AIScan corpus study cover: Homepage HTML, four parts measured. 238 rendered pages with no JSON-LD, 855 homepages scanned, with a bar chart motif
AIScan corpus study cover: Homepage HTML, four parts measured. 238 rendered pages with no JSON-LD, 855 homepages scanned, with a bar chart motif
AI Readiness

Structured HTML on 855 Homepages: JSON-LD Is Missing More Often Than the h1

855 homepages scanned for the four C3 parts: 30.1% of rendered pages lack JSON-LD, 20.1% lack an h1 and 14.8% a meta description. Full counts by platform.

18 Nov 2026 17 min read
#corpus study#structured data#h1#AI readiness

This guide covers C3 · Content, E3 · Content.

Table of contents

Of 855 homepages in the AIScan scan database on 18 November 2026, 454 (53.1%) pass the structured HTML check, C3, with all four parts in place: one <h1>, a <title>, a meta description and a JSON-LD block. The part most often missing is not the h1. Among the 792 pages that rendered any HTML at all, 238 (30.1%) had no JSON-LD, 159 (20.1%) had no h1 and 117 (14.8%) had no meta description.

Every count below comes from the latest site scan per host on rubric 2026.08.2, with aiscan.site itself left out. No scanned site is named. You can run the same check on your own homepage in about ten seconds with npx aiscan-cli yoursite.com or by pasting the URL at aiscan.site, free and without an account.

Quick summary

What we measured (855 homepages, 18 Nov 2026)CountShare
C3 pass: exactly one h1, title, meta description, JSON-LD45453.1%
C3 partial: one to three of the four parts missing or wrong33839.5%
C3 fail: all four missing (60 of 63 also fail E3)637.4%
Rendered pages with no JSON-LD block238 of 79230.1%
Rendered pages with no h1159 of 79220.1%
Rendered pages with no meta description117 of 79214.8%
Rendered pages with two or more h1 elements48 of 7926.1%
Rendered pages with no title12 of 7921.5%
Title, meta and JSON-LD all present, h1 count wrong799.2% of all
  • The biggest single gap is JSON-LD. 74 pages miss only that one part. Adding one Organization or WebSite block would turn each of them into a pass.
  • A missing h1 is three times as common as a doubled one among rendered pages: 159 against 48.
  • 58 pages with no h1 still pass E3, so they carry 150 or more words of server-rendered text and two or more h2 headings. Those pages have content and simply never mark up the main title.
  • Platform matters. Astro passed 41 of 52 (78.8%), Next.js 152 of 208 (73.1%), WordPress 129 of 211 (61.1%). Hosts where no platform was detected passed 102 of 319 (32.0%).

What C3 actually checks, and what it does not

According to AIScan's own content checks reference, C3 looks for "four basic signals on your homepage: a single <h1>, a non-empty <title>, a <meta name="description">, and at least one application/ld+json block. One point each." The check has a weight of 4 in the content dimension.

The evidence string makes the four parts visible. Every C3 result in the database reads like h1: 1, title: true, meta description: true, JSON-LD: true, which is what made this study possible: we parsed each string into an h1 count and three yes/no values and counted them.

Three limits matter before reading any number here.

  1. C3 asserts JSON-LD presence, not its type or validity. We documented this in the evidence-string audit on 8 September 2026. A block containing an empty object counts the same as a full Organization with sameAs links. "Has JSON-LD" in this post means "has at least one script tag of that type", nothing more.
  2. An h1 hidden with CSS still counts. The check reads the HTML the server returns. A visually hidden h1, or one placed off-screen, satisfies it.
  3. It is the homepage only. A site scan grades the URL you give it. Inner pages can be better or worse, and this corpus says nothing about them.

The fourth caveat is about the population. Everyone in this database chose to scan their own site, usually while working on it. It is a self-selected sample of people who care about AI readiness, not a random slice of the web.

