Schema markup does not separate a cited page from its neighbours, because almost every page on the websites AI engines cite already has it. In the Machine Relations Index answer ledger, 597 of 687 cited pages (86.9 percent, 95 percent interval 84.2 to 89.2) carried JSON-LD. So did 583 of 687 pages (84.9 percent, 82.0 to 87.3) picked at random from the sitemaps of the same websites, pages the ledger never recorded as cited. The paired difference is 2.0 percentage points (0.6 to 3.6).
Markup is not required either. 80 of the 687 cited pages (11.6 percent) carried neither JSON-LD nor microdata. The one type with a visible gap is FAQPage: 29.0 percent of cited pages against 23.1 percent of their same-site neighbours, 5.8 points (3.2 to 8.9). That gap is a count of co-occurrence, not a finding that adding FAQ markup earns a citation.
What was counted #
- Population. Cited addresses in the Index's buying-question panel, successful runs from 2026-07-06 to 2026-11-07, five engines (ChatGPT, Claude, Gemini, Google AI Mode, Perplexity) through their APIs, runs that name a brand in the prompt excluded. This is not the consumer ChatGPT or Claude app. Google AI Overviews are left out. The ledger holds 78,194 distinct cited addresses on 29,282 root domains, the same population as the subdomain study and the URL slug study.
- Sample. 1,000 root domains drawn at random (fixed seed) from the cited domains, social platforms removed (Reddit, YouTube, LinkedIn, Facebook, Instagram, X, TikTok, Quora, Medium), and one cited address drawn at random from each. 991 of those addresses fall in the window above. Each domain counts once, so the sample describes websites that get cited, not citations weighted by how often a big publisher is cited.
- Cited pages fetched. On 2026-11-09 each address was fetched by plain HTTP (15 second limit, browser user agent). 771 of 991 (77.8 percent) returned readable HTML. The other 220 (22.2 percent) timed out, were refused, or were not HTML, and are unmeasured, not zero.
- Same-site neighbours. For each readable cited page, one page was drawn at random from its website's sitemap, excluding any address the ledger records as cited, and from the same top-level folder when the sitemap held at least five addresses there (425 of the 687 pairs; the rest come from anywhere in the sitemap). 687 of the 771 cited pages ended up with a readable neighbour. All comparisons below are on those 687 pairs.
- What counts as schema. A JSON-LD script block on the page as served, with its types read from every
@typein it, including nested and@graphentries; microdata means anitemscopewith anitemtype. Markup added by JavaScript after load is not seen. The engines' own fetchers differ in what they render, and the ledger does not record which of them read which page.
The paired result #
Share of pages carrying each markup, 687 cited pages against their 687 same-site neighbours (difference intervals resample pairs, 2,000 draws, fixed seed):
| Markup | Cited pages | Same-site neighbours | Difference, points |
|---|---|---|---|
| Any JSON-LD | 86.9% (597) | 84.9% (583) | +2.0 (0.6 to 3.6) |
| JSON-LD or microdata | 88.4% (607) | 85.9% (590) | +2.5 (1.0 to 4.1) |
| Organization | 75.4% | 73.8% | +1.6 (-0.3 to 3.5) |
| BreadcrumbList | 56.9% | 56.8% | +0.1 (-2.0 to 2.3) |
| Article, BlogPosting, NewsArticle or TechArticle | 61.6% | 58.7% | +2.9 (0.1 to 5.8) |
| Person | 50.7% | 47.2% | +3.5 (0.9 to 6.1) |
| FAQPage | 29.0% (199) | 23.1% (159) | +5.8 (3.2 to 8.9) |
| Question (inside FAQ markup) | 29.4% | 23.4% | +6.0 (3.3 to 9.0) |
| Product | 4.4% | 2.8% | +1.6 (0.6 to 2.8) |
| HowTo | 1.5% | 1.2% | +0.3 (-0.6 to 1.2) |
Of the 30 pairs where exactly one page carried JSON-LD, 22 were the cited page and 8 the neighbour. In 575 pairs both had it and in 82 neither did. Restricted to the 425 same-folder pairs, the result is the same: 88.0 percent of cited pages against 85.9 percent of neighbours. Across all 771 readable cited pages, those cited on more questions were not more likely to carry it: 84.0 percent of the 631 cited on a single question, 84.9 percent of the 139 cited on two to four.
The gap in any markup is real but small, 2 points on a base of 85. Most of the types that SEO checklists name are within a few points of what the same website's ordinary pages carry.
Three tests of the same question, three designs #
Ahrefs reported two numbers that look contradictory until the design is read. Across 6 million URLs, AI-cited pages were almost three times as likely to carry JSON-LD as pages that were not cited. When it then tracked 1,885 pages that added JSON-LD between August 2025 and March 2026 against 4,000 matched controls, adding schema produced no major uplift on any platform: -4.6 percent in Google AI Overviews, +2.4 percent in AI Mode and +2.2 percent in ChatGPT, the last two statistically indistinguishable from zero. An October 2026 arXiv preprint, analysing 158,099 answers and 468,037 citations, compares pages of the same website and finds that pages whose title and paragraphs match the questions people ask are cited more, while FAQ blocks and structured data appear effective only across websites.
