I Would Not Treat Medium as Earned Media

I would quote Medium's citation count and I would not call it earned media. The Machine Relations Index files medium.com as a community and social platform, not an editorial publication. On release mri_score_v2.1+2026-11-04+94f24cb38da0, medium.com is cited in 1,047 of 35,834 monitored answer runs, 2.92%, on 148 of 172 days, by all six engines, at confidence A. It is the fifth most-cited of 35,814 observed domains. On the 2026-09-20 release the Index listed it at class rank 1 of 1,230 in the editorial class. On this release it has no editorial class rank, because the Index files it outside that class.

The domain record I would open before repeating any Medium number is Medium in the Machine Relations Index. That page is the deeper resource, the dataset. This essay is how I would use it as a founder: which number goes on a slide, which noun stays off it, and what the label settles.

The release, and the peers that make the label visible

Release mri_score_v2.1+2026-11-04+94f24cb38da0, window May 10 to November 4, 2026, 172 observed days, 35,834 monitored answer runs across six engines, 35,814 observed cited domains. A domain counts once per answer run in which it is cited. Each domain carries a confidence grade of A, B, C or Collecting. A Collecting domain shows its overall rate and cited runs on its profile but no rank. The release marks window coverage uneven for three engines: Gemini returned citation-bearing answers on 147 of 172 observed days, Google AI Mode on 129 and Google AI Overviews on 112. A rate here is a stated rate on its stated denominator, nothing more.

Domain	Source role in this release	Citation rate	Cited runs	Days cited	Engines	Confidence	Standing
forbes.com	Editorial publication	3.46%	1,241	150 of 172	6	A	class rank 1 of 1,293 editorial publications
medium.com	Community and social platform	2.92%	1,047	148 of 172	6	A	5 of 35,814 observed domains
amazon.com	Other observed source	1.31%	468	104 of 172	6	A	16 of 35,814 observed domains
dev.to	Community and social platform	1.11%	399	89 of 172	6	A	19 of 35,814 observed domains
substack.com	Community and social platform	0.96%	345	112 of 172	6	A	23 of 35,814 observed domains
prnewswire.com	Wire and press-release distribution	0.74%	264	108 of 172	6	B	37 of 35,814 observed domains
apple.com	Other observed source	0.61%	220	38 of 172	6	B	48 of 35,814 observed domains
wordpress.com	Other observed source	0.05%	19	19 of 172	6	Collecting	Collecting, no rank
blogspot.com	Vendor-owned source	0.03%	12	12 of 172	4	Collecting	Collecting, no rank

Medium's 1,047 of 35,834, the 148 days, the six engines and confidence A do not depend on the class. I would cite them without apology. The class label is a separate layer. For Blogspot, an engine count of 4 means Claude and Google AI Mode returned no citation to it in this window. The release marks Google AI Mode's window coverage uneven, so that absence is not a finding about the engine. Substack is a publishing platform for independent writers. Dev.to is a community for software developers. WordPress.com hosts sites for anyone who signs up. Medium is that kind of product, and the Index now files Medium with Substack and Dev.to. Forbes has a masthead, editors who can refuse a piece, and the class rank that goes with them. The masthead is the difference a source-role column is supposed to encode. The editorial class has 1,293 rows. Membership rests on site evidence for 989 of them and on a reviewer's decision for 304.

How the 2026-09-20 release filed these six domains

On the 2026-09-20 release, mri_score_v2.0+2026-09-20+47973f373a20 (window May 10 to September 20, 2026, 127 observed days, 16,039 answer runs, 22,320 cited domains), six domains with structural peers in other classes sat inside the editorial class: medium.com at class rank 1, amazon.com at 7, prnewswire.com at 8, apple.com at 16, and wordpress.com and blogspot.com, which sat below the floor for a rank. Together they held 1,358 of the class's 13,525 cited runs, 10.04%. On the 2026-11-04 release none of the six is in the editorial class. Medium sits with the community platforms, PR Newswire with the wires, Amazon, Apple and WordPress.com with other observed sources, and Blogspot with vendor-owned sources. The editorial class is 1,293 domains with 20,766 of the 217,583 domain-run citations across the nine classes. A run that cites two domains counts twice in those totals.

The Index's own record states that a named role rests on a reviewer's decision, a qualifying domain suffix or evidence read from the domain's own site, that in the 2026-10-28 build a pattern in a domain name alone does not establish a role, and that AuthorityTech's placement catalog is not a classification input. In the 2026-11-04 file, 143 of the 1,293 editorial domains carry the catalog label. The label is recorded and never used to decide what is scored. I would read that record before trusting any class label.

