The address of a page that an AI engine cites is usually a five-word slug. In the Machine Relations Index answer ledger, 197,384 cited-URL events came from 23,183 answer runs on 1,008 buying questions. The median slug, the last part of the address, had 5 words and 31 characters. 58.3 percent of cited addresses had a slug of 3 to 7 words (115,169 events). 16.5 percent had 1 or 2 words, 20.7 percent had 8 or more, and 4.4 percent had no words at all: a bare homepage (2.2 percent) or a slug made only of numbers (2.2 percent).
What the buyer asked changes the length. On problem-first questions, where the buyer describes a situation and names no product, 28.9 percent of cited slugs had 8 or more words. On best-X and top-list questions the figure was 15.4 and 14.6 percent. The intervals do not overlap.
This page counts what cited addresses look like. It does not say that a slug of any length earns a citation: the Index holds no count of how long slugs are on the web in general, so a typical cited slug may simply be a typical slug.
What was counted #
- Population. Successful runs in the Index's "mri_v2" panel, 2026-07-06 to 2026-11-07, on its 1,008 buying questions (168 in each of six question shapes). Runs flagged as naming a brand in the prompt are excluded. The ledger export is dated 2026-11-08 08:10 UTC. The events and runs are the same 197,384 and 23,183 as the URL slug study, the year-in-the-URL study, the site-sections study and the tracking-tag study.
- Engines. ChatGPT, Claude, Gemini, Google AI Mode and Perplexity, called through their APIs. 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 same address cited in two runs counts twice. The set holds 78,194 distinct addresses.
- Slug length. The last non-empty segment of the path, taken before any query string or fragment, with a trailing
.html,.htm,.php,.asp,.aspx,.shtmlor.jspremoved. Its length is the number of non-empty pieces after splitting on hyphens and underscores. Numbers count as pieces, so2026is one word and a leading article ID adds one. - Two kinds of address without a word count. A homepage has an empty path. An ID-only slug has no letter in it (for example
/p/12345). - No page was fetched to build the counts. Length is a fact about the address the engine returned.
The typical cited slug is 5 words and 31 characters #
Distribution of slug length across the 188,646 cited addresses that have a word slug (events; the homepage and ID-only cases are counted separately in the next table):
| Percentile | Words | Characters |
|---|---|---|
| 10th | 2 | 12 |
| 25th | 3 | 20 |
| Median | 5 | 31 |
| 75th | 7 | 45 |
| 90th | 10 | 62 |
The mean is 5.3 words. Counting each of the 78,194 distinct addresses once gives the same median (5) and a mean of 5.5, so repeated citations of a few addresses do not set the result.
Share of all 197,384 cited addresses, by length of slug (95 percent intervals resample the 1,008 questions, with every cited address of a question kept together; 2,000 draws, fixed seed):
| Slug | Events | Share | 95% interval |
|---|---|---|---|
| Homepage (no path) | 4,317 | 2.2% | 1.9 to 2.5 |
| ID only (no letters) | 4,421 | 2.2% | 2.0 to 2.5 |
| 1 to 2 words | 32,537 | 16.5% | 15.8 to 17.1 |
| 3 to 7 words | 115,169 | 58.3% | 57.4 to 59.2 |
| 8 or more words | 40,940 | 20.7% | 20.0 to 21.5 |
Within the band, 3 to 4 words hold 60,388 events (30.6 percent) and 5 to 7 words hold 54,781 (27.8 percent). Words 8 to 10 hold 25,129 (12.7 percent) and 11 or more hold 15,811 (8.0 percent).
The 1-to-2-word band is the most concentrated. youtube.com/watch alone is 2,921 of its 32,537 events, and YouTube links in all are 2,978 (9.2 percent). Without YouTube the band is 29,559 events (15.0 percent). In the 8-or-more band, reddit.com is the largest domain with 1,724 of 40,940 (4.2 percent): Reddit's slugs are post titles joined by underscores. Both bands are spread across many publishers, 6,170 root domains in the short band and 8,787 in the long one.