This page adds the same-website comparison on the Index's own ledger, and it lands where the preprint does. Ahrefs' own reading of the three-fold gap is that markup tends to live on better-maintained, more technically sophisticated sites that also publish stronger content. Here, inside sites that are all cited, the cited page and the page the ledger never recorded as cited look alike in markup, which is what a gap between websites predicts. The ledger cannot say which came first, and it was not built to test whether adding markup changes a page's chance of being cited. Ahrefs' before-and-after test is the one that tries to, and it found no clear lift.
FAQPage: the gap to read carefully #
FAQPage and its Question entries are the only types with a gap whose interval stays clear of zero by more than a point or two, and the cited page was the one carrying it in 68 pairs against 28 for the neighbour. Two things keep that from being advice.
- Markup travels with content. A page with FAQPage markup usually has a visible question-and-answer section. The ledger counts the markup; it cannot separate it from the questions and answers it describes. The preprint finds that content matching the questions people ask works inside a website, and that FAQ blocks appear effective only across websites.
- Size. 71 percent of cited pages carry no FAQPage markup at all.
What this changes #
- Do not sell markup as a citation lever. On 85 percent of ordinary pages from cited websites it is already present, so having it earns no distinction.
- Do not strip it. Nothing here shows markup hurts. Ahrefs' small AI Overviews decline came with treated and control pages both falling, and Ahrefs says its data cannot tell whether schema caused it.
- Do not wait for it. 11.6 percent of cited pages carried no markup of either kind, so a page can be cited without it.
- Spend the effort on what engines retrieve. The preprint finds that which pages ChatGPT's own search retrieves explains far more of the variation between pages (32 percent) than page properties (2.8 percent). What a page says and whether it matches the questions an engine searches for are the levers the markup studies leave standing. The citation-factors page places this beside the other studies on this ledger.
What this does not show #
- Cause. Nothing here shows markup does or does not change whether a page is cited.
- JavaScript-injected markup. Pages whose schema appears only after rendering count as having none. If that were common among cited pages, the cited share would be understated.
- The 22.2 percent not read. Sites that refuse plain requests may differ from the ones that answer. Cited pages and neighbours were fetched the same way, so the comparison holds for pages that could be read, not for all pages.
- Neighbours are not guaranteed uncited. They are addresses the ledger did not record, drawn from a panel of 1,008 questions. A neighbour might be cited on a question the panel never asked.
- Sitemap pages differ from cited pages in other ways. Age, depth and traffic were not matched. The same-folder subset narrows the difference but does not remove it.
- All websites. Every site in the sample was cited at least once. This page makes no claim about websites the engines never cite, where the gap Ahrefs found across 6 million URLs may sit.
Method note #
Source: the Index answer ledger's citation-events table, daily export of 2026-11-08 08:10 UTC, read with DuckDB. Filters: the buying-question panel, brand-prompted runs excluded, engines ChatGPT, Claude, Gemini, Google AI Mode and Perplexity, run dates 2026-07-06 to 2026-11-07. One cited address per sampled root domain, drawn with a fixed seed (20261109) from domains outside the social list above. Fetch: plain HTTP GET on 2026-11-09 between 02:37 and 02:40 UTC, 15 second limit, 2.5 MB read limit, responses counted as readable when status 200, a content type containing html and a body over 2,000 characters. Same-site neighbour: robots.txt sitemap entries (or /sitemap.xml), up to eight sitemap files per site with blog, post, article, page, news and resource files first, addresses on the same host family, image, PDF and XML files removed, any address in the ledger's cited set removed, shuffled with the cited address as seed, first readable of three tried. Types: every string @type found by walking the parsed JSON-LD, with a pattern fallback for blocks that did not parse. Intervals on shares are Wilson 95 percent; intervals on differences resample pairs with replacement, 2,000 draws, fixed seed. The cited-address sample can be re-drawn from the ledger export with the seed above. The fetches and sitemap draws read live pages on 2026-11-09, so a later re-run reads pages that may have changed.
Frequently asked questions #
Does schema markup help AI citations? #
Not in a way this ledger can see. 86.9 percent of cited pages and 84.9 percent of same-site pages the ledger never saw cited carried JSON-LD, a 2.0 point gap. Ahrefs' test of 1,885 pages that added schema found no major citation uplift on Google AI Overviews, AI Mode or ChatGPT.
Do cited pages have more schema markup than other pages? #
Than pages on other websites, yes: Ahrefs found AI-cited pages almost three times as likely to carry JSON-LD across 6 million URLs. Than other pages on the same website, barely: 86.9 percent against 84.9 percent here.
Which schema type shows the biggest difference? #
FAQPage, on 29.0 percent of cited pages against 23.1 percent of same-site neighbours. FAQ markup usually comes with visible questions and answers, so the gap may reflect that content and not the markup.
Can a page be cited without schema markup? #
Yes. 80 of the 687 cited pages (11.6 percent) had neither JSON-LD nor microdata in the HTML as served.
Should I remove schema markup? #
Nothing here or in Ahrefs' test supports removing it. Markup is present on most ordinary pages of cited sites, and what is missing is evidence that it earns citations, not evidence that it costs them.