---
title: "Do AI Engines Cite Subdomains? 7.9% of Citations, and Docs Sit on Subdomains While Blogs Sit in Folders"
description: "7.9% of 197,384 AI-cited URLs sat on a subdomain. 96.5% of blog citations were in a /blog/ folder; docs went the other way. By engine, question and site type."
canonical: "https://machinerelations.ai/research/do-ai-engines-cite-subdomains-or-subfolders-2026"
---

# Do AI Engines Cite Subdomains? 7.9% of Citations, and Docs Sit on Subdomains While Blogs Sit in Folders

7.9% of 197,384 AI-cited URLs sat on a subdomain. 96.5% of blog citations were in a /blog/ folder; docs went the other way. By engine, question and site type.

Canonical URL: https://machinerelations.ai/research/do-ai-engines-cite-subdomains-or-subfolders-2026
Published: 2026-11-08
Research type: Index Analysis
Tags: citation-behavior, url-structure, site-architecture, machine-relations-index

## Current Index data

Updated from the Index release through **2026-11-08** (mri_score_v2.1); observation window 2026-05-10–2026-11-08. This card updates with the published dataset. The article below remains a dated analysis; these figures do not retroactively validate its claims.

- [g2.com](https://machinerelations.ai/index/domains/g2.com): 2.16% citation rate; 809 of 37,536 observed runs, tier A.
- [nih.gov](https://machinerelations.ai/index/domains/nih.gov): 1.91% citation rate; 716 of 37,536 observed runs, tier A.
- [yahoo.com](https://machinerelations.ai/index/domains/yahoo.com): 1.87% citation rate; 702 of 37,536 observed runs, tier A.
- Engines named here and observed in this dataset: ChatGPT, Claude, Gemini, Google AI Mode, Google AI Overviews, Perplexity. Coverage varies by engine; see the [release manifest](https://machinerelations.ai/data/mri-release-manifest.json) for returned runs and availability.

Citation presence measures observed source selection, not endorsement or causation. [Explore the Index](https://machinerelations.ai/index) · [Read the latest weekly report](https://machinerelations.ai/research/machine-relations-index-report).

## Source Body

AI engines cite subdomains, and not often. In the Machine Relations Index answer ledger, 15,531 of 197,384 cited-URL events (7.9 percent, 95 percent interval 7.4 to 8.4) pointed at a subdomain such as blog.example.com or docs.example.com. The other 92.1 percent were on the main domain or its www host. The share barely moves with the engine (6.2 to 8.3 percent) or the question (6.9 to 9.5 percent). What moves it is what the subdomain is for and who owns it.

Blogs live in folders. Of the cited addresses that sit in a blog section, 96.5 percent (49,499 events on 8,447 domains) were in a /blog/ or /blogs/ folder on the main domain and 3.5 percent (1,785 events on 305 domains) were on a blog subdomain. Documentation goes the other way: 67.3 percent of cited docs addresses (967 events on 207 domains) were on a docs, developer or API subdomain, and 32.7 percent (469 events on 103 domains) in a /docs/ folder. Community, store and shop sections lean the same way as docs. Government and university sources are mostly on subdomains: 65 percent of .gov and 66 percent of .edu citations, against 6.8 percent for .com and 2.9 percent for .ai.

This page counts where engines pointed. The ledger has no count of how many sites run a blog on a subdomain and how many in a folder, so a 3.5 percent share of blog citations is not a finding that engines cite subdomain blogs less per page. It is how the citations are distributed.

## What was counted

- **Population.** Successful runs in the Index's buying-question panel, 2026-07-06 to 2026-11-07, on its 1,008 buying questions (168 in each of six question shapes, 24 industries). Runs flagged as naming a brand in the prompt are excluded. The ledger export is dated 2026-11-08. The 197,384 events and 23,183 runs are the same as [the site-sections study](/research/which-site-sections-ai-engines-cite-2026), [the URL slug study](/research/url-slug-length-ai-engines-cite-2026), [the year-in-the-URL study](/research/year-in-cited-url-ai-engines-2026) and [the social-platform study](/research/do-ai-engines-cite-reddit-and-youtube-2026).
- **Engines.** ChatGPT, Claude, Gemini, Google AI Mode and Perplexity, called through their APIs. This is not the consumer ChatGPT or Claude app. Google AI Overviews are left out. The engines do not answer the same questions: Perplexity cites on all 1,008, ChatGPT on 999, Claude on 895, Gemini on 436 and Google AI Mode on 138.
- **Unit.** One cited URL in one answer run, counted once however often the run repeated it. The set holds 78,194 distinct addresses on 29,282 root domains.
- **A subdomain.** The cited address's host, with a leading www removed, is not the same as its registered root domain. pmc.ncbi.nlm.nih.gov is a subdomain of nih.gov. www.example.com and example.com are the same host.
- **No page was fetched.** Subdomain and folder are facts about the address, not about the page.

