Do AI engines actually rely on structured schema, visible author bylines, and recently updated content when choosing what to cite, or do those signals differ sharply across engines and page types?
We inspected parsed HTML and JSON-LD metadata on 329,519 unique pages cited by organic ChatGPT Search, Google AI Mode, and Perplexity answers from November 1 to November 19, 2026. We measured Schema.org types, author bylines, HTML/JSON-LD dates, content age across 16 source categories, and multi-question reach.
Across 329,519 unique AI-cited HTML pages inspected in November 2026, 70.86% carry JSON-LD schema, 30.86% expose an author byline, and 41.92% expose a published or modified date (median age 74.5 days across 138,134 valid dated pages). Perplexity cites pages with author bylines (55.35%) and HTML dates (68.07%) at roughly double ChatGPT Search's rates (26.42% and 37.88%).
Perplexity and Google AI Mode favor schema, bylines, and dates far more than ChatGPT Search
Structured signals separate the three engines cleanly. On Perplexity (5,371 HTML pages), 85.55% of cited pages carry JSON-LD schema, 52.71% carry Article-family schema, 55.35% expose an author byline, and 68.07% expose a parsed HTML date.
Google AI Mode (79,824 HTML pages) sits in the middle at 75.86% JSON-LD, 38.23% Article-family schema, 44.77% author bylines, and 55.18% HTML dates. ChatGPT Search (263,838 HTML pages) cites docs, product pages, and undated evergreen URLs much more freely: only 26.42% carry a byline and 37.88% carry a parsed date.
- Carry JSON-LD schema
- 70.86%
233,513 of 329,519 cited HTML pages
From 69.48% on ChatGPT Search to 85.55% on Perplexity
- Expose an author byline
- 30.86%
101,703 of 329,519 pages
Perplexity (55.35%) is over 2x ChatGPT Search (26.42%)
- Carry a parsed HTML/JSON-LD date
- 41.92%
138,147 pages total; 138,134 valid in age cohort
58.08% expose no parsed HTML date
- Median age of dated pages
- 74.5 days
Perplexity 56.8d · ChatGPT 63.9d · Google AI Mode 101.9d
Google AI Mode cites pages 79% older than Perplexity
Structured metadata presence on AI-cited HTML pages by engine
- ChatGPT Search
- Google AI Mode
- Perplexity
- Any JSON-LD
- ChatGPT Search69.48%
- Google AI Mode75.86%
- Perplexity85.55%
- Article family schema
- ChatGPT Search20.83%
- Google AI Mode38.23%
- Perplexity52.71%
- Author byline
- ChatGPT Search26.42%
- Google AI Mode44.77%
- Perplexity55.35%
- Any parsed date
- ChatGPT Search37.88%
- Google AI Mode55.18%
- Perplexity68.07%
Takeaway
FAQPage schema spikes to 30.7% on Perplexity, while Product schema holds near 14% on ChatGPT and Perplexity
Across all 329,519 unique cited HTML pages, Article-family markup (Article, BlogPosting, or NewsArticle) appears on 24.99%, FAQPage on 19.4%, Product or SoftwareApplication on 13.34%, and Review or AggregateRating on 7.44%.
Perplexity leads every editorial schema category, hitting 30.7% FAQPage and 9.2% Review-family markup. Google AI Mode carries the lowest Product-family schema share (10.68% vs 14.15% on ChatGPT Search and 14.17% on Perplexity), aligning with Google's tilt toward editorial commentary over vendor product pages.
Full schema, byline, and date profile by AI engine
JSON-LD (%) · 3 results
- Perplexity85.55%
- Google AI Mode75.86%
- ChatGPT Search69.48%
Explore the data (3 rows)
| ChatGPT Search | 263,838 | 69.48 | 20.83 | 19.38 | 14.15 | 7.79 | 26.42 | 37.88 |
| Google AI Mode | 79,824 | 75.86 | 38.23 | 19.91 | 10.68 | 6.26 | 44.77 | 55.18 |
| Perplexity | 5,371 | 85.55 | 52.71 | 30.7 | 14.17 | 9.2 | 55.35 | 68.07 |
Takeaway
Entity and navigation schema appear on half of cited pages, while FAQPage (19.4%) tops Article (16.21%)
Looking at individual Schema.org @type tags across all 329,519 HTML pages shows why raw JSON-LD totals look high. Baseline site-identity and navigation types lead the list: Organization appears on 50.86% (167,577 pages), ListItem on 44.19% (145,611), BreadcrumbList on 43.14% (142,164), ImageObject on 36.29%, WebPage on 35.62%, and WebSite on 34.58%.
