Three in ten of the addresses AI engines cite sit in a blog or article folder. In the Machine Relations Index answer ledger, 60,306 of 197,384 cited-URL events (30.6 percent) had /blog/, /blogs/, /articles/, /article/, /post/ or /posts/ as the first folder of the path. Product, solutions, services, features and platform folders held 6,215 (3.1 percent). Homepages held 4,478 (2.3 percent). Three in four runs (76.2 percent) cited at least one blog-folder address; one in seven (15.1 percent) cited a product-folder address; one in ten (9.9 percent) cited a homepage.
The engines do not agree. Claude puts 41.6 percent of its citations in blog folders and ChatGPT 18.0 percent. ChatGPT cites a homepage in 8.7 percent of its citations; Claude, Perplexity and Google AI Mode cite one in about 0.6 to 0.8 percent. In the AI visibility and GEO vertical, 45.0 percent of cited addresses are blog or article folders and 1.6 percent are product or solutions folders.
A further 18.1 percent of cited addresses have no folder at all: the page is a single slug directly under the domain (example.com/best-crm-software). Writing a blog is not what the citations track. Having a page with its own address that answers one question is.
This page counts the addresses engines cite. It does not show that engines prefer one folder: the ledger has no count of how many pages of each kind exist on the web, so a share of citations is not a rate of preference.
What was counted #
- Population. Successful runs in the Index's "mri_v2" panel, from 2026-07-06 to 2026-11-07, on its 1,008 buying questions: 168 in each of six question shapes (best X, top lists, how to choose, is it worth it, problem first, X vs Y). Runs flagged as naming a brand in the prompt are excluded. The ledger export is dated 2026-11-07 08:10 UTC. It is the same population as the URL slug study and the year-in-the-URL study.
- Engines. ChatGPT, Claude, Gemini, Google AI Mode and Perplexity. Google AI Overviews are left out because this analysis was not run on them.
- Unit. One cited URL in one answer run, counted once however many times the run repeated it. The set holds 197,384 events from 23,183 runs, 73,370 distinct addresses (query strings and fragments removed) and 29,282 root domains. An average run cites 8.5 distinct addresses.
- A folder. The first folder of the URL path, lowercased, after dropping every segment that looks like a language code: two letters (
/en/), or two letters, a hyphen and two to four letters (/de-de/). Where nothing is left, the address is a homepage. Where exactly one segment is left, it is a root-level slug. - Folder groups. Blog or article: blog, blogs, article, articles, post, posts. Resource or learning: resources, guides, guide, learn, insights, academy, library, knowledge, faq, help, support, docs, glossary, wiki, resource-center, learning, education. Product or solution: products, product, solutions, solution, services, service, pricing, platform, features. List, compare or review: best, compare, reviews, review, vs, alternatives, top, comparison, comparisons. News or press: news, press, newsroom, press-releases, media. Everything else is left ungrouped.
- No page was fetched or read. A folder is a fact about the address, not about the page. A
/products/address on a review site is a review of a product, not a vendor's sales page, and is counted in the product group.
Where cited addresses sit #
Cited-URL events by the first folder of the path (share of all 197,384; the 95 percent intervals resample the 1,008 questions, 2,000 draws):
| Where the address sits | Events | Share | 95% interval |
|---|---|---|---|
| Blog or article folder | 60,306 | 30.6% | 29.7 to 31.4 |
| Root-level slug, no folder | 35,719 | 18.1% | 17.6 to 18.6 |
| Resource or learning folder | 15,758 | 8.0% | 7.6 to 8.4 |
| Product or solution folder | 6,215 | 3.1% | 2.9 to 3.4 |
| Homepage | 4,478 | 2.3% | 2.0 to 2.5 |
| List, compare or review folder | 4,032 | 2.0% | 1.9 to 2.2 |
| News or press folder | 2,587 | 1.3% | 1.2 to 1.4 |
| Everything else | 68,289 | 34.6% |
The single biggest folder is /blog/, at 44,241 events (22.4 percent) across 7,313 domains; /blogs/ follows at 6,452 (3.3 percent), then /resources/ (5,685; 2.9), /articles/ (5,145; 2.6), /r/ (3,362; 1.7, on six domains), /products/ (2,862; 1.5), /insights/ (2,427; 1.2), /guides/ (2,396; 1.2) and /news/ (2,322; 1.2). A further 1,677 events (0.8 percent) are on a blog. subdomain and are not in the blog-folder row.
The blog folder is not one publisher's habit. The 60,306 blog-folder events come from 9,958 root domains; the ten largest hold 3,395 (5.6 percent) and the hundred largest 12,030 (19.9 percent). Of the 1,719 domains cited at least 20 times, 873 (50.8 percent) had a blog-folder address cited.
How deep the address sits is narrow. Path depth is 2 segments for 50.8 percent of cited addresses, 3 for 18.3 percent, 1 for 17.2 percent, 4 for 6.2 percent, 5 for 4.0 percent and 6 or more for 1.3 percent; 2.2 percent are the homepage. Cited addresses are on the main domain or www for 92.1 percent of events, and on a subdomain for 7.9 percent. The most common subdomain labels are blog (1,677 events), docs (683) and learn (638).
