Schema to Avoid

“Validates in the Rich Results Test” and “appears in official documentation” do not mean “produces visible output for you.” The types and files below are common good-faith implementations that spend engineering time for no return — because support was deprecated, because eligibility is gated behind accreditation, because no rich-result format exists, or because the evidence for the effort doesn’t hold up. None of them are wrong to implement in the sense of causing a penalty; the cost is time, not risk.

Deprecated — Google removed support

These types still validate. The Rich Results Test does not flag them as errors — the deprecation was announced separately from the validation tooling.

SearchAction / SiteLinksSearchBox

Google announced the retirement of the Sitelinks Searchbox feature on October 21, 2024, and completed removal from Search results on November 21, 2024. The SearchAction type and potentialAction property remain valid in the Schema.org spec, and WebSite structured data itself is still active and recommended for site-name determination — only the sitelinks-searchbox consumption of potentialAction on WebSite was removed. As of March 2026, no major search engine publicly documents any result feature driven by this markup. The Google-specific query-input: "required name=search_term_string" annotation was never part of the Schema.org specification; it now does nothing.

A significant number of sites — including well-maintained CMS installs running popular SEO plugins — still ship this block on every page load, usually because it arrived via a default template years ago and was never revisited. It causes no penalty, but it creates false confidence in structured-data audits and adds stale, non-standard annotations (query-input specifically) that some strict validators flag as warnings.

// Search for this pattern in WebSite JSON-LD and remove the potentialAction block —
// it does nothing in any current search feature.
{
  "@type": "WebSite",
  "name": "...",
  "url": "...",
  "potentialAction": {
    "@type": "SearchAction",
    "target": "https://example.com/search?q={search_term_string}",
    "query-input": "required name=search_term_string"
  }
}

Keep the WebSite block with name, url, and optionally alternateName — those properties are still active. Remove potentialAction entirely; do not add it to new implementations.

SpecialAnnouncement

Created in March 2020 for COVID-19 health alerts and emergency notices. Google stopped processing it for rich results after the pandemic period. The type and Google’s original documentation page remain; the feature is inactive, and no current type substitutes for it.

Speakable

A Google Assistant feature for designating text extracts for voice readout in Google News. The Assistant integration that powered it has been discontinued; Google Search does not process Speakable for any current rich result or voice feature.

Gated — requires accreditation

These types produce rich results, but only for a narrow set of credentialed publishers. Valid markup without credentials is ignored.

ClaimReview

Triggers fact-check labels in Google Search and Google News. Eligibility requires recognition as a fact-checking organization through the International Fact-Checking Network (IFCN) or equivalent. Implementing ClaimReview on non-fact-checking content — reviews, opinion pieces, editorial corrections — may attract spam-policy review as a misuse of the type.

Medical schema (MedicalCondition, Drug, Physician)

Google documents support for the medical vocabulary, but rich results in practice require a health-authority context general publishers don’t meet — the effective implementation threshold is major health reference publishers (WebMD, Mayo Clinic, NHS).

Misapplied — no format, or wrong context

Q&APage

Google scaled back Q&A rich results in September 2023, in the same announcement that reduced FAQPage display. Current eligibility follows the same trajectory: restricted to high-authority domains in specific categories. Treat it as technically implementable, not worth the effort for most publishers.

WebPage (bare type)

The base class for all page types in the Schema.org hierarchy. It produces no rich result and no documented crawling, entity, or indexing benefit. If the page is an Article, use Article; if it’s a Product, use Product. Bare WebPage as a fallback is markup noise.

LocalBusiness on non-local pages

LocalBusiness is correct for businesses with a physical location serving local customers. Applying it to SaaS products, national publishers, or ecommerce operations without physical retail is a category error — no ranking benefit, and it may miscategorize the entity. Use Organization instead.

llms.txt — a skeptical verdict

A June 2026 Ahrefs study of 137,210 domains found 97% of published llms.txt files received zero requests from anything — no AI bot ever probed for the file on sites that lacked one, and Slackbot fetched it more often than PerplexityBot did. Google has publicly declined to support it: John Mueller compared it to the keywords meta tag — self-declared, unverifiable, and therefore ignored — and Gary Illyes confirmed Google has no plans to consume it.

The spec is also structurally too thin for an enterprise, multi-market catalog: a flat Markdown file has no market mechanism, no hreflang analog, and no reliable way to represent per-market pricing, availability, or launch timing. Hand-curating it doesn’t scale across a real product portfolio; auto-generating it risks reproducing data better handled by feeds, sitemaps, and structured data that engines actually consume.

The one narrow, evidenced use case: AI coding tools (IDE assistants, CLI agents) do fetch /llms.txt on developer-documentation sites as a token-saving shortcut. Mueller frames this as “not done for search” — a parsing crutch for docs, not a discovery mechanism.

What to do: verify your own server logs before spending anything on this — let your data, not advocacy, set the priority. If you ship an SDK, API, or developer program, an auto-generated llms.txt scoped to the documentation property alone is near-zero cost and serves that one real audience. Do not build a sitewide or per-market llms.txt catalog; redirect any planned effort there into product feeds, JSON-LD, and IndexNow instead.

Notes

• Check the Google Search Central Blog and Search Gallery before implementing any schema type that saw prominence years ago — deprecation announcements are not always prominently flagged in documentation.
• “Validates in Rich Results Test” confirms markup syntax only, not eligibility for display. The Search Gallery is the canonical list of types Google supports for rich results — if a type isn’t listed there, there is no supported rich result for it.
• Non-rich-result schema blocks do no ranking damage — the cost is implementation and maintenance time, not an algorithmic penalty.
• FAQPage remains partially active with narrowed eligibility (see FAQPage Schema); HowTo rich results are fully deprecated on all devices as of September 2023.