Among content-specific types, Question (19.52%), Answer (19.51%), and FAQPage (19.4%, 63,926 pages) edge out the bare Article type (16.21%, 53,416 pages) and more than double Product (8.03%, 26,476 pages) and ItemList (7.34%, 24,187 pages).
Top 20 JSON-LD Schema.org @type occurrences across 329,519 AI-cited pages
Share of cited pages (%) · Top 10 of 20
- Organization50.86%
- ListItem44.19%
- BreadcrumbList43.14%
- ImageObject36.29%
- WebPage35.62%
- WebSite34.58%
- Person26.02%
- Question19.52%
- Answer19.51%
- FAQPage19.4%
Explore the data (20 rows)
| Organization | 167,577 | 50.86 |
| ListItem | 145,611 | 44.19 |
| BreadcrumbList | 142,164 | 43.14 |
| ImageObject | 119,571 | 36.29 |
| WebPage | 117,387 | 35.62 |
| WebSite | 113,941 | 34.58 |
| Person | 85,725 | 26.02 |
| Question | 64,315 | 19.52 |
| Answer | 64,293 | 19.51 |
| FAQPage | 63,926 | 19.4 |
| SearchAction | 60,516 | 18.36 |
| Article | 53,416 | 16.21 |
| EntryPoint | 49,837 | 15.12 |
| PostalAddress | 45,896 | 13.93 |
| ReadAction | 41,256 | 12.52 |
| Offer | 37,517 | 11.39 |
| ContactPoint | 36,603 | 11.11 |
| PropertyValueSpecification | 36,396 | 11.05 |
| Product | 26,476 | 8.03 |
| ItemList | 24,187 | 7.34 |
Takeaway
Only 30.86% of cited pages carry an author byline and 41.92% expose a parsed date
While 70.86% (233,513) of cited HTML pages carry JSON-LD, explicit editorial signals are far less common. Only 30.86% (101,703 pages) expose an author byline in parsed HTML or metadata, and only 41.92% (138,147 pages: 119,966 with a published timestamp and 124,627 with a modified timestamp) expose a machine-readable page date.
Nearly three in five cited HTML pages (58.08%) carry no parsed published or modified date at all. Missing dates or author blocks do not disqualify documentation, product pages, or vendor homepages from earning citations.
- Pages with JSON-LD schema
- 233,513
70.86% of 329,519 HTML pages
- Pages with an author byline
- 101,703
30.86% of 329,519 HTML pages
- Pages with a parsed date
- 138,147
41.92% (119,966 published; 124,627 modified)
- Dated pages in 2005-2026 window
- 138,134
74.5 days median age (13 invalid dates excluded)
Takeaway
Google AI Mode cites pages 79% older (101.9 days median) than Perplexity (56.8 days)
Across the 138,134 dated pages with valid published or modified timestamps between 2005 and November 2026, median page age at inspection is 74.5 days. Freshness differs sharply by engine.
Perplexity leans newest with a 56.8-day median (35.09% updated within 30 days and 61.68% within 90 days), followed by ChatGPT Search at 63.9 days (34.28% within 30 days and 57.58% within 90 days). Google AI Mode pulls a noticeably older mix at 101.9 days median (79% older than Perplexity), with only 26.47% within 30 days and 19.36% over one year old.
Share of dated AI-cited pages across content age brackets by engine
- ChatGPT Search
- Google AI Mode
- Perplexity
Explore the data
| <= 30d | 34.28 | 26.47 | 35.09 |
| 31-90d | 23.3 | 20.66 | 26.59 |
| 91-180d | 14.86 | 17.17 | 15.21 |
| 181-365d | 12.57 | 16.33 | 12.96 |
| 1-2y | 6.85 | 9.78 | 5.58 |
| 2y+ | 8.14 | 9.58 | 4.57 |
Takeaway
Editorial pages turn over 4.7x faster (42.7 days median) than government sources (202.1 days)
Content age follows the publisher's update model more than a blanket AI recency rule. Across 16 source categories with at least 500 dated pages, editorial sites (42.7 days median; 80.83% within six months), directories (46.1 days), and blogs (52.4 days) carry the newest dates, followed by documentation (63.3 days) and SaaS sites (66.8 days).
At the other end, developer hubs (111.8 days), education sites (131.6 days), health publishers (160.9 days), social threads (192.9 days), government domains (202.1 days, 4.7x older than editorial), and community forums (240.3 days; 26.19% over two years old) stay cited years after publication.
