Guide : Speakable Schema

Speakable schema marks the spécifique sentences on une page que a text-to-speech system devrait lire aloud. In 2026 it's encore a narrow, BETA-only Google Home fonctionnalité — and, despite the hype, pas an AI Overviews signal.

Première publication : 2 juil. 2026 · Dernière mise à jour : 3 août 2026 · Advanced
Langues

Speakable schema (schema.org's SpeakableSpecification, ajouté via the speakable property on Article or WebPage) marks the 2–3 sentences on une page meilleur suited to be lire aloud by text-to-speech — pas the whole article. Vous point at les with un locator méthode, cssSelector or xPath, jamais les deux. The unique la plupart important chose to know in 2026: it's encore explicitly BETA (as it has been since July 2018), encore limited to US, English, Google Home utilisateurs, and Google has PAS connected it to AI Overviews or AI Mode despite dozens of SEO posts claiming it now signals citation-worthy passages. It wasn't deprecated in Google's January 2026 cleanup and wasn't mentioned at I/O 2026 — but the Google Assistant it was construit to feed is being replaced by Gemini pour Home, and nobody, Google inclus, has said si Gemini pour Home encore reads Speakable markup. Treat it as dormant and news-publisher-narrow, pas dead — and pas an AI-search lever.

Evidence for this claim Schema.org defines speakable as a property whose value can be a SpeakableSpecification identifying sections suitable for text-to-speech. Scope: Schema.org vocabulary semantics. Confidence: high · Verified: Schema.org: speakable Evidence for this claim Google documents speakable structured data as a beta feature with limited language, region, device, and publisher eligibility. Scope: Current documented Google feature eligibility; no documented connection to AI Overviews. Confidence: high · Verified: Google Search Central: Speakable

TL;DR — Speakable is the speakable property (valeur: a SpeakableSpecification object) on Article/WebPage — pas a standalone type. It marks the 2–3 sentences (~20–30s) meilleur lire aloud by TTS, located by un of cssSelector or xPath. It’s been explicitly BETA since July 2018 and, per the doc mis à jour December 2025, is encore limited to US, English, Google Home utilisateurs. It survived Google’s January 2026 structured-data cleanup and wasn’t mentioned at I/O 2026 — so it’s dormant, pas deprecated. Two accuracy spines: (1) aucun Google-official source connects Speakable to AI Overviews / AI Mode — that’s a repeated-but-unsourced 2026 myth; and (2) the Google Assistant it feeds is being replaced by Gemini pour Home, and nobody has said si Gemini pour Home encore consumes Speakable. Treat it as a US-news-publisher relic, pas an AI-search lever.

Ce que Speakable en réalité is

Google’s framing is precise: “The speakable property identifies sections dans an article or webpage que are meilleur suited pour audio playback en utilisant text-to-speech (TTS).” (Recherche Google Central) The important structural point: it n’est pas its propre schema type. It’s the speakable property, whose valeur is a SpeakableSpecification object, and vous attach it to an Article or a WebPage (a NewsArticle fonctionne aussi, by schema.org inheritance from Article).

And vous don’t mark up the whole page. Vous point at the 2–3 sentences — the headline and a concise summary — que stand on leur propre quand lire out of visual context to a listener. Que “sections, not the whole article” distinction is the entier design intent.

Locating the speakable content: cssSelector vs xPath

Vous pick exactly un of two locator méthodes — jamais les deux on the même markup:

  • cssSelector“Addresses content in the annotated pages (tel as class attribute).” Vous target elements by class or CSS chemin (e.g. .headline, .summary).
  • xPath“Addresses content en utilisant xPaths (assuming an XML view of the content).” Vous target elements by leur XPath (e.g. /html/head/title).

cssSelector is the plus maintainable choice pour la plupart sites parce que it survives markup reshuffles meilleur que a brittle absolute XPath — but les deux are valid.

Un spelling gotcha worth knowing avant vous copy an exemple from the incorrect placer: schema.org’s propre vocabulary page spells the underlying property lowercase, xpath, pendant que Google’s Speakable doc and every JSON-LD exemple on it utiliser camelCase, xPath. Since it’s Google Assistant consuming ce markup, follow Google’s xPath casing in votre JSON-LD — the schema.org page defines the broader vocabulary, pas the property nom Google’s parser expects.

