Google's August 2023 FAQ rich-result rollback (and the periodic "schema is dead" cycle that follows every Google deprecation) keeps generating the same takes. Schema is not dead. Two-thirds of the noise mistook Google's rich result removal for all engines no longer using schema. Those are different facts.
This is the 2026 hierarchy: what each AI engine actually reads, what survived the deprecations, and the under-used signal that pays off because almost nobody ships it.
TL;DR — read these five lines
- Google has deprecated rich-result coverage for FAQPage (Aug 2023), HowTo (Sep 2023 desktop, fully retired), Practice Problem, Dataset, SpecialAnnouncement, and Q&A. The schema types are still valid and AI engines still benefit. Don't strip the markup.
- In direct-fetch retrieval, ChatGPT / Claude / Gemini / Perplexity / Google AI Mode treat JSON-LD as plain text in the page — they do not run a dedicated parser. Schema benefits indirectly via Google's index, Knowledge Graph, and Bing Copilot.
- The single highest-leverage 2026 play is
Organization+Personwith richsameAschains to Wikidata — drives entity verification that AI Overviews and AI Mode now rank-condition on. - Empty schemas hurt. Sites with minimal schema get cited less than sites with no schema. Fully populate before adding more.
Speakableis the most under-used citation lever — sub-1% adoption, direct verbatim-quote signal for Perplexity + ChatGPT browse.
The 2026 priority order
If you've never touched schema on this property, ship in this order:
1. Organization + Wikidata sameAs
This is the substrate. Every other piece of schema gets cross-referenced through it.
{
"@context": "https://schema.org",
"@type": "Organization",
"name": "Your Brand",
"url": "https://yourbrand.com",
"logo": "https://yourbrand.com/logo.png",
"sameAs": [
"https://www.wikidata.org/wiki/Q12345678",
"https://en.wikipedia.org/wiki/Your_Brand",
"https://www.crunchbase.com/organization/your-brand",
"https://www.linkedin.com/company/your-brand",
"https://twitter.com/yourbrand"
]
}
The Wikidata QID is the keystone — it's what AI engines use to resolve "are these two Your Brand mentions the same entity?" Without it, your authority signal fragments across name variants.
2. Person schema for every author
Claude cites UGC and named-author content at 2-4× the rate of other engines. AI Mode rank-conditions on author E-E-A-T. Generic "by Editorial Team" lines lose to named, schema-linked authors.
{
"@type": "Person",
"name": "Jane Smith",
"jobTitle": "Senior Engineer",
"url": "https://yourbrand.com/team/jane-smith",
"sameAs": [
"https://www.linkedin.com/in/janesmith",
"https://orcid.org/0000-0001-2345-6789",
"https://github.com/janesmith"
]
}
ORCID + LinkedIn + a GitHub or industry directory closes the entity loop.
3. Article (or BlogPosting / NewsArticle) on every long-form page
{
"@type": "Article",
"headline": "...",
"author": { "@type": "Person", "name": "Jane Smith", "url": "..." },
"publisher": { "@type": "Organization", "name": "Your Brand", "logo": "..." },
"datePublished": "2026-05-08",
"dateModified": "2026-05-08",
"mainEntityOfPage": "https://yourbrand.com/blog/your-post"
}
dateModified is non-negotiable in 2026 — freshness is a documented ranking factor across AI Overviews and AI Mode (see Ahrefs Q1 2026 study linked in Sources). Stale dateModified reads as abandoned content.
4. Product / Service with aggregateRating (only if real)
Critical caveat: review-richness must match the HTML. Schema-HTML mismatches now actively demote you — our audit's detectHtmlMismatches() catches this. Claiming aggregateRating: 4.9 with no visible reviews on page is worse than no aggregateRating at all.
5. Speakable — the underused one
Speakable marks the passages a voice assistant should read aloud verbatim. Sub-1% of sites use it. Perplexity and ChatGPT browse both prefer Speakable-marked passages when the user query has a "summary" intent.
{
"@type": "Article",
"speakable": {
"@type": "SpeakableSpecification",
"cssSelector": [".article-summary", ".key-takeaway"]
}
}
Add it to your hero paragraph + TL;DR sections. Zero cost, real upside.
6. FAQPage / HowTo — still ship them
Despite the rich-result removals, the schema types are still valid and AI engines still extract Q&A structures from them. Just drop any "this will give you rich results in Google search" framing from your internal docs. They're for AI consumption now.
What sub-1% of sites get right
Sample disclosure: numbers below are from our own audit pipeline (~3,000 sites/week), biased toward English-speaking B2B SaaS, marketing-led DTC, and SMB local-services sites — not a random web sample. Treat as directional, not authoritative. Field data, May 2026:
| Schema type | % of sites with any | % with rich-enough to count |
|---|---|---|
Organization | 71% | 38% (full sameAs chain) |
WebSite | 64% | 51% |
BreadcrumbList | 42% | 35% |
Article / BlogPosting | 38% | 22% (with author entity) |
Product | 28% | 19% (with valid aggregateRating) |
FAQPage | 19% | 17% |
Organization.sameAs → Wikidata QID | 5% | — |
Speakable | <1% | — |
Person schema for authors | 6% | 3% (with sameAs) |
The two lowest-adoption rows — Wikidata QID and Person — are the two that AI engines weight most heavily for entity authority. That's the arbitrage.
Schema-HTML mismatch — the silent demotion
Three checks our audit added in May 2026, based on field evidence that Google + Perplexity are penalizing them:
Product.price≠ on-page price — common when CMS updates the page but not the JSON-LD.AggregateRating.reviewCount≠ visible review count — common when imported from a review widget that's been removed.JobPosting.baseSalary≠ visible salary range — common in late edits.
If you mark up rich data, audit it monthly. The cost of stale schema is bigger than the cost of no schema.
The bottom line
Schema in 2026 is no longer about rich results. It's about entity verification — proving you are who you say you are, in a format AI engines and their upstream indexes can cross-reference. The Wikidata QID + author Person entity chain is the highest-leverage move you can make in a single afternoon.
Sources
- Google deprecation timeline — https://developers.google.com/search/docs/appearance/structured-data
- Speakable spec — https://schema.org/Speakable
- Wikidata QIDs explained — https://www.wikidata.org/wiki/Help:Items
- AIO Knowledge Graph ranking signals (Ahrefs Q1 2026) — https://ahrefs.com/blog/ai-overviews-vs-ai-mode/
Want to know which schema types you have, which are mis-matched, and which are missing? Run our free audit — every JSON-LD block on every crawled page, scored for AI consumption.