Key takeaways
Both page groups use the full desktop or mobile website sample as their denominator.
In the July 2025 crawl, JSON-LD detection was more common on homepages in both samples. Microdata showed the reverse pattern.
Track format detection, markup accuracy and observed citations as separate checks.
Use saved answers and cited URLs to decide which product, article or documentation pages to review.
Read a structured data benchmark by checking what was counted and what it was divided by. HTTP Archive's July 2025 crawl provides a useful example. Its published results count websites with a detected format on sampled pages. For each desktop or mobile group, the denominator is the entire website sample. That choice shapes how a content team should interpret the difference between homepages and secondary pages. [1]
Use the benchmark to frame a page review. If a homepage already has markup, inspect the product, article or documentation pages that answer your audience's questions. Then compare what those pages say, what their markup describes and which URLs appear in saved AI answers. Each observation supports a different decision.
Start with the unit of measurement
The two charts use figures 5.50 and 5.51 of the 2025 Web Almanac and the results and query linked from those figures. HTTP Archive collected the data in July 2025 and published the SEO chapter in January 2026. The four formats in this analysis are JSON-LD, Microdata, RDFa and Microformats2. The crawl sample comes from sites eligible for the Chrome UX Report, with separate desktop and mobile runs. [1]
For each format and page group, the query counts distinct websites where the collector detected structured data after rendering. A site with a detection on any sampled secondary page contributes once to that format's secondary-page count. Each desktop rate divides by all 12,155,374 desktop sites. Each mobile rate divides by all 15,426,398 mobile sites. [1]
The denominator makes the comparison intelligible. A count of all marked-up URLs would give a site with many sampled pages more weight. Here, a website contributes at most one count per format and group. When a team wants to know how many of its own product URLs have accurate markup, it can build a separate measure using the product URLs eligible for its review as the denominator. That result describes the team's work directly.
Read the differences by format
Website-level detection in sampled homepages and secondary pages, from the July 2025 HTTP Archive crawl.
- Homepage detection
- Secondary-page detection
Full data table
| Detected format | Homepage detection | Secondary-page detection |
|---|---|---|
| JSON-LD | 43.0805 | 38.9663 |
| Microdata | 16.8125 | 19.1033 |
| RDFa | 1.0809 | 1.9585 |
| Microformats2 | 0.1377 | 0.2109 |
Each bar uses all 12,155,374 websites in the desktop sample as its denominator. The secondary-page bar counts websites with a detection on a sampled secondary page. Sites can contribute to several formats. Ansvra recalculated the rates from HTTP Archive's counts. [1]
Website-level detection in sampled homepages and secondary pages, from the July 2025 HTTP Archive crawl.
- Homepage detection
- Secondary-page detection
Full data table
| Detected format | Homepage detection | Secondary-page detection |
|---|---|---|
| JSON-LD | 42.8346 | 37.2852 |
| Microdata | 17.5294 | 18.8551 |
| RDFa | 1.0299 | 1.9361 |
| Microformats2 | 0.1717 | 0.2411 |
Each bar uses all 15,426,398 websites in the mobile sample as its denominator. The secondary-page bar counts websites with a detection on a sampled secondary page. Sites can contribute to several formats. Ansvra recalculated the rates from HTTP Archive's counts. [1]
JSON-LD has the highest detection share in both page groups for both clients. It appears on sampled homepages for 43.08% of desktop sites and 42.83% of mobile sites. The corresponding secondary-page shares are 38.97% and 37.29%. Based on the original counts, the homepage share is higher by 4.11 percentage points for desktop and 5.55 for mobile. [1]
Microdata shows a different pattern. The secondary-page share is higher in each client sample. Reading these formats individually gives the team better questions to ask about its templates. A product page could carry one form of markup while an article page on the same site uses another. The review should identify the page's purpose and the relevant properties before anyone chooses a format to standardize.
Consider a team moving product markup from Microdata to JSON-LD. Track format and fact coverage in separate columns throughout the rollout. Both encodings might appear on some pages during the transition. An increase in JSON-LD detections could reflect an encoding change even if the same product facts are available before and after. A property check tells the team whether important details survived the move.
Apply the same discipline when comparing desktop and mobile. The charts show separate client samples. To investigate a difference on your own site, choose the same URLs for both checks and record the rendered facts on each version. A matched review gives the template owner a specific discrepancy to reproduce, such as a mobile page where a product detail disappears after rendering.