How deep the address sits #
Path depth, counted as the number of path segments (the slug is the last one):
| Depth | Example | Events | Share |
|---|---|---|---|
| 0 | example.com |
4,317 | 2.2% |
| 1 | example.com/pricing |
34,011 | 17.2% |
| 2 | example.com/blog/post-title |
100,287 | 50.8% |
| 3 | example.com/blog/2026/post-title |
36,060 | 18.3% |
| 4 | 12,258 | 6.2% | |
| 5 | 7,848 | 4.0% | |
| 6 or more | 2,603 | 1.3% |
Half of all cited addresses are exactly two segments: one folder and a slug. 88.5 percent (174,675 events) sit at three segments or fewer. The folder itself is counted in the site-sections study.
The question changes the length #
Share of cited addresses in each length band, by the shape of the question that produced the citation (events matching, out of events; intervals resample the 168 questions of a shape):
| Question shape | Cited URLs | 1 to 2 words | 3 to 7 words | 8 or more words |
|---|---|---|---|---|
| Problem first | 26,258 | 15.1% (13.9 to 16.5) | 52.5% (50.7 to 54.3) | 28.9% (26.9 to 30.7) |
| Is it worth it | 30,652 | 13.8% (12.4 to 15.4) | 58.7% (56.9 to 60.7) | 24.3% (22.6 to 26.0) |
| X vs Y | 30,246 | 14.5% (13.0 to 16.1) | 59.6% (57.4 to 61.8) | 22.9% (21.2 to 24.7) |
| How to choose | 31,480 | 15.2% (13.6 to 16.9) | 58.2% (56.3 to 60.0) | 22.8% (21.1 to 24.3) |
| Best X | 38,847 | 17.7% (16.2 to 19.2) | 61.1% (58.8 to 63.4) | 15.4% (14.1 to 16.7) |
| Top lists | 39,901 | 20.7% (19.1 to 22.3) | 58.3% (56.1 to 60.7) | 14.6% (13.3 to 15.9) |
Shares in a row do not add to 100 because homepage and ID-only addresses are left out of the table; they are 0.4 to 3.8 percent and 1.7 to 2.6 percent of a shape.
The shapes fall into three groups. Ranking questions (best X and top lists) draw the shortest slugs: one in five top-list citations has 1 or 2 words, and only 14.6 to 15.4 percent have 8 or more. The three evaluation questions (is it worth it, X vs Y, how to choose) sit near 23 to 24 percent long. Problem-first questions draw the longest slugs, 28.9 percent at 8 or more words, and the fewest in the middle band. Problem-first is the highest cell in its column and its interval does not overlap the top-list or best-X intervals.
The ordering holds when each distinct address is counted once per question shape: 8 or more words is 29.0 percent on problem-first, 25.6 on is-it-worth-it, 24.9 on X vs Y, 23.6 on how to choose, 17.8 on best X and 16.6 on top lists.
One reading is that a ranking page is titled around a category and a problem page around a sentence describing the situation. The ledger does not read the pages, so it cannot confirm that.
Engines differ in how short they go #
Share of each engine's cited addresses in each length band (events matching, out of events; intervals resample the questions the engine cited on):
| Engine | Cited URLs | Homepage | 1 to 2 words | 3 to 7 words | 8 or more words |
|---|---|---|---|---|---|
| ChatGPT | 35,357 | 8.5% (7.3 to 9.7) | 21.7% (20.5 to 22.9) | 48.8% (47.2 to 50.3) | 18.2% (17.1 to 19.4) |
| Gemini | 17,416 | 1.4% (1.0 to 1.9) | 19.0% (17.6 to 20.3) | 53.3% (51.5 to 54.9) | 23.7% (22.4 to 25.0) |
| Google AI Mode | 6,124 | 0.6% (0.3 to 0.9) | 17.7% (15.4 to 19.9) | 50.9% (47.4 to 54.5) | 28.6% (25.5 to 31.6) |
| Perplexity | 105,295 | 0.7% (0.6 to 0.9) | 16.6% (15.7 to 17.4) | 61.1% (60.1 to 62.1) | 19.8% (19.0 to 20.6) |
| Claude | 33,192 | 0.8% (0.6 to 1.1) | 9.1% (8.4 to 9.8) | 63.9% (62.5 to 65.2) | 23.5% (22.5 to 24.6) |
ChatGPT cites the shortest addresses. 30.2 percent of its citations are a homepage or a slug of 1 or 2 words, against 9.9 percent for Claude. Claude puts 63.9 percent of its citations in the 3-to-7-word band. Google AI Mode is highest on long slugs (28.6 percent) but answers only 138 of the 1,008 questions, so its interval is wide and its question mix is not the same as the others'.