## One share, flat across engines and questions

Share of cited addresses on a subdomain, by engine (95 percent intervals resample the questions the engine cited on, 2,000 draws, fixed seed):

| Engine | Cited URLs | Subdomain events | Share |
|---|---:|---:|---:|
| Perplexity | 105,295 | 8,762 | 8.3% (7.7 to 9.0) |
| Gemini | 17,416 | 1,426 | 8.2% (7.4 to 9.1) |
| ChatGPT | 35,357 | 2,840 | 8.0% (7.4 to 8.7) |
| Google AI Mode | 6,124 | 444 | 7.3% (5.6 to 9.1) |
| Claude | 33,192 | 2,059 | **6.2%** (5.6 to 6.8) |

Claude is the low engine, and its interval does not overlap Perplexity's. The other four are within each other's intervals. By question shape:

| Question shape | Cited URLs | Subdomain events | Share |
|---|---:|---:|---:|
| Problem first | 26,258 | 2,506 | **9.5%** (8.1 to 11.2) |
| X vs Y | 30,246 | 2,471 | 8.2% (6.8 to 9.6) |
| How to choose | 31,480 | 2,521 | 8.0% (6.7 to 9.6) |
| Best X | 38,847 | 3,053 | 7.9% (7.0 to 8.8) |
| Is it worth it | 30,652 | 2,213 | 7.2% (6.1 to 8.5) |
| Top lists | 39,901 | 2,767 | 6.9% (6.2 to 7.7) |

Problem-first questions, where the buyer describes a situation and names no product, draw the most, and their interval does not overlap the top-list interval. The spread is 2.6 points. Counting each distinct address once, 9.0 percent of the 78,194 addresses (7,073) are on a subdomain.

## What lives on subdomains and what lives in folders

The same section of a site can sit in either place. This table takes seven sections and counts cited addresses on a subdomain whose first label names the section (blog or blogs; docs, doc, documentation, developer, developers, dev, api or devdocs; help, support, kb, faq, helpcenter, answers, knowledge or helpdesk; community, forum, forums, discuss, discourse or ask; learn, academy, training, university, education or courses; news, press, newsroom, investor, investors, ir or media; shop or store) against addresses on the main domain whose first folder is the same word (after dropping language folders such as /en/ and /de-de/).

| Section | On a subdomain: events (domains) | In a folder: events (domains) | Share on a subdomain |
|---|---:|---:|---:|
| Blog | 1,785 (305) | 49,499 (8,447) | 3.5% |
| News or press | 225 (58) | 2,239 (537) | 9.1% |
| Learn or academy | 659 (30) | 2,470 (341) | 21.1% |
| Help or support | 471 (140) | 819 (152) | 36.5% |
| Community or forum | 397 (86) | 251 (40) | 61.3% |
| Docs or developer | 967 (207) | 469 (103) | 67.3% |
| Shop or store | 266 (61) | 143 (49) | 65.0% |

Two cautions come before any reading. The first is concentration. On the blog subdomain side, three domains hold 671 of the 1,785 events, and one of them, hubspot.com, holds 541: in this ledger HubSpot's cited blog addresses are on blog.hubspot.com and none is at hubspot.com/blog. For learn subdomains, three domains hold 601 of 659. For docs, nvidia.com (181), amazon.com (122) and github.com (90) hold 393 of 967. In the blog folders the top three domains hold 1,019 of 49,499 (2.1 percent), so the folder pattern is a habit across thousands of sites while the subdomain pattern is a handful of large publishers plus a long tail. The second caution is the denominator. The folder columns for docs, community and shop rest on 40 to 103 domains, so the shares in those rows are suggestive and not stable.

