundaries apply. The unclassified class is the long tail of 18,623 domains the Index has observed but not yet assigned a role, so its 24 first places are a statement about classification backlog, not about a source type. And the editorial class includes media properties operated by vendors; the Index publishes the class as measured and the domain profile says who runs each domain.

The seven news-shaped segments #

Seven categories carry a news-topic segment that has been sampled since May on a fixed cadence. These are the most settled numbers in the release.

Healthcare is the one news segment where an academic and government source leads by a wide margin, and the only one where a trade publication is second. In cybersecurity, fintech and martech the news question is led or closely followed by vendor-owned domains. Healthcare Services, HR & Talent and Martech & Advertising publish their news segment while every one of their buyer-question segments is still collecting, so a reader should never average a category across shapes.

Where the six engines differ #

The public release records, for each domain, which of the six engines cited it at least once in the window. Nineteen of the twenty-five most-cited domains are cited by all six. The six exceptions are the engine story of this release.

The pattern splits on retrieval lineage. The two Google surfaces and Gemini cite Reddit and Facebook, and ChatGPT and Claude do not, across 15,782 runs. The two Google surfaces are also the two that do not cite TechRadar, which ChatGPT, Claude, Gemini and Perplexity all do. Perplexity is the only engine that appears in every row. The release records per-engine presence in the window, meaning whether each engine was observed citing the domain at least once, not a per-engine rate; a per-engine citation rate is not part of the public view and is not inferred here.

What the Index cannot yet say #

Coverage is 85 of 151 measurable segments. The 66 unpublished segments are two different states wearing one label. Seventeen have observed runs and are accumulating toward the floor, concentrated in three categories: Healthcare Services on all six buyer shapes at 72 to 102 runs across 4 to 6 run dates, HR & Talent on all six at 35 to 72 runs, and iGaming & Betting on five. Forty-nine have zero observed runs, across Industrial, Local Services, Logistics & Freight, Martech & Advertising, Physical Consumer Products, Professional Services, Sales GTM & CRM, VC & Private Equity, and the comparison shape in iGaming. A zero-run segment says the prompt basket has not reached that category yet; it says nothing about how engines behave there.

The segments nearest to publication are five Healthcare Services buyer shapes at six of the seven required run dates: problem-first at 102 runs, best-tools, how-buyers-choose and is-it-worth-it at 101 each, and top lists at 78. If sampling continues on cadence they clear the floor on their next run date, and the next edition of this report will carry their first published leaders.

Movement since the previous release #

This edition sets the baseline for per-domain movers; the second edition will report rank changes for every domain in the top 100 of each published segment. What can be read now is the universe and the sampling.

The cited-domain universe grew by 398 domains, 1.8%, over the five daily releases from September 14 to September 18. Against the September 17 release, observed runs rose from 15,678 to 15,782 and citation events from 123,538 to 124,397, while the published segment count held at 85. Daily observed runs in the drift-monitoring sample over the last eight run dates were 48, 42, 38, 36, 48, 48, 18 and 32, with September 17 the lightest sampling day in the series.

Method notes #

The Index observes six answer engines against a fixed basket of 912 eligible buyer and research queries across 24 measured subject categories plus a legacy news bucket. Every domain an engine cites is measured on equal terms; membership in any external catalog is recorded as a label and never used to decide what is scored. A segment publishes only after 10 observed runs across 7 distinct run dates. The public release reports rates, rankings and evidence counts and excludes raw cited URLs, internal query identifiers and provider payloads. First-place counts in this report break ties by awarding first place to every domain sharing the top rate in a segment. Rates are rounded to one decimal place in the segment tables and two in the whole-Index table, as published.

The methodology is described in full at machinerelations.ai/index and the release contract at machinerelations.ai/machine-relations-index.md. The engine set follows each provider's own published crawling and citation documentation: OpenAI's crawler documentation, Google Search Central on AI features, Anthropic's crawler documentation and Perplexity's crawler documentation. Two companion studies from the same release read one category in depth: the editorial citation share by category and the how-buyers-choose editorial gap in AI visibility.

FAQ #

What is the Machine Relations Index Report? A weekly reading of the daily public release of the Machine Relations Index: segment leaders, engine differences, coverage and movement, each figure tied to a release id so it can be re-checked.

Which engines does the Index observe? ChatGPT, Claude, Gemini, Google AI Mode, Google AI Overviews and Perplexity, in the same window and against the same query basket.

What does a citation rate mean? The share of a segment's observed answer runs in which an engine cited the domain. A 46.8% rate for medium.com in AI Infrastructure best-tools means 36 of 77 observed runs cited it. It measures engine source selection only, not traffic, not sentiment and not whether the cited page supported the answer.

Why does the Index publish 85 segments and not 157? Six of the 157 are legacy-unmapped buyer-shape pairings that receive no prompts by construction. Of the 151 measurable, 85 have cleared the evidence floor, 17 are collecting with observed runs and 49 have not yet been sampled.

Does a first-place domain in a segment mean it is the best source? No. It means engines selected it most often when answering that shape of question in that category during the window. Several first places belong to unclassified long-tail domains and to vendor-operated media properties.

Where is the per-domain data? Every ranked domain has a public profile under machinerelations.ai/index/domains, every category and shape a leaderboard under machinerelations.ai/index/categories, and the full release is the public JSON linked at the top of this report.