A named author, a visible date and outbound links all look like citation signals until the two pages being compared are the same length. In the Machine Relations Index answer ledger, 659 pages that AI engines cited were each paired with another page from the same website that the ledger never recorded as cited. 60.1 percent of the cited pages named an author against 51.7 percent of the neighbours (+8.3 points, 5.5 to 11.4). Where the two pages were within one and a half times each other's length (294 pairs), the gap was +1.4 points (-1.7 to +4.4). Where the cited page was more than one and a half times longer (303 pairs), it was +16.2.
The gap follows length and page type, not the byline. This is a count of what cited pages look like, not a test that adding an author earns a citation.
What was counted #
- Pairs. The 707 pairs of the opening study: one cited address drawn at random from each of 1,000 random root domains in the Index's buying-question panel (runs 2026-07-06 to 2026-11-18, five engines through their APIs, brand-prompted runs and social platforms excluded), and one same-site page from the website's sitemap that the ledger never recorded as cited, from the same top-level folder where the sitemap held at least five addresses there. The same two addresses were fetched again on 2026-11-19 between 11:40 and 11:46 UTC by plain HTTP (12 second limit, browser user agent).
- Readable. Both pages returned readable HTML with at least 30 words of main content in 659 of the 707 pairs. The other 48 (24 cited pages, 26 neighbours that no longer loaded or no longer had text) are unmeasured, not zero.
- Features. Whether each page carried a named author, an Article-type JSON-LD block, a visible or machine-readable date, outbound links to other sites, three or more images and a table. Detection rules are in the method note. Content added by JavaScript after load is not seen.
The paired result #
Share of pages with each feature, 659 cited pages against their 659 same-site neighbours. Difference intervals resample pairs, 2,000 draws, fixed seed.
| Feature | Cited pages | Same-site neighbours | Difference, points |
|---|---|---|---|
| A named author (any rule below) | 60.1% (396) | 51.7% (341) | +8.3 (5.5 to 11.4) |
| Author in JSON-LD as a person | 33.4% (220) | 27.5% (181) | +5.9 (3.2 to 8.5) |
| Author in a meta tag | 36.9% (243) | 32.0% (211) | +4.9 (2.6 to 7.1) |
| Visible "By First Last" text in the first 150 words | 7.7% (51) | 8.5% (56) | -0.8 (-2.6 to +1.2) |
| Article-type JSON-LD | 66.9% (441) | 54.8% (361) | +12.1 (8.8 to 15.5) |
| Any publish or update date | 74.4% (490) | 65.6% (432) | +8.8 (5.5 to 12.1) |
| Outbound link to another website | 66.0% (435) | 55.7% (367) | +10.3 (6.2 to 14.3) |
| Three or more images | 59.9% (395) | 58.3% (384) | +1.7 (-1.5 to +5.0) |
| A table | 40.7% (268) | 26.7% (176) | +14.0 (10.0 to 18.1) |
The median cited page ran 1,745 words against 1,085 for its neighbour, a paired median difference of +354 words (261 to 484): longer in 441 pairs, shorter in 210.
The gap tracks how much longer the cited page is #
Pairs sorted by the ratio of the two pages' word counts. Each cell is the difference in points between cited page and neighbour, with its interval.
| Feature | Neighbour 1.5x+ longer (62 pairs) | Within 1.5x (294) | Within 1.25x (190) | Cited 1.5x+ longer (303) | Cited 3x+ longer (167) |
|---|---|---|---|---|---|
| A named author | +3.2 (-6.5 to +12.9) | +1.4 (-1.7 to +4.4) | +1.1 (-2.6 to +5.3) | +16.2 (10.9 to 21.5) | +24.0 (15.6 to 32.3) |
| Any date | -4.8 (-17.7 to +8.1) | +1.7 (-1.7 to +5.1) | +3.7 (-0.5 to +7.9) | +18.5 (12.9 to 24.1) | +25.7 (18.0 to 34.7) |
| Outbound link | -8.1 (-22.6 to +6.5) | +3.1 (-1.0 to +7.1) | +5.3 (0.5 to 10.5) | +21.1 (14.2 to 27.7) | +34.7 (25.1 to 44.3) |
| A table | -6.5 (-21.0 to +8.3) | +4.8 (0.0 to 9.9) | +2.6 (-3.2 to +8.4) | +27.1 (20.8 to 33.3) | +37.7 (29.3 to 46.1) |
| Article-type JSON-LD | +3.2 (-8.1 to +14.5) | +0.7 (-2.4 to +3.4) | +2.6 (-1.1 to +6.3) | +25.1 (19.1 to 31.0) | +37.1 (28.1 to 45.5) |
Where the cited page was far longer, its neighbour was often not an article: 42.6 percent of those neighbours carried Article-type JSON-LD against 67.7 percent of the cited pages. The author, date and link gaps in the whole sample come largely from pairing a cited article with a shorter tag, category or utility page. Within 1.25x, the author gap is +1.1 points; the outbound-link gap is the one that stays marginally above zero (+5.3, 0.5 to 10.5), and a table is borderline in the within-1.5x band (+4.8, 0.0 to 9.9). The 62 pairs where the neighbour was longer are too few to read.