The 1-to-2-word band includes YouTube watch links: 626 of ChatGPT's 7,671 events in that band and 1,350 of Perplexity's 17,464. Without YouTube, ChatGPT's band is 19.9 percent and Perplexity's is 15.3.
What this changes #
- Write the slug as the question, in about five words. That is where 58 percent of cited addresses sit. Inside the band, 3 to 4 words and 5 to 7 words are almost equal (30.6 and 27.8 percent), so there is no single best length to hit.
- Do not cut a slug down to fit a rule. One cited address in five has 8 or more words, and one in twelve has 11 or more. Problem-first pages in particular cite long slugs. A slug that states the situation is within the cited range.
- Short slugs belong on category and hub pages. Top-list questions draw the most 1-to-2-word slugs (20.7 percent), and ChatGPT cites them and homepages more than any other engine. A page whose job is to be the single page for a category fits that shape.
- Keep important pages within two or three folders. 88.5 percent of cited addresses are three segments or fewer.
- This site's own research slugs run long. On 2026-11-08 the 91 addresses in the research sitemap of machinerelations.ai had a median slug of 7 words, 42.9 percent had 8 or more and none had 1 or 2, against 20.7 percent at 8 or more in what engines cited. The data does not show that this has a cost. It is a gap between how this site names its pages and how cited pages are named, and it is not a reason to rename pages: a rename breaks inbound links unless it is redirected, and the ledger shows what engines cited, not what a rename would earn.
What this page does not show #
- Cause. A slug of a given length may label a page that was cited for its content. No page was read, so this does not separate the two.
- A web baseline. The Index does not hold the slug lengths of pages that were not cited. If most pages on the web have slugs of 3 to 7 words, the cited band would be the web's band. Nothing here says a five-word slug is cited more often than a ten-word one.
- 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).
- Non-Latin slugs. A slug with no Latin letter counts as ID-only, so addresses in other scripts are in that 2.2 percent.
- Google AI Overviews. Not included.
- Trend. The page does not test whether these shares change over time.
Method note #
Source: the Index answer ledger's citation-events table, daily export of 2026-11-08 08:10 UTC, read with DuckDB, joined to the run table (successful runs only) and the question table (panel "mri_v2", for the question shape). Filters: panel "mri_v2", 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. The path is the address with the scheme and host, the query string and fragment, and trailing slashes removed. The slug is the last path segment with a trailing page extension removed. A homepage has an empty path. A slug with no letter A to Z is ID-only. Otherwise word count is the number of non-empty pieces after splitting on hyphens and underscores. Depth is the number of non-empty path segments. Bands: 1 to 2, 3 to 7, 8 or more. Percentiles are over the 188,646 addresses with a word slug. Intervals resample the questions in a group with replacement, keeping each question's cited addresses together, 2,000 draws, fixed seed; shares are pooled event counts, not averages of per-question shares. The distinct-address check counts each cited address once per question shape. The site's own slug count uses the 91 addresses in https://machinerelations.ai/research/sitemap.xml on 2026-11-08, with the same word rule.
Frequently asked questions #
How long should a URL slug be? #
For the addresses AI engines cited in the Machine Relations Index, the median slug was 5 words and 31 characters, and 58.3 percent had 3 to 7 words (115,169 of 197,384). That describes cited addresses. The ledger does not show that a slug of that length earns a citation.
Do AI engines prefer short URLs? #
Not in this ledger. 16.5 percent of cited addresses had a slug of 1 or 2 words and 20.7 percent had 8 or more. ChatGPT cited the most short addresses (30.2 percent were a homepage or a 1-or-2-word slug) and Claude the fewest (9.9 percent).
Do long slugs get cited? #
Yes. 40,940 of 197,384 cited addresses (20.7 percent) had 8 or more words, and on problem-first questions the share was 28.9 percent (26.9 to 30.7).
Should I change my existing URLs to match this? #
Not on this evidence. The data shows what engines cited, not what a rename would earn, and a rename breaks inbound links unless it is redirected. For new pages, name the page for the question it answers in about five words.
How many folders deep are the pages AI engines cite? #
Half (50.8 percent) are two segments, a folder and a slug. 88.5 percent are three segments or fewer. Only 1.3 percent are six or more.
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