Method: how the 855 homepages were counted

The scans table in AIScan's production database stores every report as JSON. On 18 November 2026 we ran four aggregate SQL queries against it:

  • one latest site-scope scan per host, hosts compared without scheme, path or case
  • rubric version 2026.08.2 only, so every result was graded by the same rules
  • aiscan.site excluded, because we scan it constantly and it always passes
  • the C3 evidence string parsed into its four parts, then cross-tabulated with the detected platform and with the E3 result

Aggregates only. No query selected a row, and no scanned host appears in this article. The only sites named are our own four, checked separately with curl the same morning.

Population filterHosts
Hosts with a C3 result on rubric 2026.08.2 (site scope, aiscan.site excluded)855
C3 fail, every part false63
Pages that rendered HTML (pass or partial)792
Hosts with a detected platform536
Hosts with no platform detected319

The refill run that queued this study counted 854 hosts on 17 November. The population moves every day as people scan, and every figure in this post is the 18 November re-run.

The 63 fails: every part missing at once

Every one of the 63 C3 fails has the same evidence: h1: 0, title: false, meta description: false, JSON-LD: false. Not a single fail has some of the parts. Of those 63, 60 also fail E3, the server-rendered text check, and their mean overall AIScan score is 18.

A C3 fail means the scanner got a document with none of the four parts. That happens two ways: a page that sends near-empty HTML and builds everything in the browser, or a site that answered the scanner with a bot challenge or an error body. Stored evidence cannot split the two. When the empty DOM study checked a sample of C3 failures by hand, only 4 of 26 were truly empty shells, so treat 63 as a ceiling on empty pages. If your own site is one of them, open the raw HTML: an empty body means server rendering or pre-rendering is the fix, and a challenge page means your firewall is turning crawlers away. For the rest of this post, the useful population is the 792 pages that did render something.

Which part is missing: the full breakdown

Missing part (rendered pages only)HostsShare of 792
JSON-LD23830.1%
h1 (zero h1 elements)15920.1%
Meta description11714.8%
h1 count above one486.1%
Title121.5%

A page can miss more than one part, so the rows add up to more than the 338 partials. The next table splits the partials by how many parts they miss.

C3 partials by parts missingHostsMean overall scoreOf those, E3 pass
One part missing163see next table123
Two parts missing1144144
Three parts missing613116
(C3 pass, nothing missing)45471436
(C3 fail, all four missing)63180

The overall score also covers 17 other checks, so this is a correlation and not a claim that C3 moves a score by 40 points. C3 is worth 4 points of weight. What the table does show is that pages missing several of these basics tend to miss other things too.

The 163 pages that are one fix away from a pass

The one missing partHostsMean overall scoreE3 pass
JSON-LD only744262
No h1, everything else present575831
Two or more h1, everything else present225922
Meta description only9527
Title only1351

This is the most practical table in the study. 163 homepages, 19.1% of everything we graded, would pass C3 with one edit. For almost half of them, that edit is a JSON-LD block. For another 79 it is fixing the h1 count: 57 have none and 22 have too many.

The h1 question: missing is the bigger problem

Search for "missing H1" and you will find years of forum threads, from the Shopify community to theme support boards, nearly all reacting to an SEO audit warning. Most of the advice online is about removing extra h1 elements. Our data points the other way.

h1 elements on the homepageAll 855Rendered 792
0222159
1585585
22424
377
433
5 or more1414

The highest count on a single homepage was 83 h1 elements, almost certainly a page builder or template that wraps every card title in an h1.

Among rendered pages, zero h1 is more than three times as common as two or more. And 58 of the 159 pages without an h1 pass E3: by the definition in the content checks reference, they carry at least 150 words of server-rendered text and at least two h2 headings. The page has a structure and text. It simply never says which heading is the page's title.

What the HTML standard and Google actually say

The WHATWG HTML Living Standard, in its section on headings and outlines (fetched from html.spec.whatwg.org on 18 November 2026), says verbatim: "If a document has one or more headings, at least a single heading within the outline should have a heading level of 1." That is the rule a zero-h1 page with h2 sections breaks.