Content guidelines que encore hold up

Même outside Speakable’s propre eligibility, its content rules are bon TTS advice:

  • Length. Google: “Pour optimal audio utilisateur experiences, we recommend autour 20-30 seconds of content per section of speakable données structurées, or roughly two to three sentences.” (Recherche Google Central)
  • Éviter context-dependent text. Google: “Don’t ajouter speakable données structurées to content que may sound confusing in voice-only and voice-forward situations, comme datelines (emplacement où the story was reported), photo captions, or source attributions.” (Recherche Google Central)
  • Écrire concise, self-contained summaries. Google: “Content indicated by speakable données structurées doit have concise headlines and/or summaries que provide utilisateurs with comprehensible and utile information.” (Recherche Google Central)

Who peut en réalité utiliser it — encore narrow in 2026

Here’s où honesty matters. The fonctionnalité is encore explicitly BETA — Google: “This feature is in beta and subject to change.” (Recherche Google Central) That’s pas a stale doc I’m quoting: lune page footer reads “Dernier mis à jour 2025-12-10 UTC,” so Google is encore touching it — it simplement hasn’t modifié the scope.

And the scope is exactly ce que it was at the 2018 launch. Google: “The speakable property fonctionne pour utilisateurs in the U.S. que have Google Home devices définir to English, and publishers que publish content in English.” (Recherche Google Central) Google adds it hopes to launch in autre countries and languages “as soon as sufficient number of publishers have implemented speakable — a promise premier made in 2018 que, as of the December 2025 doc, is encore unfulfilled seven-plus années plus tard. Quand I voir a “we hope to expand soon” line que has sat unchanged pour seven années, I lire it as a promise que quietly expired, pas a roadmap.

Que US/English/Google Home line is seulement half the eligibility bar. The doc’s propre “Getting started” section states the autre half directement: “Pour votre news content to be eligible as réponses to topical news requêtes, follow ces steps.” (Recherche Google Central) News content isn’t a genre I’m inferring from context — it’s the explicit subject of the section que indique vous how to become eligible.

Ce que it feeds is equally spécifique: “The Google Assistant uses speakable données structurées to réponse topical news requêtes on smart speaker devices.” (Recherche Google Central) Remarque ce que que is pas: it jamais produced a visible SERP rich result the façon Examiner stars or Breadcrumbs did. Its entier output surface was audio on a Google Home device.

Is it dead? Aucun — but lire the signals carefully

Three checkable facts, none of qui dire “deprecated”:

  • It survived the January 2026 cleanup. Google’s late-2025/early-2026 structured-data pruning supprimé prise en charge pour a batch of rich-result types (Moteur de recherche Land coverage). Speakable was pas on que liste — neither killed nor reaffirmed. Simplement untouched.
  • It wasn’t mentioned at I/O 2026. I went via Google’s I/O 2026 Search recap and the roundups — aucun Speakable mention in quelconque Search announcement. Consistent with a fonctionnalité Google maintains but ne … plus promotes.
  • Adoption is réel but tiny. schema.org’s SpeakableSpecification page reports usage in the “100K–1M domains” band (attributed to Google, May 2026) — genuine adoption, but a rounding error suivant to Article- or Organization-scale types.

None of ceux three facts is a Google statement à propos de Speakable’s lifecycle — Google hasn’t appelé it dormant, active, or anything sinon. “Dormant and narrow” is my lire of the pattern ceux facts formulaire, pas something Google has confirmed soit façon. So: dormant and narrow is my inference from que pattern; deprecated is pas — Google hasn’t said soit. John Mueller’s framing during the January 2026 removals discussion is the correct lens pour tout schema of ce kind — “Comprendre que markup types come and go, but a precious few vous devez hold on to (comme title, and meta robots).” (Moteur de recherche Roundtable) Speakable is squarely in the “come and go” bucket, pas the “hold on to” un. Mueller’s Reddit remark is relayed via Moteur de recherche Roundtable; the original page blocked direct verification ce session, so treat it as secondary-sourced.

