Skip to content

See where you stand

Change it

Works withWordPressShopifyWebflowGitHubGoogleSlackAll integrations
FixesFixes now ship to WordPress, Shopify, Webflow and GitHub

By team

By goal

Done for you

Not sure where to start? Take a guided tour.Book a demo
Customer storyHow the voice AI company Futuro grew its traffic3×AI traffic in 60 days

Learn

Data and research

FreeAI Site GraderAI Query Scannerllms.txt GeneratorAll free tools
Open dataWhere AI sends people, across more than 6,000 sites

The anatomy of an AI citation.

We crawled 1,465 pages that AI answers cite, across 950 domains, and measured what they have in common. Most of it is plain good publishing. A few traits stand out.

Mack GrenfellFounder, Trakkr. Ran the crawl and the analysis.
Published
9 April 2026
Updated
16 April 2026
Sample
1,465 cited pages, 950 domains
Window
10 January to 27 March 2026
Reading
7 minutes
Start free trialGet a free snapshot
lope.example/guides/choosing-a-daily-trainer
Guides · Daily running

How to choose a daily running shoe

PSBy Priya Shah, running coachUpdated 2 March 2026

A daily trainer does most of your miles. Here is what to look for, in the order it matters, from fit to foam to how long it lasts.

Which shoe for which run

ShoeFeelBest for
Lope DailyBalancedMost of your miles
Lope TempoFirm, lightFaster days
Lope CushionSoft, protectiveLong, easy runs

Questions runners ask

How long does a daily trainer last?
Can I race in a daily trainer?
Do I need a wide fit?
  1. 91.4%set a canonical tag. 89.2% add Open Graph tags.
  2. 67.8%carry schema. The web: 38.5%.
  3. 9.4×the web’s rate of Person schema
  4. 2,290 wordson average. The web: 750.
  5. 40%of the most-cited use tables
  6. +45% citationswith FAQ schema and content (n=23)
Figure 1A cited page, labelled. The page is an illustration; every rate comes from the 1,465 pages in the study. Correlation, not causation.Jump to the findings

In short

Four findings, strongest first
  1. +45%

    more citations for pages with FAQ schema and FAQ content than for pages with neither. The sample is small: 23 pages.

    Figure 2
  2. 67.8%

    of cited pages carry schema, against 38.5% of the web. Person schema, which names the author, turns up 9.4 times as often.

    Figure 3
  3. 1 to 20

    schema fields is the sweet spot. Light markup averages 30.5 citations; pages with more than 100 fields average 23.7.

    Figure 4
  4. 3×

    the length of a typical web page. Cited pages average 2,290 words, and 78% run past 1,000.

    Figure 5

Every AI answer that cites a source is a small vote for one page over all the others that could have answered. We wanted to know what the winning pages look like.

So we took the pages that ChatGPT, Perplexity and Gemini cited across 28,033 citation appearances and crawled them. 1,465 pages from 950 domains made it into the analysis. For each one we pulled its schema, its structure and its metadata, then compared it with the web as a whole.

Most of what we found is plain good publishing: canonical tags, real authors, long and well-organised pages. A few traits stand out from that baseline, and one is a surprise.

FAQ schema is the strongest signal.

Pages that pair FAQ schema with a visible FAQ section average 36.9 citation appearances. Pages with no FAQ signal average 25.4. That 45% gap is the largest single association in the data.

Pages with FAQ content but no markup sit in between, at 27.2, so the markup seems to add something beyond the questions themselves. But only 23 pages carry both, and they run long: 4,290 words on average. Length may explain part of the lift.

Figure 2

Pages with FAQ schema and content averaged 11.5 more citations in this sample

Each square is one citation appearance on an average page in the group. Grey squares sit under the no-FAQ baseline.

FAQ schema and content23 pages · 4,290 words
36.9+45%
FAQ content only161 pages · 2,530 words
27.2+7%
No FAQ signal269 pages · 1,980 words
25.4
Average citation appearances by FAQ signal
GroupPagesAverage wordsAverage citations
FAQ schema and content23429036.9
FAQ content only161253027.2
No FAQ signal269198025.4
Source: Trakkr crawl of AI-cited pages, March 2026. 453 pages with an FAQ reading.Data

Cited pages say who wrote them.