The shape that holds is this: blog content is cited from folders almost everywhere, while documentation, community and store content is cited from subdomains about two times in three. Those are also the sections that tend to run on separate systems from the marketing site.

## Site type changes the answer more than engine or question

Share of cited addresses on a subdomain, by the top-level domain of the root domain:

| Site type | Events | Subdomain events | Share | Root domains (with a subdomain citation) |
|---|---:|---:|---:|---:|
| .gov | 1,846 | 1,208 | **65.4%** | 145 (66) |
| .edu and academic | 1,662 | 1,097 | **66.0%** | 268 (193) |
| .org | 10,047 | 1,268 | 12.6% | 1,171 (131) |
| .com | 148,868 | 10,103 | 6.8% | 20,413 (1,300) |
| .io | 7,177 | 343 | 4.8% | 1,083 (74) |
| .ai | 9,677 | 276 | 2.9% | 1,543 (99) |
| All other | 18,107 | 1,236 | 6.8% | 4,659 (383) |

Government and university institutions publish through department and journal hosts. The three largest .gov and .edu subdomain hosts are pmc.ncbi.nlm.nih.gov (539 events), health.harvard.edu (189) and image-ppubs.uspto.gov (95). Together .gov and .edu supply 2,305 subdomain events, 14.8 percent of the total, from 1.8 percent of all citations. The .ai and .io domains, which are mostly software vendors, sit at 2.9 and 4.8 percent. Taking out the ten root domains with the most subdomain events, the .com, .io, .ai and .co domains are 5.0 percent subdomain across 164,581 events.

## Concentration

The subdomain share is carried by a few large hosts. The ten root domains with the most subdomain events (nih.gov 773, yahoo.com 582, hubspot.com 545, apple.com 424, microsoft.com 347, g2.com 319, amazon.com 307, clevelandclinic.org 292, wikipedia.org 260, harvard.edu 248) hold 4,097 of the 15,531 events, 26.4 percent. Without those ten the subdomain share of the rest is 5.9 percent (11,434 of 192,455).

It is also a minority habit. 2,246 of the 29,282 root domains (7.7 percent) had at least one cited address on a subdomain. Among the 1,719 domains cited at least 20 times, 312 (18.1 percent) did, and for 90 (5.2 percent) a majority of their citations were on subdomains.

## What this changes

- **Do not move a blog on this evidence.** 305 domains had a blog subdomain cited, 1,785 times. Nothing here shows a blog subdomain is cited less per page than a blog folder, and the ledger cannot, because it has no count of pages that exist.
- **Do not build a new blog on a subdomain because HubSpot did.** One domain is 30 percent of blog-subdomain citations. The pattern most sites show is the folder.
- **Documentation and community are different.** When the docs, help or community site is its own host, engines cite it, and in this ledger that is the more common arrangement for docs. The question for those sections is whether the pages answer a buyer question, not which host they sit on.
- **A cited host is not a recommendation.** A subdomain citation may be a vendor's own help page, a news item or a third party's review.
- **The question and the engine still dominate.** 92.1 percent of cited addresses are on the main domain whichever engine is asked. The larger gaps in this series are in [what each engine cites](/research/do-ai-engines-cite-same-sources-2026) and in [which sections of a site get cited](/research/which-site-sections-ai-engines-cite-2026).

## What this page does not show

- **Cause.** A page is cited because of what it says and whether an engine retrieves it. This page does not test whether moving a section between a subdomain and a folder changes any citation.
- **Preference.** The ledger counts citations, not the sites that could have been cited. A rare architecture can be cited rarely because it is rare.
- **Crawling and indexing.** How engines or search crawlers treat subdomains is not measured.
- **Consumer products.** The engines are called through their APIs. The consumer apps and Google AI Overviews may cite differently, and Google AI Overviews are left out of every number here.
- **Buyer demand.** The 1,008 questions are the Index's own panel, written for coverage of buying decisions. They are not a sample of what buyers type.
- **Section names.** A section is identified by the word in the host or folder. A /resources/ or /insights/ folder, or a subdomain called something else, is not counted in the section table. The sections page covers folders in more detail.
- **Trend.** The page does not test whether these shares are changing.

