---
title: "Healthline Is the #15 Most-Cited Domain in AI Answers and Has No Source Category"
description: "In today's Machine Relations Index release, 189 of the 506 domains that clear the evidence floor carry no source role at all. Together they hold more cited answer runs than the entire rated editorial class — and the split runs straight through matched competitors."
canonical: https://authoritytech.io/curated/ai-citation-source-role-coverage-unclassified-2026
last-updated: 2026-09-20
---

# Healthline Is the #15 Most-Cited Domain in AI Answers and Has No Source Category

In today's Machine Relations Index release, 189 of the 506 domains that clear the evidence floor carry no source role at all. Together they hold more cited answer runs than the entire rated editorial class — and the split runs straight through matched competitors.

Canonical URL: https://authoritytech.io/curated/ai-citation-source-role-coverage-unclassified-2026
Published: 2026-09-20
Author: Jaxon Parrott
Tags: Afternoon Brief, AI Search & Discovery, Measurement

Every claim of the form "editorial publications hold X% of AI citations and wire services hold Y%" is computed over a classified set. Today's [Machine Relations Index](https://machinerelations.ai/index) release lets you measure how much of the citation market that set actually contains, and the answer is smaller than the claims imply.

[Healthline](https://machinerelations.ai/index/domains/healthline.com) is the 15th most-cited domain in the index — cited in 298 of 16,039 monitored answer runs, by all six observed engines. Its source role in the release is **Other observed source**. [Bankrate](https://machinerelations.ai/index/domains/bankrate.com) at #36, [WebMD](https://machinerelations.ai/index/domains/webmd.com) at #57 and [Tom's Guide](https://machinerelations.ai/index/domains/tomsguide.com) at #73 are the same: rated, cited daily, in no category.

## What the release publishes today

Release `mri_score_v2.0+2026-09-20+47973f373a20`, generated 2026-09-20, observation window 2026-05-10 through 2026-09-20, 127 days observed, 16,039 monitored answer runs across six engines (ChatGPT, Claude, Gemini, Google AI Mode, Google AI Overviews, Perplexity), 22,320 cited domains, 112,518 cited-domain observations.

A domain counts once per answer run in which it is cited. A domain publishes a rate and a confidence grade only after it clears the index's evidence floor of at least 10 observations across at least 7 distinct dates; below that line it is marked collecting rather than scored.

| Source role | Domains | Cited runs | Share of all cited-domain observations |
|---|---:|---:|---:|
| Other observed source | 18,755 | 68,238 | 60.6% |
| Editorial publication | 1,230 | 13,525 | 12.0% |
| Vendor-owned source | 823 | 11,250 | 10.0% |
| Market and company database | 689 | 6,130 | 5.4% |
| Community and social platform | 27 | 4,333 | 3.9% |
| Academic and government source | 413 | 3,889 | 3.5% |
| Analyst and consulting research | 364 | 3,404 | 3.0% |
| Search or media platform | 10 | 1,517 | 1.3% |
| Wire and press-release distribution | 9 | 232 | 0.2% |

Sixty percent of everything the index has observed being cited sits in the catch-all. That number on its own proves little — most of those 18,755 domains are long tail, cited a handful of times, below the evidence floor, and no honest taxonomy will ever reach them all.

The number that matters is what happens inside the rated set.

## Among domains that clear the floor, the catch-all is the largest single block

506 domains carry a published confidence grade of A, B or C in today's release. They account for 39,037 cited runs. Here is how they divide.

| Source role | Rated domains | Cited runs | Share of rated citation |
|---|---:|---:|---:|
| **Other observed source** | **189** | **8,991** | **23.0%** |
| Editorial publication | 99 | 8,427 | 21.6% |
| Vendor-owned source | 104 | 7,553 | 19.3% |
| Community and social platform | 10 | 4,216 | 10.8% |
| Market and company database | 49 | 3,906 | 10.0% |
| Academic and government source | 22 | 2,308 | 5.9% |
| Analyst and consulting research | 28 | 1,958 | 5.0% |
| Search or media platform | 2 | 1,469 | 3.8% |
| Wire and press-release distribution | 3 | 209 | 0.5% |

Read the top row again. The domains with no source category are the largest single block of evidence-qualified citation in the index, and they hold more of it than the entire rated editorial class — 8,991 cited runs against 8,427. Thirty-seven percent of every rated domain is uncategorized. Eighty of those 189 are cited by all six engines, against 40 of the 99 rated editorial domains.

So when a vendor study or one of our own pages says "editorial holds this much and wire holds that much," the comparison is being made against a class that is, in rated citation volume, smaller than the pile nobody classified.

## The split runs through matched competitors

If the gap were a long-tail artifact it would sit below the head and nobody would need to care. It does not. It runs straight through pairs of direct competitors at the top of the index, in three separate verticals.

| Vertical | In the editorial class | Not in any class | Rank gap |
|---|---|---|---|
| Consumer health | [medicalnewstoday.com](https://machinerelations.ai/index/domains/medicalnewstoday.com) — #216, 0.31% | [healthline.com](https://machinerelations.ai/index/domains/healthline.com) — #15, 1.86%; [webmd.com](https://machinerelations.ai/index/domains/webmd.com) — #57, 0.72% | The uncategorized one is cited six times as often |
| Personal finance | [nerdwallet.com](https://machinerelations.ai/index/domains/nerdwallet.com) — #17, 1.58%; [fool.com](https://machinerelations.ai/index/domains/fool.com) — #112, 0.46% | [bankrate.com](https://machinerelations.ai/index/domains/bankrate.com) — #36, 0.90%; [kiplinger.com](https://machinerelations.ai/index/domains/kiplinger.com) — #94, 0.52%; [investopedia.com](https://machinerelations.ai/index/domains/investopedia.com) — #178, 0.36% | Bankrate outranks Fool by 76 places and is not in the class |
| Consumer tech | [techradar.com](https://machinerelations.ai/index/domains/techradar.com) — #7, 2.87%; [pcmag.com](https://machinerelations.ai/index/domains/pcmag.com) — #53, 0.75%; [cnet.com](https://machinerelations.ai/index/domains/cnet.com) — #109, 0.46% | [tomsguide.com](https://machinerelations.ai/index/domains/tomsguide.com) — #73, 0.62% | Tom's Guide sits between PCMag and CNET, outside the class |