The standard does not say a page must have exactly one h1. C3 is stricter than the spec on that point, and we say so plainly. AIScan asks for one because an agent extracting a page's title has to pick something, and one h1 removes the guess.

Google's own guidance points the same way. According to its title links documentation, fetched on 18 November 2026, says Google "looks at various sources when creating title links, including the main visual title, heading elements, and other large and prominent text, and it can be confusing if multiple headings carry the same visual weight and prominence." It suggests "putting the title text in the first visible <h1> element on the page." That is advice for search snippets, but an AI crawler extracting a page faces the same choice.

If your homepage has two or more h1 elements, our multiple h1 tags fix walks through the common causes (a logo wrapped in an h1 plus a hero heading is the usual one).

Meta descriptions: present on most, missing where a plugin would add them

115 homepages have a title but no meta description. Twelve more have neither. Google's snippet documentation, fetched 18 November 2026, says Google "will sometimes use the <meta name="description"> tag from a page to generate a snippet in search results" when it describes the page better than the on-page text would. For an AI agent, the meta description is often the one sentence it quotes when it summarises what a site is.

The meta description gap is concentrated. Of the 117 rendered pages without one, 36 are WordPress and 69 have no detected platform. Next.js had only 3 of 208 missing.

Results by platform

The scanner detects a platform from the page's own markers. Rows with fewer than four hosts are left out. 62 of the 63 empty shells had no platform detected and one was Vue, so every other named-platform row counts rendered pages only.

PlatformHostsC3 passPass rateNo h12+ h1No metaNo JSON-LD
WordPress21112961.1%51153627
Next.js20815273.1%146346
Astro524178.8%6149
Shopify16956.3%3343
Lovable14750.0%6103
SvelteKit7685.7%0001
Vue6116.7%3015
Nuxt55100%0000
Framer500%1305
Wix400%4001
None detected (rendered only)25710239.7%701869135

The small rows (Nuxt, Framer, Wix, Vue) are too small to say anything about the platform. They are listed so the totals add up, not as a verdict. A few patterns hold up in the larger rows.

WordPress: JSON-LD is fine, the h1 is not

WordPress pages almost always have JSON-LD (only 27 of 211 missing), which fits how common SEO plugins are on WordPress. The weak part is the h1: 51 WordPress homepages have none, and 36 of those pass E3. They have text and h2 sections and no top heading. In 41 WordPress cases the title, meta description and JSON-LD are all present and only the h1 count is wrong. We did not inspect the themes, so we cannot tell you which ones do this. Check your own front page with the command below; a theme that shows the site name as a logo image, or as styled text in a <p> or <div>, is the first place to look.

36 WordPress homepages also lack a meta description. WordPress core does not write one. ThinkRank is the plugin we build for this. According to its own feature page, fetched 18 November 2026, it writes meta titles and descriptions (in bulk with its AI tools) and outputs Organization schema through its Site Identity settings, and it also handles robots.txt, sitemaps and llms.txt, so one plugin covers what otherwise takes three that can conflict over the same files. It imports existing settings from Rank Math, Yoast, AIOSEO and SEOPress. Yoast SEO, Rank Math and All in One SEO also write meta descriptions and JSON-LD, and if one of them already runs on your site, the C3 parts it covers are already handled. Per-platform steps are in the title, meta and schema guide by platform.

Next.js: the h1 is there, JSON-LD is not

Next.js is the reverse. 188 of 208 Next.js homepages have exactly one h1, the best ratio of any large group, and only 3 lack a meta description, because the framework's metadata API makes both easy. But 46 have no JSON-LD. Next.js has no built-in structured data helper; you write the script tag yourself. The Next.js section of the platform guide has a working example.