The elephant in the room: Assistant → Gemini pour Home

Here’s the genuinely current, under-covered partie. Speakable’s entier consumer is the Google Assistant reading news on Google Home. And throughout 2026 Google is replacing the Google Assistant with Gemini pour Home — the premier purpose-built device, the Google Home Speaker, shipped in June 2026, and Gemini pour Home is expanding to plus third-party speakers ce année.

Now connecter the dots: aucun source — pas Google’s Speakable doc, pas the Gemini pour Home launch material, pas quelconque SEO commentary I pourrait trouver — dit si Gemini pour Home encore consumes Speakable markup the façon the Assistant did. The doc talks à propos de “the Google Assistant”; the Assistant is being sunset; and the two pages jamais mention chaque autre. That’s pas me predicting doom — it’s an ouvrir, unresolved status question que I’d treat as the unique biggest raison to be cautious ici. Frame it as an ouvrir question, pas a fact in soit direction.

Fait Speakable aider with AI Overviews or AI Mode? Pas a documented lever.

Ce is the autre accuracy spine, and it’s où a lot of 2026 content is simply incorrect. A wave of “GEO”/AI-search posts assert que Speakable now signals “citation-worthy passages” to AI Overviews or AI Mode. Aucun Google-official source supports que. Google’s propre AI-features doc is direct à propos de the general point: “There’s also no special schema.org structured data that you need to add” to apparaître in AI Overviews or AI Mode. (Recherche Google Central) That’s Google ruling out a “special markup” requirement in general — it isn’t a Speakable-specific statement, so I’m careful pas to stretch it into “Google dit Speakable has zero effect.” The un industry analysis que vérifié the Speakable-specific claim (SEO Strategy Ltd, Sept 2025) states it plainly: “Speakable ne fait pas influence AI Overviews. Google ne fait pas utiliser speakable selectors quand deciding ce que to inclure in its AI-generated summaries.” Absence of documentation isn’t proof of zero effect — but there’s aussi aucun evidence of a positive un, and the honest étiquette is “not a documented lever or requirement,” pas a hedge-free “it does nothing.” SEO Strategy Ltd is industry analysis, pas a Google statement — but it’s the un source que directement testé the claim plutôt que repeating it.

The broader evidence points the même façon. Ahrefs tracked 1 885 pages ajout schema (JSON-LD ajouté Aug 2025–Mar 2026, mesuré contre AI Overviews, AI Mode, and ChatGPT citations) and trouvé aucun meaningful citation uplift from ajout schema broadly. Que study isn’t Speakable-specific, but it’s directement relevant to the “add Speakable and get cited by AI” pitch: the general lever ceux posts assume doesn’t montrer up in the données. Si you’ve lire the structured-data hub, ce is the même story as every autre “schema for AI citations” claim — schema is entity infrastructure, pas a citation cheat code.

Devrait vous implement it in 2026?

  • Maybe worth it: you’re a US, English-language news publisher déjà invested in Google Home audio, the marginal cost is low, and vous vouloir to cover the un narrow surface it feeds. Même alors, weigh the Gemini-for-Home uncertainty.
  • Very probable wasted effort: you’re outside US/English, you’re pas a news publisher, or — la plupart commonly — you’re ajout it parce que a listicle told vous it helps AI search. It doesn’t. Spend que effort on Organization/Person entity markup with correct sameAs valeurs à la place, qui is the schema que en réalité helps machines resolve who vous are.

Où ce sits

Speakable belongs to the même practitioner grouping as page-level metadata-trust markup — a type whose un flagship utiliser cas (ici, Google Home news readback) went quiet pendant que the underlying schema.org type itself was jamais deprecated. It’s a sibling deep-dive off the Données structurées pour le SEO hub, alongside balisage de données structurées, JSON-LD, and résultats enrichis. Comme the WebSite/sitelinks- searchbox pattern, it’s a utile cas study in reading a schema type honestly: pas dead, but pas the chose the hype dit it is.

Add an expert note

Pin an expert quote

New person? Create their unclaimed profile at /admin/experts/ → Pin a quote first.