Two in three cited pages carry schema markup, against 38.5% of the web. Among the most-cited tenth, it rises to 80%.

The types that over-index most describe a person and a publication. Person schema, which names an author, sits on 18.9% of cited pages and 2% of the web: 9.4 times as often. ImageObject and NewsArticle follow.

Over-indexing is not the same as lift. Within cited pages, most types don’t predict more citations. They describe the kind of page AI tends to cite: edited, attributed and kept up to date. FAQPage is the one type tied to a higher count.

Figure 3: Pages with Person schema, out of every 100

Person schema, which names an author, is 9.4 times as common on cited pages as on the web.

Figure 3

Pages with Person schema, out of every 100

2in 100 web pages

19in 100 cited pages

The next six typesCitedWebRatio
ImageObject21.4%2.4%8.9×
NewsArticle10.4%1.2%8.7×
SoftwareApplication2.4%0.3%8.0×
Service1.3%0.2%6.5×
BreadcrumbList37.7%7.3%5.2×
WebPage29.3%5.8%5.1×
Source: Trakkr crawl of 1,465 AI-cited pages, March 2026. Web rates: HTTP Archive Web Almanac 2024. Any schema: 67.8% of cited pages, 38.5% of the web.Data

Light schema beats heavy schema.

More markup doesn’t mean more citations. Pages with light schema, 1 to 20 fields, average 30.5 citation appearances, the most of any tier. After that, the count falls with every step up.

Pages with no schema at all average 24.1, about the same as the heaviest tier. Focused, accurate markup seems to beat exhaustive markup. Whether AI prefers it, or it simply marks better-kept pages, is still open. None of this is a reason to strip useful schema from a page that has it.

Figure 4

Citations peak at light schema, then fall with every step up

Average citation appearances per page

Average citations by number of schema fields
Schema fieldsPagesAverage wordsAverage citations
No schema146184324.1
1 to 20 fields135255230.5
21 to 50 fields89231026.6
51 to 100 fields72264724.8
Over 100 fields12247923.7
Pages grouped by the number of schema fields they carry. Source: Trakkr crawl of AI-cited pages, March 2026. 454 pages in these groups.Data

Cited pages run long.

The average cited page runs 2,290 words, about three times the 750 of a typical web page. The median is 1,723, so a tail of very deep pages pulls the average up. 78% pass 1,000 words.

Length is also the strongest rival explanation for everything else here. The FAQ pages in chapter 01 average 4,290 words. Depth may be doing more of the work than any single tag.

Figure 5

A cited page is three times the length of a typical one

Pages drawn to scale by word count. The dashed line marks 1,000 words; 78% of cited pages run past it.

Words per page
Typical web page750 words
Cited page, median1,723 words
Cited page, average2,290 words
FAQ schema pages4,290 words
Words per page. Source: Trakkr crawl of AI-cited pages, March 2026. Web average: HTTP Archive Web Almanac 2024.Data

Depth may be doing more of the work than any single tag.

From chapter 04

The best-cited pages differ less than you’d think.

Put the most-cited tenth of pages beside the bottom half and the gaps are modest. Schema shows the clearest split: 80% of the top tenth carry it, against 65.6% of the bottom half. Article schema and tables come next.

Structure barely moves. Headings, FAQ sections and how-to content look alike in both groups. Cited pages already clear a high bar, so no single on-page tweak lifts one above the rest.

Figure 6

The most-cited pages look a lot like the rest

Share of pages with each trait. The shorter the green line, the smaller the difference.