Shopify: small sample, mixed results

Only 16 Shopify homepages are in this corpus, so treat the row with care: 9 pass, 4 lack a meta description and 3 have two or more h1 elements. On Shopify the theme decides both. StoreSEO is the Shopify app we build for this. According to storeseo.com, fetched 18 November 2026, it writes meta titles and descriptions in bulk, builds JSON-LD product schema, and generates llms.txt from the store's products, collections and pages. Other SEO apps on the Shopify App Store cover meta tags and schema too.

No platform detected: where most of the gaps are

Hosts with no detected platform are a mix of hand-built sites, static generators the scanner does not fingerprint, and custom stacks. Even after removing the 62 empty shells, they pass C3 only 39.7% of the time. 135 of the 257 rendered pages in this group have no JSON-LD, and 69 have no meta description. If you built your site by hand, this is the group you are most likely in.

Our own four sites, checked the same morning

We ran the same four parts against our own homepages with curl on 18 November 2026, using a browser user-agent, and counted the elements in the raw HTML.

Our siteh1titlemeta descriptionJSON-LD blocksC3
aiscan.site1yesyes1pass
xspeedcache.com1yesyes1pass
storeseo.com1yesyes4pass
bettershield.ai1yesyes1pass

The storeseo.com scan through AIScan's public API returned h1: 1, title: true, meta description: true, JSON-LD: true for C3 and 3989 words of server-rendered text, 16 <h2>, 58 <h3> for E3 the same morning. Passing C3 is a low bar, and these numbers show it: a page with four parts in place can still have a JSON-LD block that says very little.

How to check your homepage

The quickest check is AIScan itself. Run:

npx aiscan-cli yoursite.com

or paste your URL at aiscan.site. The report lists C3 with its evidence string, so you see the h1 count and the three yes/no values directly, next to E3 (server-rendered text and headings) and the rest of the content checks. It is free and needs no account.

If you would rather check by hand, these four commands count the same things C3 counts in the HTML your server returns:

URL=https://yoursite.com/
curl -sL "$URL" | grep -o '<h1' | wc -l                               # want 1
curl -sL "$URL" | grep -c '<title'                                    # want 1 or more
curl -sL "$URL" | grep -o '<meta name="description"' | wc -l          # want 1
curl -sL "$URL" | grep -o 'application/ld+json' | wc -l               # want 1 or more

grep on raw HTML is rough. It misses attributes written in a different order (<meta content="..." name="description">) and counts <h1 inside comments. AIScan parses the document, so where the two disagree, trust the scan.

Fixing each part

Part (hosts in this corpus)Common causeFix
No JSON-LD (238)Nothing in the stack writes it by default (135 hosts with no platform detected, 46 Next.js)Add one Organization or WebSite block sitewide; see JSON-LD schema types for AI
No h1 (159)Site name shown as a logo or styled text; hero heading written as h2Mark the main page title as the one h1
2+ h1 (48)Logo in an h1 plus a hero h1; page builders adding h1 per sectionKeep one; demote the rest to h2. See multiple h1 tags fix
No meta description (117)No SEO plugin or head template writes it (36 WordPress, 69 no platform detected)Write one sentence, roughly 150 characters, that says what the site is
No title (12)Single-page apps that set the title only in JavaScriptRender the <title> on the server
All four (63)Empty client-rendered shellServer-render or pre-render the page first; see server-rendered text and headings

A minimal JSON-LD block that turns the 74 "JSON-LD only" partials into a pass looks like this:

<script type="application/ld+json">
{"@context":"https://schema.org","@type":"Organization",
 "name":"Example Co","url":"https://example.com/",
 "logo":"https://example.com/logo.png"}
</script>

According to Google's introduction to structured data, verified on 18 November 2026, structured data is "a standardized format for providing information about a page and classifying the page content"; the same page explains which types Google reads, and the Schema.org validator checks the syntax. Remember the limit above: C3 only asks whether a block exists. A good block names the organisation, its URL and its logo at a minimum.