I do not read the two releases against each other as a trend. Medium's rate is 2.92% here and was 5.04% there, but the window grew from 127 to 172 days, the run set from 16,039 to 35,834, and the classes changed. A different denominator is a different number. I read each release by its own id.

What I would still do

People act on the quotable row. "Medium is the number one editorial publication AI cites" was the sentence most likely to escape a briefing, and read as written it said that publishing on Medium is earned media. Earned media means somebody who does not work for you decided the piece was worth publishing. A Medium post is something you wrote and published yourself, on a host that did not gate it. The Index files Medium with community platforms. The citation rate is real. The distribution is real. The host is a place I can publish, and a retrieved one.

A leaderboard is only as trustworthy as a reader's ability to check it. The Index is public, including the release file and the per-domain profile. I would open both before I repeated a rank. If I cannot show a colleague the observation and the classification as separate fields, I do not yet understand the row.

Before I quote any "top publications" list, including ours, I would open the top row and ask what the publication is. If the answer is a place where anyone can post, I am looking at a host, and I would read every class rank below it with that host removed from the noun. Regrouping can move a leaderboard without changing a count. Medium's class rank differs between these two releases because its class differs, and that is the reason the research note I would hand someone who wants it shown formally is the classification-sensitivity note.

Independent counts of what answer engines cite are published by BuzzStream, Fractl, Fullintel, Presenc AI, and Ahrefs. Each uses its own sample and its own denominator. I would read the method before I set any of those percentages next to Medium's 2.92% of 35,834 runs. A news-publication share from one firm's crawl and a class share from this release are not the same fraction, and averaging them would recreate the class-label problem in a new costume. The mechanism paper I would hand a writer who wants to know why some pages extract more cleanly than others is the GEO paper from KDD 2024. It is about extraction. It is not a ruling on whether Medium is a newsroom.

The habit of keeping a count and a label in different sentences is the one I want from Machine Relations. When I need a word for the citation rate that does not smuggle the class in with it, I use the glossary.

The decision

I would cite medium.com as 1,047 of 35,834 runs, 2.92%, 148 of 172 days, six engines, confidence A, fifth of 35,814 observed domains, on release mri_score_v2.1+2026-11-04+94f24cb38da0. I would call it a community and social platform, which is what the Index calls it. I would not move budget onto Medium to collect an earned-media placement. A post there can be cited, and the rate says the host is retrieved often. The citation does not turn the post into a newsroom decision. Using Medium as a distribution channel for a piece we were going to publish ourselves is fine, and I would report it as owned.

Scope: reading source-role classes in release mri_score_v2.1+2026-11-04+94f24cb38da0, and any release that places hosts in the editorial class. The same rule applies to Amazon at 468 cited runs, PR Newswire at 264 and Apple at 220 when someone calls those editorial wins. It does not say the 1,047 is unreal.

What would change my mind: a release that files Medium in the editorial class on a written rule, with evidence that an editor there can refuse a piece. I would then quote the new rank and read the rule. I would not change my mind because Medium's rate rose. A higher rate makes the host more important as a retrieved domain. It does not create an editor.

FAQ

Why does Medium rank first among editorial publications?

It does not on this release. On mri_score_v2.1+2026-11-04+94f24cb38da0, medium.com is filed as a community and social platform and ranks 5 of 35,814 observed domains: 1,047 of 35,834 runs, 2.92%, 148 of 172 days, six engines, confidence A. Forbes is the first editorial publication, at class rank 1 of 1,293. On the 2026-09-20 release the Index filed Medium as editorial and ranked it first in that class. The two releases file the same domain under different classes. The measured unit is domain-level citation presence in this run set. It does not establish why any engine selected the domain, and it does not promise a future citation, a recommendation or revenue.

Does publishing on Medium count as earned media?

It counts as a post we published on a host that does not gate the piece. Earned media is an independent party's decision to publish. Medium's citation rate can still be real, and I would still report the 1,047 runs. Forbes at 1,241 runs and 3.46%, with a masthead, is the comparison that keeps the noun honest.

Does the classification change the citation rates?

Not within a release. Rate, cited runs, days, engines and confidence are per-domain observations over that release's run set. The class changes anything computed across a class: an editorial share, a class rank, an editorial-versus-wire comparison. Six domains held 1,358 of the editorial class's 13,525 cited runs on the 2026-09-20 release, and none of them is in the class on the 2026-11-04 release.

What should I use instead of the class rank?

The per-domain figures: citation rate against the stated denominator of 35,834 runs, cited runs, days cited out of 172, engine breadth and confidence. If I need a comparison across source types, I would use the Index's own classes and read the release id with them. I can check the row in the public release file and on the domain profiles for Medium, Substack, Amazon, Apple and PR Newswire. Independent compositions from other firms use other samples. I would read their methods before I compared a percentage to 2.92%.