The engines differ more than the question shapes do #
Share of cited URLs by first folder, by engine (events matching, out of events; 95 percent interval resampling questions in brackets for the three rows that move most):
| Engine | Cited URLs | Blog or article | Root-level slug | Resource or learning | Product or solution | Homepage |
|---|---|---|---|---|---|---|
| Claude | 33,192 | 41.6% (40.1 to 43.2) | 19.1% | 7.9% | 1.8% (1.4 to 2.2) | 0.8% (0.6 to 1.1) |
| AI Mode | 6,124 | 34.7% (31.0 to 38.3) | 20.1% | 6.6% | 1.7% | 0.6% |
| Gemini | 17,416 | 31.1% (28.6 to 33.7) | 19.5% | 5.3% | 5.0% (3.6 to 6.5) | 1.4% |
| Perplexity | 105,295 | 30.9% (29.8 to 32.1) | 16.3% | 8.4% | 1.8% (1.6 to 2.0) | 0.8% |
| ChatGPT | 35,357 | 18.0% (17.0 to 19.1) | 21.4% | 8.5% | 7.8% (6.8 to 8.8) | 8.7% (7.5 to 10.0) |
ChatGPT is the outlier. It cites a blog or article folder less than any other engine (18.0 percent against 30.9 to 41.6) and it is the one engine that routinely cites homepages and product folders: 8.7 percent homepages, against 0.6 to 1.4 percent for the other four, and 7.8 percent product folders, against 1.7 to 5.0.
Per run the gap is the same. Share of runs that cite at least one address of the kind (23,183 runs; Gemini has 1,397 and Google AI Mode 855, so those two are rougher):
| Engine | Runs | At least one blog-folder address | At least one product-folder address | At least one homepage |
|---|---|---|---|---|
| Perplexity | 7,778 | 89.5% | 15.7% | 7.3% |
| Claude | 5,979 | 84.6% | 7.7% | 3.8% |
| Gemini | 1,397 | 84.3% | 24.3% | 13.0% |
| Google AI Mode | 855 | 68.1% | 8.4% | 2.3% |
| ChatGPT | 7,174 | 54.1% | 19.7% | 18.0% |
ChatGPT's homepage citations are broad, not a few brands: 3,087 events on 1,248 root domains, with no domain above 21 events. Almost one ChatGPT run in five (18.0 percent) cites a homepage.
The question changes the mix, but less #
Share of cited URLs by first folder, by the shape of the question that produced the citation (events matching, out of events):
| Question shape | Cited URLs | Blog or article | Resource or learning | Product or solution | Homepage | List, compare or review |
|---|---|---|---|---|---|---|
| How to choose | 31,293 | 32.6% | 9.1% | 2.3% | 2.2% | 1.6% |
| Is it worth it | 31,616 | 32.9% | 9.0% | 2.4% | 1.7% | 1.2% |
| Problem first | 33,977 | 31.8% | 9.3% | 2.3% | 1.4% | 0.8% |
| X vs Y | 30,590 | 31.2% | 8.4% | 1.9% | 1.4% | 2.6% |
| Best X | 35,628 | 28.4% (26.4 to 30.5) | 6.2% | 4.6% (3.9 to 5.4) | 3.0% | 3.0% |
| Top lists | 34,280 | 26.9% (24.9 to 28.9) | 6.3% | 5.0% (4.1 to 6.0) | 3.8% (3.1 to 4.6) | 3.0% |
Questions that ask for a ranked list (best X, top lists) pull in the most product folders and homepages and the fewest blog folders; questions that ask whether something is worth it, how to choose, or describe a problem pull in the most blog and resource folders. The range across shapes is 26.9 to 32.9 percent for blogs, a spread of six points, against 23.6 points across engines.
The vertical moves the blog share most #
Blog or article share of cited URLs, highest and lowest of the 24 verticals, with the homepage share beside it (events matching, out of events; pooled counts, no interval):
| Vertical | Cited URLs | Blog or article | Homepage |
|---|---|---|---|
| Sales, GTM and CRM | 2,539 | 51.2% | 5.2% |
| AI visibility and GEO | 12,680 | 45.0% | 4.3% |
| Martech and advertising | 4,026 | 41.9% | 1.3% |
| AI infrastructure | 10,974 | 40.3% | 1.3% |
| Fintech | 8,277 | 40.1% | 2.8% |
| Consumer products | 12,698 | 22.2% | 1.1% |
| Review-sparse industrial | 5,044 | 18.2% | 1.0% |
| Local services | 5,639 | 17.5% | 7.4% |
| Consumer finance | 12,385 | 17.1% | 1.1% |
The spread from top to bottom is three to one. Software and AI categories, where vendors publish heavily, cite blog folders most. Consumer finance, local services and industrial categories cite them least, and local services cites homepages most (7.4 percent, where a business's one page is its homepage). In AI visibility and GEO, the category this site covers, 45.0 percent of cited addresses are blog or article folders and 1.6 percent are product or solutions folders.