Common mistakes we saw in the evidence

  • Writing the hero heading as an h2. It is one way a page ends up like the 58 that have h2 sections and no h1. The visual size does not matter to a parser; the tag does.
  • Fixing two h1 elements by deleting both. The audit warning goes away and the page joins the larger group with none. Demote the extras, keep one.
  • Assuming the SEO plugin covers the homepage. Check the front page itself after installing one: when the homepage is a static page, its meta description usually comes from that page's own SEO fields, and they start empty.
  • Adding JSON-LD with JavaScript after load. C3 reads the server's HTML. A block injected by a tag manager after the page loads is invisible to it, and to most AI crawlers.
  • Treating a C3 pass as "schema done". C3 counts a block's presence. It does not check the @type, the required properties, or whether the data is true.

What to do next

Run npx aiscan-cli yoursite.com or scan at aiscan.site, read the C3 evidence string, and fix the part it names. If it says h1: 0, look at E3 in the same report: if E3 passes, you need one tag change; if E3 fails, start with server rendering. Then work through the rest of the content dimension (C1 Markdown negotiation, C2 llms.txt, E3 headings, E5 feed) in the step-by-step AIScan guides.

Frequently asked questions

AIScan says C3 is partial with "h1: 0". What does that mean?

Your homepage HTML, as the server returns it, contains no h1 element. Check E3 in the same report: if E3 passes, the page has text and headings and you only need to mark the main title as the one h1. If E3 also fails, the page is probably rendered in the browser, and server rendering comes first. In our 18 November 2026 corpus, 159 of 792 rendered homepages had no h1.

My C3 result says "JSON-LD: false" even though I installed a schema plugin. Why?

C3 reads the HTML your server sends. If the plugin is switched off for the front page, or the JSON-LD is injected by a tag manager after load, the block is not in that HTML. View the page source (not the inspector) and search for application/ld+json. JSON-LD was the most common missing part in our corpus: 238 of 792 rendered pages.

I get "h1: 2" on C3 after adding a hero heading. How do I fix it?

Keep one h1 for the page's main title and change the other to an h2. The usual cause is a logo or site name wrapped in an h1 in the header plus an h1 in the hero. Do not delete both: a page with no h1 is also partial, and in our data zero h1 was more than three times as common as two or more.

Does the HTML standard require exactly one h1?

No. The WHATWG HTML Living Standard says that if a document has headings, at least one should have a heading level of 1. C3 is stricter and asks for exactly one, because an agent extracting a page title has to choose a heading, and a single h1 removes the guess.

Does C3 check whether my JSON-LD is valid?

No. C3 checks that at least one application/ld+json block exists. It does not check the @type, required properties or syntax beyond that. Use the Schema.org validator for syntax and Google's structured data documentation for which types it reads.

My C3 fails with every part false. Is my HTML broken?

Not always. All 63 C3 fails in our corpus had all four parts false, and 60 also failed E3. That is either a page that builds itself with JavaScript and sends near-empty HTML, or a site that answered the scanner with a bot challenge or error page. Open the raw HTML: an empty body needs server rendering; a challenge page means your firewall is turning crawlers away.

Which platforms pass C3 most often?

In our 18 November 2026 corpus: Astro 41 of 52 (78.8%), Next.js 152 of 208 (73.1%), WordPress 129 of 211 (61.1%). Hosts with no detected platform passed 102 of 319 (32.0%). WordPress pages mostly miss the h1; Next.js pages mostly miss JSON-LD.

How do I check C3 on my own site?

Run npx aiscan-cli yoursite.com or paste your URL at aiscan.site. It is free and needs no account. The C3 evidence string shows the h1 count and whether the title, meta description and JSON-LD were found, next to the other content checks.

M Asif Rahman

About the author

Entrepreneur, curious about AI

I have been building companies for a long time, and lately most of my curiosity goes to AI. AIScan is a small personal project I built on a flight from Dubai to Narita. I use it to understand how ready the web is for AI agents, and I share what the scans teach me.

Related guides

Browse every guide by check and platform