Words per page
2,521.1vs 2,304.7
Headings per page
33.7vs 31.4
List items per page
146.9vs 120.3
Top 10% most-cited pages against the bottom 50%
FeatureTop 10%Bottom 50%Difference
Any schema80.0%65.6%+14.4 percentage points
Article schema37.8%23.3%+14.5 percentage points
FAQ schema11.1%5.3%+5.8 percentage points
Person schema17.8%19.4%−1.6 percentage points
Words per page2,521.12,304.7+216.4
Headings per page33.731.4+2.3
List items per page146.9120.3+26.6
Tables40.0%28.2%+11.8 percentage points
FAQ content42.2%38.3%+3.9 percentage points
How-to content73.3%70.0%+3.3 percentage points
Pages ranked by citation appearances: top 10% against the lowest 50%. Source: Trakkr crawl of AI-cited pages, March 2026. Gaps are in percentage points.Data

What we think it means.

Four things we’d do with a page we wanted cited. None is a guarantee; each is grounded in the data above.

  1. Write the whole answer first

    Depth is the most consistent trait of cited pages. Cover the question fully before you polish any tag.

  2. Name your authors

    Person schema is the most over-represented type. A named, credible author is common ground among cited pages.

  3. Mark up the questions you already answer

    FAQ schema is the one type tied to more citations. The sample is small, so try it on a few pages and measure.

  4. Keep markup light and true

    Pages with up to 20 fields were cited more than pages with over 100. Add what describes the page and skip the rest.

How we ran it.

Everything you need to check our work, repeat it or cite it: who ran the study, when, on what data, and where it falls short.

Trakkr Research · Study 008

The anatomy of an AI citation

Mack GrenfellFounder, Trakkr. Author and analyst.
Sample
1,465 pagesfrom 950 domains, out of 3,317 we crawled
Drawn from
28,033citation appearances in tracked AI answers
Window
11 weeks10 January to 27 March 2026, analysed 30 March
Answers from
3 modelsChatGPT, Perplexity and Gemini

Method

We took the most-cited URLs from Trakkr’s citation tracking and crawled each one live. From every page we pulled its JSON-LD schema, counted words, headings, lists and tables, detected FAQ and how-to patterns, and read its canonical, Open Graph and description tags. Then we compared each trait with web-wide rates.

Benchmark

HTTP Archive Web Almanac 2024

Data

schema-advantage.jsonAggregates only. No page or customer data.

Where it falls short

  • A small FAQ sample

    Only 23 pages carry FAQ schema and FAQ content. Treat the 45% as an early signal.

  • No uncited control

    We compare cited pages with the web, not with similar pages that weren’t cited.

  • Presence, not quality

    We check that schema exists, not that it is valid or correct.

  • A B2B and DTC tilt

    The pages come from brands Trakkr tracks, so the findings fit B2B, SaaS and DTC best.

Cite this study

Grenfell, M. (2026, April 9). The anatomy of an AI citation (Trakkr Research Study 008). Trakkr. https://trakkr.ai/trakkr-research/anatomy-of-an-ai-citation

Questions,
answered.

The ones we hear most about this study.

Partly. 67.8% of cited pages carry schema, against 38.5% of the web. Only FAQ schema paired with a visible FAQ section is tied to more citations: 45% more, across just 23 pages. Those pages also run long, so treat it as an early signal.

Ask anything about this studyAsk anything about this study

The Agent answers from our docs, methods and pricing, and links to the source.

Keep reading.

All Trakkr Research
Study 009 · 22 April 202673.5%

of AI citation URLs appear once and never return. Brand presence halves in a median 31 days.

The half-life of AI citationsStudy 010 · 6 May 202675.9%

smaller median payload as Markdown. Four of the five crawlers showed no clear preference.

Do AI crawlers prefer Markdown?Study 011 · 2 July 2026+3.6pp

lift in recommendations in the week a brand page is first cited. Being cited is not being chosen.

Cited, not chosen

Page audit

lope.example
Schema markupArticle and BreadcrumbList
Named authorPriya Shah, with Person schema
Depth2,140 words, 14 headings
FAQ markupMissing on 3 questionsFix drafted

Drafted FAQPage markup

{ "@type": "FAQPage",
  "mainEntity": [{ "@type": "Question",
    "name": "How long does a daily trainer last?" …

Is your page built to be cited?

Trakkr checks your pages against what AI cites, then drafts the fixes for you to approve.

Start free trialGet a free snapshot