## Method note

Source: the Index answer ledger's citation-events table, daily export of 2026-11-08, read with DuckDB, joined to the run table (successful runs only) and the question table (the buying-question panel, for the question shape). Filters: the buying-question panel, runs not flagged as brand-prompted, a non-empty cited URL, engines ChatGPT, Claude, Gemini, Google AI Mode and Perplexity, run dates 2026-07-06 to 2026-11-07. Unit: distinct (answer run, cited URL) pairs, 197,384, from 23,183 runs and 1,008 questions; the counts reproduce the earlier studies exactly before any new cut. Host: the cited address's host, lowercased, with a leading www removed. Subdomain: the host differs from the registered root domain in the ledger. Section on a subdomain: the first label of the host. Section in a folder: the first path segment of a main-domain address, after removing up to two leading language segments (two letters, or two letters, a hyphen and two to four letters). Shares are pooled event counts, not averages of per-question shares. Intervals resample the questions in a group with replacement, keeping each question's cited addresses together, 2,000 draws, fixed seed. Site type is read from the root domain's ending; .edu includes academic domains ending .ac. The 1,785 blog-subdomain events here count blog and blogs hosts; [the sections study](/research/which-site-sections-ai-engines-cite-2026) counts blog hosts only (1,677) and counts a folder on any host, with further folder names such as articles and posts in its blog group, so its folder counts are not comparable to the main-domain folder counts here.

## Frequently asked questions

### Do AI engines cite subdomains?

Yes, but for a minority of citations. In the Machine Relations Index ledger, 7.9 percent of 197,384 cited URLs were on a subdomain. 2,246 of 29,282 cited root domains (7.7 percent) were cited on one at least once.

### Is a subdomain or a subfolder better for AI citations?

This data does not say. It shows that 96.5 percent of cited blog addresses are in folders and that documentation, community and store addresses are more often on subdomains. It cannot say which arrangement is cited more per page, because it does not count the pages that exist or the sites that use each arrangement.

### Should a blog be on a subdomain or in a /blog/ folder?

In this ledger the folder is the common case: 49,499 cited blog-folder events on 8,447 domains against 1,785 events on 305 domains for blog subdomains, 30 percent of which are one domain. Nothing in it shows a reason to move an existing blog.

### Do .gov and .edu sites get cited on subdomains?

Mostly. 65.4 percent of .gov and 66.0 percent of .edu citations were on a subdomain, for example pmc.ncbi.nlm.nih.gov, health.harvard.edu and image-ppubs.uspto.gov, against 6.8 percent for .com.

### Does the engine matter?

A little. Claude had the lowest subdomain share at 6.2 percent and Perplexity the highest at 8.3. The question matters about as much: 6.9 percent on top-list questions and 9.5 percent on problem-first questions.

## Attribution

This research is published by Machine Relations Research, the research program of machinerelations.ai — the public research initiative that publishes the glossary, research, evidence, and measurements for the Machine Relations discipline. Provenance and editorial standards: https://machinerelations.ai/about

## Machine-readable related links

### Related concepts

- [Machine Relations Index (MRI)](https://machinerelations.ai/glossary/machine-relations-index)
- [AI Citations](https://machinerelations.ai/glossary/ai-citations)
- [Machine Relations (MR)](https://machinerelations.ai/glossary/machine-relations)
- [MRI Score](https://machinerelations.ai/glossary/mri-score)

### Supporting research

- [AI Engines Cite Blog Folders Ten Times as Often as Product Folders: 30.6% of Cited Addresses Against 3.1%](https://machinerelations.ai/research/which-site-sections-ai-engines-cite-2026)
- [AI Engines Cite Five-Word URL Slugs: 58% of Cited Addresses Have 3 to 7 Words, and Problem-First Questions Pull the Longest](https://machinerelations.ai/research/url-slug-length-ai-engines-cite-2026)
- [AI Engines Cite URLs That Match the Question: 'vs' in 37% of Head-to-Head Citations, 'best' or 'top' in 38% of 'Best X' Ones](https://machinerelations.ai/research/cited-url-slug-matches-question-type-2026)
- [A Year in the URL: 2026 in 5.5% of the Addresses AI Engines Cite, 2025 in 1.1%](https://machinerelations.ai/research/year-in-cited-url-ai-engines-2026)

### Framework context

- [Machine Relations Stack](https://machinerelations.ai/stack)
- [Evidence Base](https://machinerelations.ai/evidence)