Which page sections does Google recommend marking as speakable?

The Citead team, updated

In short. Google recommends concise news headlines or summaries focused on key points and lasting about 20 to 30 seconds per speakable section.

Definition: Google recommends marking concise, self-contained news headlines or summaries that communicate key points clearly when read aloud. Its speakable guidance recommends about 20 to 30 seconds of content, or roughly two to three sentences, for each marked section.

Key facts

  • Google's speakable documentation describes the feature as beta and subject to change.
  • Marked content should contain concise headlines or summaries with comprehensible and useful information.
  • Google recommends selecting key points instead of marking an entire article.
  • The target is around 20 to 30 seconds of content per section, roughly two to three sentences.
  • Datelines, photo captions and source attributions can sound confusing in voice-only or voice-forward situations and should not be marked.

What text should a section contain?

Google's content guidelines ask for concise headlines or summaries that give users comprehensible and useful information. The guidance is about the text a listener receives, so a selected passage needs to stand on its own as a short account of the relevant news point.

Google recommends focusing on key points rather than highlighting an entire article. The same technical guidance explains that this lets listeners get an idea of the story and avoids having text-to-speech readout cut off important details. A long body section, a caption that depends on an image, or an attribution that only makes sense in the page layout is therefore outside the recommended selection.

How long should the marked section be?

Google recommends around 20 to 30 seconds of content per speakable section, or roughly two to three sentences. This is a duration and sentence-count guide for an audio experience, rather than a stated word limit. It applies to each section, so a page can use more than one concise section when the markup identifies them separately.

If the selected text is at the top of the story, Google suggests rewriting it into individual sentences so that it reads more clearly for text-to-speech. The recommendation concerns the wording of the marked news text, not a separate summary hidden from readers.

Which parts of the page should be left out?

Google's technical guidance says not to add speakable structured data to content that may sound confusing in voice-only and voice-forward situations. It names datelines, which identify where a story was reported, photo captions and source attributions as examples.

The Schema.org SpeakableSpecification definition describes the type as identifying document sections highlighted as particularly speakable, usually through xpath or cssSelector. Schema.org's speakable property also says that other page sections can be useful in particular circumstances, while the property identifies the parts most likely to be generally useful for speech. Google's news guidance supplies the narrower editorial test: the selected passage should be concise, understandable and useful when heard alone.

For Google's documented implementation, the locator uses one of two alternatives. A page can identify the selected elements with cssSelector or with xPath, but Google's structured data table says to use either one and not both.

"speakable": {
  "@type": "SpeakableSpecification",
  "cssSelector": [".headline", ".summary"]
}

FAQ

Does Google give a word count for speakable sections?

No word count is stated in Google's guidance. It gives around 20 to 30 seconds or roughly two to three sentences per section.

Should a whole news article be marked as speakable?

Google recommends focusing on key points rather than marking an entire article. The selected text should be a concise headline or summary that remains comprehensible and useful when read aloud.

Can a speakable section include a photo caption?

Google names photo captions as content that may sound confusing in voice-only and voice-forward situations. Its technical guidance says not to add speakable structured data to such content.

Sources

  1. Google Search Central, "Speakable (Article, WebPage) structured data (BETA)", 8 September 2026
  2. Schema.org, "SpeakableSpecification", accessed 17 October 2026

Check how a page reads to answer engines. Open the GEO score analyzer

Test a page and see what to fix first

The analysis usually takes less than 10 seconds and needs no account.

400,000+ pages scored in the public benchmark