Two other cuts leave the whole-sample gap smaller but above zero: among the 524 pairs where both pages ran 300 words or more it was +5.7 points (2.7 to 8.8), and among the 350 pairs from the same top-level folder +4.0 (0.9 to 6.9). Both still mix pages of different lengths.
How this fits the other studies #
The length study found cited pages longer and more often carrying tables, and the opening study found an FAQ gap that held in three cuts. This study shows why that kind of gap needs a length control: tables and dates travel with length and page type. Neither earlier study matched the two pages on length, so their FAQ and table gaps should be read with that caution. The schema study found JSON-LD was equally common on both sides; an Article-type block, which is the page-type marker here, was not.
None of this says authorship is irrelevant to people or to an engine's trust judgement. It says that, among pages from one website, whether a page names an author does not by itself tell cited pages from uncited ones once their length is similar.
What this changes #
- Do not add a byline to get a page cited. Cited and uncited pages of similar length named an author about equally often (58.8 percent against 57.5 within 1.5x).
- Treat a named author, a date and an Article block as table stakes. Most pages on both sides carry them: 74.8 percent of cited pages and 73.1 percent of equal-length neighbours had a date.
- Write the page that answers the question. What separates cited pages is being the substantial article on a site, not a marker added to a thin page.
- Do not read this as a recipe. Cited pages are a selected set, and the ledger cannot say why an engine picked one.
Limits #
- Author detection uses simple rules. A meta author tag can hold an organisation name, a JSON-LD author can be a person who is not a recognised expert, and this study did not test whether the author is real, independent or credentialed.
- 48 of 707 pairs could not be re-read and are unmeasured. A neighbour is a page the ledger did not record as cited in this panel, not a page no engine has cited, and may be a tag, category or utility page.
- The length bands are retrospective cuts of one sample, not a matched experiment; the 62 pairs with a longer neighbour and the 190 within 1.25x have wide intervals.
- Each of the 1,000 domains contributes one pair, so large publishers weigh the same as small sites. The fetches read live pages on 2026-11-19, later than the runs that cited them.
Method note #
Pairs: the 707 same-site pairs of the opening study (Index answer ledger citation-events table, daily export of 2026-11-19 08:11 UTC, the buying-question panel, runs 2026-07-06 to 2026-11-18, engines ChatGPT, Claude, Gemini, Google AI Mode and Perplexity, brand-prompted runs and social hosts removed, 1,000 root domains drawn with fixed seed 1119). Each cited and neighbour address was fetched again by plain HTTP GET between 11:40 and 11:46 UTC on 2026-11-19; a page counted as readable when status 200, a content type containing html and at least 30 words of main content (main, else article, else body, with navigation, aside, form, script, style, noscript, svg and footer elements removed). Named author: any of a meta tag named author, article:author, parsely-author or dc.creator (or a property article:author or og:author); a JSON-LD author object of type Person with a name, or an author given as text; a link with rel author; an element with itemprop author; or the words by, written by, authored by, posted by, reviewed by or author followed by two or three capitalised words in the first 150 words of main content. JSON-LD person: the JSON-LD part alone. Article type: JSON-LD of type Article, BlogPosting, NewsArticle, TechArticle, ScholarlyArticle or Report. Any date: JSON-LD datePublished, a time element, or the words updated, published or posted followed by a date in the first 3,000 characters. Outbound link: a link in main content to a different registrable host. Images: three or more img elements in main content. Table: a table element in main content. Length bands: cited-page words divided by neighbour words. Intervals on shares resample pairs with replacement, 2,000 draws; intervals on medians, 500 draws; fixed seed 7. The script and the per-pair file are retained in the Machine Relations Research repository with this analysis. A rerun reads pages that may have changed.
Frequently asked questions #
Do AI engines cite pages that name an author? #
More often than other pages on the same website, but the gap is mostly length. 60.1 percent of cited pages named an author against 51.7 percent of same-site neighbours (+8.3 points, 5.5 to 11.4) across 659 pairs. Where the two pages were within one and a half times each other's length it was +1.4 (-1.7 to +4.4).
Does adding a byline help a page get cited by AI? #
This study does not show it. Among pairs of similar length, 58.8 percent of cited pages and 57.5 percent of neighbours named an author. It counts what cited pages look like and does not test adding a byline.
Do cited pages show a publish date more often? #
Yes in the whole sample (74.4 percent against 65.6), but not among pages of similar length (74.8 percent against 73.1 within 1.5x, a gap of +1.7 with an interval from -1.7 to +5.1).
Why do cited pages look more authored, dated and linked? #
Because they are more often substantial articles. Where the cited page was more than one and a half times longer, only 42.6 percent of the neighbours carried Article-type JSON-LD against 67.7 percent of the cited pages, so many neighbours were thin or utility pages.
Do outbound links help a page get cited? #
The data cannot say. The outbound-link gap was +10.3 points overall and +3.1 (-1.0 to +7.1) within 1.5x, and the only one of the authorship, date and link features whose interval stays above zero within 1.25x (+5.3, 0.5 to 10.5). A table is borderline within 1.5x (+4.8, 0.0 to 9.9).