These comparisons use aggregate counts. To measure how often a site with homepage markup also has it on secondary pages, you would need to match the same websites across the two page groups. To measure adoption across all pages, you would need a page-level numerator and denominator. Those would answer different questions from the published comparison. [1]
Client | Homepage JSON-LD | Secondary-page JSON-LD | Difference in percentage points |
|---|---|---|---|
Desktop | 43.0805% | 38.9663% | 4.1141 |
Mobile | 42.8346% | 37.2852% | 5.5494 |
The differences use the unrounded site counts. Subtract the secondary-page count from the homepage count, divide by the corresponding client sample, then multiply by 100. The percentages in the table are rounded for display. [1]
Inspect the full count table
The table includes all 16 results in the selected scope. Each percentage is an Ansvra calculation from HTTP Archive's published numerator and the matching client denominator. A site may contribute to several format rows. Read each percentage on its own. [1]
Client | Page group | Format | Websites with detection | Website sample | Share |
|---|---|---|---|---|---|
Desktop | Homepage | JSON-LD | 5,236,590 | 12,155,374 | 43.0805% |
Desktop | Homepage | Microdata | 2,043,626 | 12,155,374 | 16.8125% |
Desktop | Homepage | RDFa | 131,390 | 12,155,374 | 1.0809% |
Desktop | Homepage | Microformats2 | 16,734 | 12,155,374 | 0.1377% |
Desktop | Secondary pages | JSON-LD | 4,736,505 | 12,155,374 | 38.9663% |
Desktop | Secondary pages | Microdata | 2,322,072 | 12,155,374 | 19.1033% |
Desktop | Secondary pages | RDFa | 238,057 | 12,155,374 | 1.9585% |
Desktop | Secondary pages | Microformats2 | 25,639 | 12,155,374 | 0.2109% |
Mobile | Homepage | JSON-LD | 6,607,832 | 15,426,398 | 42.8346% |
Mobile | Homepage | Microdata | 2,704,155 | 15,426,398 | 17.5294% |
Mobile | Homepage | RDFa | 158,871 | 15,426,398 | 1.0299% |
Mobile | Homepage | Microformats2 | 26,489 | 15,426,398 | 0.1717% |
Mobile | Secondary pages | JSON-LD | 5,751,756 | 15,426,398 | 37.2852% |
Mobile | Secondary pages | Microdata | 2,908,660 | 15,426,398 | 18.8551% |
Mobile | Secondary pages | RDFa | 298,667 | 15,426,398 | 1.9361% |
Mobile | Secondary pages | Microformats2 | 37,190 | 15,426,398 | 0.2411% |
The description accompanying figure 5.50 gives 16% for mobile homepage Microdata. The linked results list 2,704,155 websites with a detection. Dividing that count by 15,426,398 yields 17.5294%, consistent with the fraction in the results. The charts and tables here follow the linked counts and query. Source data from HTTP Archive's 2025 Web Almanac, licensed under Apache 2.0. [1]
Assess what the markup says about the page
Detection is a starting point for an editorial or technical review. Take a product page with a price and availability displayed on screen. The reviewer can check whether the markup describes the same product and matches those visible facts. If availability has changed, the ticket can name the affected template, the value a shopper sees and the value recorded in markup. That gives the engineer something specific to reproduce.
Google's structured data guidance lists JSON-LD, Microdata and RDFa as supported formats and recommends JSON-LD when it is easiest to maintain. The crawl also records Microformats2, which is useful for describing what the collector found. For a Google rich result review, check the supported feature, its required properties and Google's quality rules. The appearance of a rich result remains a separate observation. [2] [3]
A reviewer can make the check repeatable by saving the exact URL, the visible content, the properties inspected and the date. Group repeated findings by template to decide where a shared fix would help. If the issue belongs to one page, record the specific change and ask the page owner to confirm the corrected facts. Save what you checked so a later reviewer can reproduce the decision.
Choose which pages to investigate with answer evidence
Start with a question your audience asks. Suppose buyers want to know which of your products supports a particular material or use. Find the product or guide that explains the answer, then review the cited URLs from saved AI answers to that question. A citation gives you a specific page to read in context. It can reveal which fact deserves clearer placement or which page already gives a good explanation. Explore Ansvra AI Citations.
When several answers point to the same URL, examine what that page covers and how clearly it states the facts those questions depend on. If the audience needs compatibility details, look for an explicit compatibility statement, then check that relevant markup represents the same product information. A template issue and an incomplete explanation can have separate owners and separate verification steps. Explore Citation Source Analysis.
Follow the steps to scope a page review, check the markup and record what appears in saved answers.
- Choose pages
Choose the homepage and the product, article or documentation URLs that answer the tracked questions. Use the July 2025 benchmark as context for the page review. [1]
Next: Check the sample
- Check the sample
The published query divides distinct websites with a detected format by all distinct websites for that client, including for the secondary-page group. [1]
Next: Read formats
- Read formats
Check which format is present on the rendered page. Review the required properties and compare them with the visible facts in the next step. [1] [2]
Next: Check page facts
- Check page facts
Review the relevant supported feature, visible content, factual accuracy and applicable technical and quality requirements. Google does not guarantee rich results. [2] [3]
Next: Review citations
- Review citations
Read saved answers for the questions you track. Record cited URLs and report visits separately. Google's general Search eligibility guidance applies to its AI features. [4]
Record the evidence for each step. Revisit saved answers to observe which URLs are cited after the review. [1] [2] [3] [4]
Google's eligibility guidance for AI Overviews and AI Mode follows its general Search requirements, including indexing and eligibility for snippets. Use the guidance to assess the page on those surfaces. To observe whether the URL is actually cited, return to saved answers for the questions and surfaces you track. Record the question, date, surface and cited page so the team can compare observations in context. [4]
Keep a short review note for each change. Record the reader question, the URL, the fact that needed clarification, the template or copy updated and the date you checked the result. Track whether later answers cite the same URL and keep visits in their own reporting column. If an answer changes alongside a page update, the note helps you investigate timing and other changes before attributing the difference to one edit.
Keep the historical sample in view
The July 2025 figures describe the websites and pages HTTP Archive sampled at that time. Desktop and mobile use their own website counts. The results group detections by format, so page-by-page adoption, property accuracy and AI citations require their own evidence. [1]
When setting a target for a content team, first name the pages the target covers and the check each page must pass. A review might cover the product detail pages that answer a defined set of buyer questions, with a pass based on whether the visible facts and relevant properties agree. Keep the list of pages stable when comparing review rounds and document any additions. That gives the team a count it can investigate URL by URL, alongside the answer evidence it is tracking.
Summary
For your next review, choose URLs tied to a specific reader question. Record the sample, check the visible facts and relevant markup, then revisit saved answers for citation evidence. Give each finding a URL and a date so the next review can compare the same pages.
Discussion
Join the discussion