What this changes for where an answer lives #
- Put the answer on an article page with its own address. Blog, article and resource folders hold 38.6 percent of cited addresses together; product and solution folders hold 3.1 percent. A feature page that states what a product does is in the 3 percent lane.
- A folder is not required. 18.1 percent of cited addresses are a single slug directly under the domain. Moving a page into
/blog/is not shown to change its citations; the slug study shows the slug is what matches the question (slug and question shape). - Keep a homepage that says what you are. ChatGPT cited a homepage in 18.0 percent of its runs and no other engine does so routinely. If a buyer's engine is ChatGPT, the homepage is an address it cites, so its opening lines should name the company, what it does and who for. The ledger does not show what ChatGPT used the homepage for.
- Ranking questions are where product and home pages surface. On best-X and top-list questions, product folders (4.6 and 5.0 percent) and homepages (3.0 and 3.8 percent) are cited most: these are the questions where a vendor's own page is most likely to be among the citations. How-to-choose and problem-first questions cite blog and resource folders for 41 to 42 percent of their citations.
- Do not generalize across engines. A claim like "AI engines cite blogs" is a Claude and Perplexity claim. ChatGPT cites blog folders in fewer than one in five citations.
What this page does not show #
- Preference. The ledger does not hold how many blog, product and home pages exist for each question. A blog folder is where most publishers keep editorial content, so a 30.6 percent share may mostly reflect supply.
- Cause. Nothing here says the folder name earns a citation. 18.1 percent of cited addresses have no folder.
- Vendor intent. A
/products/address on a review or directory site counts as a product folder. The product group mixes vendor pages with reviews of products. Among its largest domains are review and directory sites as well as vendor sites. - Page content. No page was read. A blog-folder address may be a short announcement or a thousand-word guide.
- 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.
- Citation quality. A cited URL is a source an engine linked or listed. Whether the answer relied on it or named it is a separate question: the Index finds engines name only 17.6 percent of the sources they cite (AI engines cite sources they do not name).
- Google AI Overviews. Not included.
- Query strings. Addresses were read without their query strings. ChatGPT's API adds
?utm_source=openaito every address it cites and the other engines almost never add a tag (the tracking-tag study). - Folder naming. Only the first folder is read. A
/resources/blog/address is a resource-folder address. Any two-letter segment (such as/ai/or/uk/) is dropped as if it were a language code; a variant that drops only a leading language code moves the blog share to 30.4 percent and leaves the others within 0.1 points.
Method note #
Source: the Index answer ledger's citation-events table, daily export of 2026-11-07 08:10 UTC, read with DuckDB and joined to the question table for question shape and vertical. Filters: panel "mri_v2", runs not flagged as brand-prompted, a cited URL beginning with http, 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 on the cited URL as recorded (197,384). Classification: the URL is lowercased and its scheme, host, query string and fragment removed; trailing slashes are dropped; empty segments are dropped; a segment that is two letters, or two letters, a hyphen and two to four letters, is treated as a language code and ignored; the first remaining segment is the folder. No remaining segment is a homepage, one is a root-level slug. Subdomains are read from the host: an address is on the main domain if its host is the root domain or www. plus the root domain. Intervals resample the 1,008 questions, 2,000 draws, fixed seed, with every cited URL of a question kept together; engine, shape and vertical rows pool the counts. Run-level shares count a run once if any of its cited addresses is of the kind. A recount by a second expression that removes only a leading language code and tests the path with regular expressions returns 60,082 blog-folder events (30.4 percent), 6,165 product-folder (3.1), 4,476 homepage (2.3) and 35,582 root-level slugs (18.0) on the same 197,384 events.
Frequently asked questions #
What kinds of pages do AI engines cite? #
Mostly articles. In the Machine Relations Index, 30.6 percent of cited URLs (60,306 of 197,384) sat in a blog or article folder, 18.1 percent were a single slug under the domain, 8.0 percent were in a resource or learning folder, 3.1 percent were in a product or solutions folder and 2.3 percent were homepages. This describes the cited addresses and does not show that a folder earns a citation.
Do AI engines cite product pages? #
Rarely. Product, solution and service folders were 3.1 percent of cited URLs, and 15.1 percent of runs cited at least one. ChatGPT cited them most (7.8 percent of its citations), Claude, Perplexity and Google AI Mode least (about 1.7 to 1.8 percent).
Do AI engines cite homepages? #
ChatGPT does; the others almost never. A homepage was 8.7 percent of ChatGPT's cited URLs and 18.0 percent of its runs cited one, against 0.6 to 1.4 percent of citations on the other four engines.
Does a page have to be in a /blog/ folder to be cited? #
No. 18.1 percent of cited addresses are a single slug directly under the domain, with no folder. The ledger does not show that a folder changes the odds of a citation.