How this is computed, and what it can't tell you

A site that argues for honest structured data has to show its own working.

The source

The Schema.org + Google usage statistics dataset, published monthly in the public schemaorg/schemaorg repository. Each file pairs a Schema.org term with the order-of-magnitude range of domains Google's crawl observed it on. The archive currently covers May 2026.

The pipeline

Freshness is a property of the rebuild. Nothing is fetched in your browser; the entire computation is public.

  1. 1

    Heartbeat

    A daily job checks the upstream dataset directory; a new YYYY_MM.csv is committed to /data, append-only.

  2. 2

    Rebuild

    That commit triggers a static rebuild; every archived month is recomputed at build time.

  3. 3

    Archive

    Each month accumulates in the repo, the longitudinal series a late starter can't backfill.

The shape of the data

Structured data on the web is a long tail: most terms are barely evidenced, a dozen are everywhere. Counts are exact; the bands they sit in are order-of-magnitude.

Deployment spectrum

Log-scaled bars (exact counts labeled), fitting, since the bands themselves are orders of magnitude. The long tail is the point: a handful of terms are everywhere, most are barely evidenced. Click the top three bands to list the actual terms.

Distribution of terms per deployment band 5545 terms · 2026-05
Band In words Types / 958 Properties / 4587
10M+ ubiquitous 12 31
1M - 10M widely deployed 35 65
100K - 1M common 39 119
10K - 100K occasional 151 269
1K - 10K rare 236 324
< 1K negligible 485 3779
All bands total 958 4587

Read down a column: half of all types sit in the rarest band, a dozen reach the top. Counts are exact; the bands they sit in are order-of-magnitude.

Three states, never collapsed

The credibility error this whole site avoids is treating these as one question. Whether a Schema.org term is deprecated, whether a Google rich result still exists, and whether the markup still carries comprehension value are three separate things.

Term Schema.org vocabulary Google rich result Comprehension value Verdict
FAQPage source ↗ source ↗ Still a valid type, not deprecated Removed from Search May 7, 2026. SC reporting and Rich Results Test support dropping June 2026; SC API Aug 2026 Not documented. Google states only that unused structured data has no visible effect in Search; any residual value to comprehension or AI/GEO retrieval is an inference, not a platform claim Migrated, not dead
SearchAction source ↗ Valid type, not deprecated in the vocabulary No current Google SERP feature No specific downstream use evidenced; treat as inert unless you have a non-Google consumer Retired plumbing
HowTo source ↗ Valid types, not deprecated No SERP feature No residual platform use evidenced Zombie
WPHeader source ↗ Valid WebPageElement subtypes Never a rich result Structural; harmless, low-signal Legacy noise
Every dated claim is checked against Google and Schema.org primary sources, linked per row. Last verified Jun 2026. Where Google documents nothing, the table says so rather than inferring a platform claim.

Caveats that keep the read honest

Buckets are ranges, not counts.
Ordinal language only, widely deployed, common, rare, not evidenced, and never a precise ranking within a band.
A sampled, rendered crawl, not a census.
Google's public crawl is sampled and applies per-URL size limits, so schema buried deep in very large pages, or injected late by JavaScript, may be undercounted. A map of deployment, not a recommendation.
Three states, never collapsed.
Vocabulary status, rich-result status, and comprehension value are answered separately everywhere.

Claims we refuse to make

No specific FAQ→AI-Overview multiplier (e.g. "3.2x"), traces to a single unverified vendor blog. Cite direction only.
No calling these "deprecated schema" without the qualifier "rich result".

Where the data is clear we are blunt. Where it isn't, we say so. The numbers are the raw material; the judgment about what they mean is the work.

Request an audit

Have structured data you're not sure about? Send a URL and get a keep / cut / re-aim read on what it actually ships, the same call the verdicts above make.