# Speakable Schema: What It Is and Where It Works in 2026

> What the Schema.org speakable property is, what Google has used it for since 2018, where it works (U.S., English, news) and whether it helps a store.

URL: https://clione.ai/glossary/speakable/
Lang: en

Speakable tends to show up on lists of "markup for voice search," and the name helps: it sounds like the piece that makes an assistant read your site aloud. It exists, it is a property of [Schema.org](/glossary/schema-org/) for [structured data](/glossary/structured-data/) and Google documents it, but **its real use is much narrower than the name suggests**: news, in English, in the United States, on speakers. A store should know what it does so as not to spend time on it, and to recognize it when it turns up in an audit or an agency proposal.

## How it works

In Schema.org, `speakable` is a property, not a type. It indicates which sections of a web page are especially "speakable," in the sense that they are the best suited to being converted from text to speech. It is used on Article and WebPage, and its value is a SpeakableSpecification or a URL.[^schemaorg-speakable] Schema.org allows three ways to point at the content: a URL pointing at the `id` of an element on the page, a CSS selector or an XPath expression.[^schemaorg-speakable] The SpeakableSpecification, which is an intangible type, covers the last two in its `cssSelector` and `xpath` properties.[^schemaorg-speakablespec]

Google describes what it uses it for. **Google Assistant uses it to answer queries about current news on smart speakers**: when someone asks for news on a topic, the Assistant returns up to three articles from different sites and reads aloud the sections marked with `speakable`. As it reads them, it credits the source and sends the full article's URL to the person's phone, through the Assistant app.[^google-speakable]

The implementation Google asks for is stricter than Schema.org's. Its documentation only covers two ways to point at the content, `cssSelector` or `xPath`, and asks you to use one of the two, never both; the property can be repeated as many times as needed.[^google-speakable] And it gives content guidelines:

- Do not mark what sounds confusing when you only hear it: the dateline of the story, photo captions or source attributions.[^google-speakable]
- Do not mark the whole article, but the key points, so that the reading does not get cut off before the important part.[^google-speakable]
- Concise headlines and summaries, and if you mark the opening of the story, rewrite it in separate sentences so it is easier to follow when read aloud.[^google-speakable]
- **About 20 to 30 seconds of content per section, roughly two or three sentences.**[^google-speakable]

### Since when and where

Google introduced it on July 24, 2018 on its webmaster blog. The announcement explained that it had collaborated on this new Schema.org specification so that news publishers could mark the parts of an article best suited to being read by the Assistant, that it was available to English-language users in the United States, and that it hoped to **bring it to other languages and countries once enough publishers had implemented it**. It added that, being new, they were going to experiment to refine it.[^google-speakable-2018]

More than eight years later, Google's documentation, updated on September 8, 2026, still says the feature is **in beta and subject to change**. It limits availability to U.S. users with Google Home devices set to English and to publishers who publish in English, and repeats the hope of reaching other countries and languages when there are enough publishers.[^google-speakable] Speakable is still on the list of structured data features Google supports.[^google-gallery] And according to the Schema.org page, with data from Google's index from August 2026, the property is used on between 100,000 and one million domains.[^schemaorg-speakable]

## In a store

A **fictional example**: a backpack store publishes a news post on its blog about its fall collection, with a headline and a two-sentence summary that carry the classes `headline` and `summary`. This would be its markup:

```json
{
  "@context": "https://schema.org",
  "@type": "WebPage",
  "name": "The waxed-canvas backpack fall collection is here",
  "url": "https://www.example.com/blog/fall-collection",
  "speakable": {
    "@type": "SpeakableSpecification",
    "cssSelector": [".headline", ".summary"]
  }
}
```

It reads like this: the page is a WebPage, and its `speakable` property points at two elements by their CSS selector, the headline and the summary, which together stay within the two or three sentences Google recommends. It uses `cssSelector` and not `xPath`, as its documentation asks.[^google-speakable] It is correct markup. But for a store, Google's documentation gives it no use today: the feature covers news published for the Assistant, in English and for U.S. users, so a store that sells outside the United States or writes in another language gets nothing from it.[^google-speakable] And on the [product page](/glossary/product-page/) it has no documented role: Google describes it for answering news queries, not for reading products aloud.


That `speakable` is the key to voice search in general. Google's documentation limits it to one thing: the Assistant reading news snippets aloud on speakers, for U.S. users and in English.[^google-speakable] It describes no other use. Nor is it needed for Search's AI features: Google says there is no special markup from Schema.org that you have to add for generative search.[^google-ai-guide]


## How to check it

1. **Search for `speakable` in the source code** of your pages. If you do not find it, you are missing nothing that Google uses outside the United States.[^google-speakable]
2. **Check that every selector exists.** Open the browser console on that page and type `document.querySelectorAll(".summary")` with your selector: if it returns an empty list, the property points at an element that is not there.
3. **Measure what you mark.** Read each section's text aloud: it should come to about 20 or 30 seconds, two or three sentences.[^google-speakable]
4. **Make sure `cssSelector` and `xPath` are not mixed** in the same property: Google asks for one or the other.[^google-speakable]
5. **Run the page through a [schema validator](/glossary/schema-markup-validator/)** to rule out syntax errors in the [JSON-LD](/glossary/json-ld/) block.

## Related concept: llms.txt

The two are sometimes presented as ways to "tell a machine which part of your site to read," and they resemble each other in nothing else. `speakable` is a structured data property, inside the page, that points at fragments to be read aloud.[^schemaorg-speakable] [`llms.txt`](/glossary/llms-txt/) is a separate text file, made for language models. Your presence in Google's AI features depends on neither: its guide says you do not need to create machine-readable files, AI text files or special markup to appear in Search, generative features included.[^google-ai-guide] **Speakable has one documented, very specific use; outside it, it is not a signal for AI.**

To keep going, [rich results](/glossary/rich-results/) explain which markup does change how a page appears on Google, and the [structured data reading path](/learn/structured-data/) orders the rest of the terms.