(Technical)

Speakable structured data

A Property That Tells Voice Assistants “Read This Part”

Speakable is a schema.org property that marks the sections of an article or web page best suited to be read aloud with text-to-speech (TTS, technology that turns text into sound). Google (official) With this markup, search engines and other apps can identify the content to read aloud on devices with Google Assistant. Google (official) It’s a lot like highlighting just two or three sentences in a newspaper article and saying, “Just read this part.”

schema.org defines it the same way: it indicates sections of a web page that are particularly highlighted as suitable for reading aloud. Other sections may also be read depending on the situation, but the property is meant to mark the parts that will generally be most useful. schema.org The one actually using it is Google Assistant. When a user asks a smart speaker for the latest news on a topic, Assistant picks up to three articles from the web, reads the sections marked speakable, attributes the source, and sends the full article URL to the Assistant app on the user’s phone. Google (official)

There Are Only Two Ways to Mark It: Selectors

Attach the property to an Article or WebPage type, and point to the sections to read with either a CSS selector (cssSelector) or an XPath expression (xPath). You don’t use both together. Google (official) The schema.org definition includes a third method that references an element’s id by URL, but Google’s documentation lists only the first two as supported. schema.org The example in Google’s documentation uses XPath to point to the page title and meta description. Google (official)

{
  "@context": "https://schema.org/",
  "@type": "WebPage",
  "name": "Quick Brown Fox",
  "speakable": {
    "@type": "SpeakableSpecification",
    "xPath": [
      "/html/head/title",
      "/html/head/meta[@name='description']/@content"
    ]
  },
  "url": "https://www.example.com/quick-brown-fox"
}

There’s one difference in spelling. The official schema.org property name is lowercase xpath, but Google’s documentation writes xPath in both its example and its property table. schema.org If you’re targeting Google Assistant, it’s safer to follow Google’s spelling. Google’s documentation also gives cssSelector examples that point to classes, such as .headline and .summary. Google (official) In practice, adding classes to the headline and summary paragraph and pointing to them this way is easier to manage than XPath.

Google’s Requirements for Read-Aloud Sections

Google’s guidelines on what to mark are short and specific.

  • The marked sections should be concise headlines or summaries that give listeners useful, understandable information. Google (official)
  • Don’t mark up the whole article; focus on the key points so the reading isn’t cut off midway and listeners don’t miss important content. Google (official)
  • If you include the top of the article, Google recommends rewriting the information into separate sentences so it reads clearly as speech. Google (official)
  • It recommends 20 to 30 seconds per section, roughly two or three sentences. Google (official)
  • Don’t mark parts that are confusing when only heard, such as datelines, photo captions and source attributions. Google (official)

In Beta from 2018 to Today

The vocabulary for this property predates Google’s announcement. schema.org added the SpeakableSpecification type, the cssSelector and xpath properties, and WebPage’s speakable property as newly proposed terms in version 3.2 in March 2017, and promoted them to the official core vocabulary in version 3.5 in April 2019. schema.org On July 24, 2018, Google announced the speakable specification, developed with schema.org, on the Search Central Blog, explaining that it lets eligible publishers mark the sections of news articles most suitable for Assistant to read aloud. Google (official) It opened first to English-speaking users in the US, and said it would expand to other languages and countries once enough publishers had implemented it. Google (official)

The documentation at the time, as quoted by an industry outlet the day after the announcement, carried a notice that the feature was in beta and its requirements could change. Search Engine Roundtable The current documentation, updated in September 2026, still has the same notice and the (BETA) label. Google (official)

There has been sporadic news since then. In December 2019, Danny Sullivan said speakable was no longer limited to news content, but added that putting this markup on any site doesn’t guarantee Assistant will always use it, and that it was still in beta. Search Engine Roundtable In September 2020, after an email from a Google team was made public and a rumor spread that it had been deprecated, John Mueller corrected it, saying it was not deprecated and the markup was still useful. Search Engine Roundtable When Google announced in June 2025 that it was ending support in search results for seven types of structured data, speakable was not on the list. Google (official) It is still listed as a news item in Google’s structured data feature gallery. Google (official)

But being alive on paper is not the same as being used. Despite what Sullivan said in 2019, the current documentation’s getting-started guidance and guidelines are still written around news content and news results. Google (official) schema.org shows usage, counted from Google’s index, only as a range of between 100,000 and 1 million domains. schema.org Ahrefs’ voice search guide never mentions speakable. Ahrefs Neither does Semrush’s Google voice search guide. Semrush Even within the industry, it’s a fringe property.

It Doesn’t Work on Korean-Language Sites

The most important fact is its scope. The country and language section of Google’s documentation states firmly that the speakable property works for users in the US who have Google Home devices set to English, and for publishers that publish content in English. Google (official) The condition for expanding to other countries and languages is that enough publishers implement it, and this wording is virtually the same in the 2018 announcement and the 2026 documentation. Google (official) That means the condition hasn’t changed in eight years.

So even if a Korean-language site adds speakable now, Google Assistant won’t read it aloud. There’s no harm in adding it either. In April 2025, John Mueller said structured data doesn’t raise a site’s rankings, and that using schema.org items not on Google’s feature list won’t cause problems but will make almost no visible difference in search. Search Engine Roundtable Your time is better spent on the structured data Google actually displays, such as types for articles, products and organization information.

It Has Nothing to Do with Rankings

The claim that adding speakable raises your search rankings is not true. Google’s general structured data guidelines state that even accurate markup that passes the Rich Results Test is not guaranteed to appear in search results; structured data makes a feature possible but does not guarantee it will appear. Google (official) The troubleshooting section of the speakable documentation also notes that even with markup, selection is decided by algorithmic ranking, and Assistant reads at most three articles from different publishers. Google (official) Markup makes you eligible; being eligible and being picked are two different things. This is in line with the answer optimization that AEO talks about. A page with well-organized answers comes first, and markup comes after.

Gemini Is Taking Over the Speaker

The devices speakable targeted were speakers with Google Assistant. But in March 2025, Google announced it would replace the existing Assistant on phones with Gemini, making Assistant unavailable on most devices, and previewed bringing a Gemini-powered experience to home devices such as speakers, displays and TVs. Google (official) In October 2025, it announced that the Gemini for Home voice assistant would upgrade and replace Assistant on speakers and smart displays. Google (official) It plans to start with early access in the US and expand to other countries in 2026. Google (official)

There is no official word yet that Gemini for Home reads speakable markup. The speakable documentation, updated in September 2026, still mentions only Google Assistant. Google (official) So the accurate reading is that this property is alive in the documentation, its devices are changing, and it was never opened to Korean in the first place. The broader trend of voice and AI reading our content aloud is a GEO question, and the answer is not a line of markup but short, accurate paragraphs.

References

Sep.2026By Studio JT

More from Technical

Like this term?
Give it a like!

(52)