Each of those uncategorized domains publishes the editorial apparatus the classification is supposed to detect. Bankrate publishes an [editorial policy](https://www.bankrate.com/editorial-policy/) describing its writer and editor review chain. WebMD publishes an [editorial policy](https://www.webmd.com/about-webmd-policies/about-editorial-policy) covering medical review. Healthline publishes [who it is and how it reviews](https://www.healthline.com/about/about-us). Tom's Guide and Kiplinger both publish [mastheads](https://www.tomsguide.com/about-us) with [named editorial staff](https://www.kiplinger.com/about-us). Their classified competitors — [TechRadar](https://www.techradar.com/about-us), [CNET](https://www.cnet.com/about/), [The Motley Fool](https://www.fool.com/about/) — publish the same kind of thing.

Nothing distinguishes the two columns except which side of an incomplete classification pass each domain landed on.

## Coverage decays with rank

The classification is dense at the very top and thins fast.

| Through rank | Share carrying a source role |
|---|---:|
| 100 | 88.0% |
| 200 | 76.0% |
| 400 | 66.5% |
| 500 | 62.8% |
| 750 | 54.7% |
| 1,000 | 48.0% |
| 2,000 | 34.5% |
| 5,000 | 24.2% |

Twelve of the top 100 most-cited domains in AI answers have no source category. By rank 1,000 it is a coin flip.

This is the signature of a classification built from a seed list downward rather than from the observed data upward — the familiar shape of any taxonomy applied to a set that keeps growing. It is not evidence of bad faith in anyone's index, including ours. It is evidence that class-level conclusions carry a coverage limit that class-level headlines rarely carry with them.

## Why classifying publishers is genuinely hard

It is worth saying plainly that the problem is not unique to this index, and the existing standards do not solve it either.

[IPTC Media Topics](https://www.iptc.org/standards/media-topics/) is a subject taxonomy — it classifies what an article is about, not what kind of organization published it. The [Trust Project's trust indicators](https://thetrustproject.org/trust-indicators/) and the [Journalism Trust Initiative](https://www.journalismtrustinitiative.org/) describe transparency practices a publisher can disclose, which is a different axis from "is this an edited newsroom or a hosted publishing platform." [NewsGuard's rating criteria](https://www.newsguardtech.com/ratings/rating-process-criteria/) score credibility and transparency for news and information sites, again on their own axis. The [IAB Content Taxonomy](https://www.iab.com/guidelines/content-taxonomy/) classifies content for advertising. Google's own [guidance on AI features in Search](https://developers.google.com/search/docs/appearance/ai-features) and its [statements about generative results](https://blog.google/products/search/generative-ai-google-search-may-2024/) describe eligibility and linking behavior rather than any published source-type register.

There is no widely adopted register that answers "what type of source is this domain" for the whole open web. Everyone measuring AI citations is building that column themselves, and everyone's version will have a coverage boundary like this one.

## The part that applies to us

We publish this index, and we have been quoting its source-role column. Two limits are now on the record, both from today's release:

1. **The class contains things that do not belong in it.** Medium is filed as an editorial publication and ranks [#1 of 1,230](https://machinerelations.ai/index/domains/medium.com) of them, while Substack, beehiiv and dev.to — the same hosted open-publishing product — are filed as community platforms. Jaxon wrote that up in full in [the #1 "editorial publication" in our own index has no editors](https://jaxonparrott.com/blog/medium-number-one-editorial-publication-ai-citations-classifier).
2. **The class omits things that do belong in it**, which is this page.

Both defects point the same way: a source-role rank or a class-share percentage from this release describes a partial, imperfectly bounded set. The per-domain citation rates are unaffected — those are direct observations over a fixed run set, and Healthline's 298 cited runs in 16,039 are the same number whatever label sits next to them.

So the practical rule, for us and for anyone reading a class-level AI citation statistic:

- **Argue on the measured rate and the engine breadth.** Those are observations.
- **Treat class rank and class share as provisional**, and name the coverage limit on the page where you use them.
- **Check whether your own domain carries a source role at all** before reading any class leaderboard as a competitive set. On a [domain profile](https://machinerelations.ai/index/domains/healthline.com), the "Evidence-qualified source role" line tells you which population you are actually being ranked inside — Healthline's reads #1 of 189, because 189 is the size of the uncategorized rated set.

The classifier is one release away from being better. The habit of quoting a class share without its denominator is the thing that needs to change permanently.

*Measured from Machine Relations Index release `mri_score_v2.0+2026-09-20+47973f373a20`, artifact sha256 `47973f373a2058a8eff0e36d1cb81d29237468044e9419ffab03876a817b11e6`, read 2026-09-20. Domain classifications and ranks verified against the live per-domain profiles the same day.*

## Links

- [Curated Index](https://authoritytech.io/curated.md)
- [Home](https://authoritytech.io/index.md)