Median content age and freshness shares across 16 source categories
Median age (days) · Top 10 of 16
- community240.3
- government202.1
- social192.9
- health160.9
- education131.6
- developer111.8
- reference101.4
- ecommerce91.2
- news87.5
- general77.4
Explore the data (16 rows)
| editorial | 14,785 | 42.7 | 40.98 | 80.83 | 90.59 | 5.2 |
| directory | 3,114 | 46.1 | 40.91 | 76.27 | 87.25 | 7.45 |
| blog | 4,050 | 52.4 | 37.09 | 76.99 | 88.59 | 6.59 |
| documentation | 7,157 | 63.3 | 34.99 | 74.78 | 85.93 | 6.29 |
| saas | 37,999 | 66.8 | 32.29 | 73.22 | 86.53 | 6.54 |
| finance | 3,775 | 72.8 | 31.66 | 68.4 | 81.51 | 9.75 |
| general | 33,414 | 77.4 | 30.92 | 69.67 | 85.44 | 7 |
| news | 5,660 | 87.5 | 31.93 | 65.23 | 78.99 | 12.49 |
| ecommerce | 6,247 | 91.2 | 33.02 | 62.03 | 73.57 | 16.62 |
| reference | 9,358 | 101.4 | 26.35 | 62.94 | 76.46 | 13.45 |
| developer | 1,610 | 111.8 | 26.09 | 62.24 | 81.74 | 9.5 |
| education | 1,513 | 131.6 | 23.13 | 55.72 | 71.32 | 15.99 |
| health | 1,865 | 160.9 | 21.66 | 52.87 | 71.21 | 16.41 |
| social | 3,856 | 192.9 | 15.12 | 47.38 | 69.09 | 14.5 |
| government | 2,498 | 202.1 | 16.37 | 46.56 | 66.65 | 20.06 |
| community | 733 | 240.3 | 15.14 | 45.29 | 57.16 | 26.19 |
Takeaway
Multi-question news and editorial pages carry twice the FAQPage schema and length, while SaaS stays flat
Comparing pages cited for 1 question against workhorse pages cited for 3 or more questions within the same source type separates real signals from site-category averages. On news sites, 3+-question pages double FAQPage adoption (27.78% vs 13.84%) and run 75% longer in median markdown length (17,366 bytes vs 9,928 bytes). Editorial 3+-question pages rise to 93.54% JSON-LD, 49.35% FAQPage, and 20,349 bytes; blogs climb from 66.99% to 75.72% author bylines and 76.10% to 85.12% dates.
By contrast, SaaS and documentation show almost zero metadata lift from 1 question to 3+ questions. SaaS sits flat on JSON-LD (77.46% vs 77.74%), FAQPage (26.47% vs 27.14%), and length (7,952 vs 8,491 bytes), and documentation sits flat on author bylines (16.64% vs 17.24%) and FAQPage (1.55% vs 2.21%).
Metadata and page size by source type and question reach
Filter by source type to see how JSON-LD, FAQPage, bylines, dates, and page size change between 1-question and 3+-question pages.
FAQPage (%) · Top 10 of 21
- editorial · 3+ questions49.35%
- editorial · 2 questions47.83%
- editorial · 1 question46.24%
- blog · 2 questions39.3%
- blog · 3+ questions38.44%
- blog · 1 question35.34%
- news · 3+ questions27.78%
- saas · 2 questions27.73%
- saas · 3+ questions27.14%
- saas · 1 question26.47%
Explore the data (21 rows)
| saas | 1 question | 53,419 | 77.46 | 26.47 | 30.88 | 45.92 | 7,952 |
| ecommerce | 1 question | 29,236 | 79.11 | 12.46 | 11.23 | 15.44 | 5,379 |
| documentation | 1 question | 17,978 | 41.46 | 1.55 | 16.64 | 24.89 | 5,981 |
| saas | 3+ questions | 17,572 | 77.74 | 27.14 | 29.02 | 44.74 | 8,491 |
| saas | 2 questions | 12,966 | 77.19 | 27.73 | 29.62 | 45.03 | 8,260 |
| editorial | 1 question | 11,882 | 92.23 | 46.24 | 69.14 | 81.6 | 16,738 |
| directory | 1 question | 7,717 | 74.96 | 21.8 | 20.93 | 29.96 | 8,521 |
| documentation | 3+ questions | 5,933 | 47.06 | 2.21 | 17.24 | 25.54 | 7,023 |
| news | 1 question | 5,579 | 80.61 | 13.84 | 55.51 | 68.65 | 9,928 |
| ecommerce | 2 questions | 4,932 | 78.79 | 14.46 | 11.48 | 18.21 | 5,584 |
| ecommerce | 3+ questions | 4,743 | 79.93 | 15.81 | 9.47 | 17.9 | 5,395 |
| documentation | 2 questions | 4,383 | 44.44 | 1.73 | 17.29 | 27.1 | 6,711 |
| blog | 1 question | 3,862 | 88.53 | 35.34 | 66.99 | 76.1 | 12,454 |
| editorial | 3+ questions | 3,234 | 93.54 | 49.35 | 71.21 | 86.61 | 20,349 |
| editorial | 2 questions | 2,862 | 92.56 | 47.83 | 69.88 | 83.82 | 18,155 |
| news | 3+ questions | 1,465 | 85.53 | 27.78 | 52.9 | 72.9 | 17,366 |
| directory | 3+ questions | 1,420 | 80.28 | 21.76 | 19.72 | 27.11 | 11,926 |
| directory | 2 questions | 1,388 | 80.04 | 24.64 | 22.26 | 31.77 | 11,776 |
| news | 2 questions | 1,164 | 81.87 | 21.05 | 53.52 | 69.76 | 12,699 |
| blog | 2 questions | 710 | 88.59 | 39.3 | 70.7 | 78.87 | 14,808 |
| blog | 3+ questions | 692 | 92.63 | 38.44 | 75.72 | 85.12 | 15,454 |
Takeaway
Study scope and counting rules
This analysis inspects 329,519 unique AI-cited HTML pages drawn from each page's latest November 1 to November 19, 2026 fetch that returned parsed HTML metadata across organic ChatGPT Search (263,838 pages), Google AI Mode (79,824 pages), and Perplexity (5,371 pages) answers. Because an identical URL can be cited by more than one engine, per-engine page counts sum above the unique page total.
Schema.org @type counts inspect parsed JSON-LD blocks only; microdata and RDFa outside JSON-LD are not included. Effective page age uses the parsed modified timestamp falling back to published timestamp and covers 138,134 valid dated pages between January 1, 2005 and inspection time (excluding 13 out-of-range timestamps from the 138,147 pages with any parsed date). All metrics describe properties observed on cited pages and do not constitute a randomized test against uncited pages.
What to do with these numbers
Match your metadata work to your page type and target engine instead of bolting identical Article and FAQPage templates onto every URL. On vendor blogs, guides, and editorial comparisons, keep visible author bylines, valid published/modified timestamps, and structured Q&A blocks fresh inside a 45-to-60-day cadence, especially if Perplexity and Google AI Mode matter in your category.
On SaaS product pages and technical documentation, focus on Organization identity markup, Product/Offer clarity, and factual accuracy rather than inflating page length or forcing FAQPage schema where buyers only need crisp specs.
Metadata and freshness priorities by content layer
| Technical documentation | 41.46% to 47.06% JSON-LD; 1.55% to 2.21% FAQPage | Update specs on release rather than adding synthetic bylines or FAQ wrappers. |
| SaaS product and pricing pages | 77.7% JSON-LD; flat ~8 KB size across reach tiers | Keep Organization and Product/Offer facts clean; avoid bloating product pages for length. |
| Perplexity and Google AI Mode targets | 44.77% to 55.35% author; 55.18% to 68.07% dated | Make authorship and last-updated dates unambiguous in both visible HTML and JSON-LD. |
| Editorial guides and company blog | 89% to 94% JSON-LD; 42.7d to 52.5d median age | Expose Person author bylines, accurate modified dates, and FAQPage sections on multi-question guides. |
Takeaway
How we measured
We analyzed 329,519 unique AI-cited HTML pages from their latest November 1 to November 19, 2026 crawl inspection across organic ChatGPT Search (263,838 pages), Google AI Mode (79,824 pages), and Perplexity (5,371 pages) answers. Content-age distributions cover 138,134 dated pages with valid HTML or JSON-LD dates between 2005 and November 2026.
- AI-cited HTML pages carrying JSON-LD schema
- 70.86%
233,513 of 329,519 unique cited HTML pages
- AI-cited HTML pages exposing an author byline
- 30.86%
101,703 of 329,519 pages (55.35% on Perplexity vs 26.42% on ChatGPT)
- Median content age across 138,134 dated pages
- 74.5 days
41.92% of cited pages expose a parsed HTML or JSON-LD date
- Cited pages carrying FAQPage schema
- 19.4%
63,926 pages, topping Article @type directly (16.21%)
Get the data
Sources
- Ahrefs: Does schema markup improve AI visibility? (opens in a new tab) · accessed November 20, 2026
- Ahrefs: Do AI assistants prefer to cite fresh content? (opens in a new tab) · accessed November 20, 2026
- Google Search Central: Changes to FAQ and HowTo rich results (opens in a new tab) · accessed November 20, 2026