What Speakable does and what it does not do
Speakable is a locator, not a writing format or ranking signal. It identifies existing portions of an Article or WebPage that may be suitable for speech. Schema.org’s SpeakableSpecification definition describes selectors or XPath as ways to identify those sections. Google documents a narrower product use: topical news answers with audio playback on Google Assistant enabled smart speakers.
Its feature is beta, subject to change, and its availability note covers U.S. users with Google Home set to English and publishers with English content. This is not a general promise of voice search exposure, AI Overviews placement, or answer engine citations. Markup remains one supporting layer in AEO implementation; it cannot replace useful content or a clear page purpose.
Decide whether a page is a sensible candidate
Begin with the user experience, not a schema generator. Speakable may merit consideration for timely news coverage when a concise passage can answer a topical question without nearby images or context. An editor might select a headline and factual summary of a transit disruption. A service business page about appointment hours is not automatically eligible for Google’s documented news feature just because customers could listen to it; markup alone creates no eligibility.
For broader audio strategy, distinguish this use from voice search AEO, which addresses user behavior and platform contexts, not a promise attached to this property. Record the intended audience, content type, and product use. If no specific editorial or consumer case exists, skip the annotation rather than adding it sitewide.
Write for listening before choosing a selector
Choose the passage before writing the markup; schema cannot make a confusing sentence sound clear. Google recommends concise, useful headlines or summaries, and suggests about 20 30 seconds per section roughly two or three sentences. If selecting the top of a story, rewrite it into short, self contained sentences. For example: “The city closed the eastbound bridge after an overnight inspection found structural damage. Bus routes 4 and 7 are diverted until engineers complete a safety review.” Each sentence works without a chart or surrounding paragraph.
Avoid “see the chart,” unexplained abbreviations, datelines, captions, and source attributions that sound odd in isolation. Keep marked text visible and consistent with the article. The answer first writing guide offers a related way to state the main point before background.
Map only the intended visible elements
Attach SpeakableSpecification to the page’s Article or WebPage item and use either cssSelector or XPath not both in the same specification for Google’s documented implementation. CSS selectors work well when templates provide stable classes. This example is illustrative: replace both classes with elements present in the rendered page, and use the right page type and headline.
- 01Inspect the final HTML and confirm each selector resolves to the exact headline or summary intended for audio.
- 02Use narrow selectors; a generic paragraph class may match every paragraph or a promotional block as well as the summary.
- 03Keep the marked text in the reader facing page. Do not create a hidden audio only summary to satisfy the markup.
{
"@context": "https://schema.org",
"@type": "Article",
"headline": "Bridge closed after inspection",
"speakable": {
"@type": "SpeakableSpecification",
"cssSelector": [
".article-title",
".audio-summary"
]
}
}Choose CSS selectors or XPath deliberately
CSS selectors target elements through HTML structure and classes; XPath expresses a path through a document and can, in Google’s example, address title or description metadata. Choose a method your team can maintain. A selector such as .audio summary is useful only if the production template assigns that class to the intended visible summary. Avoid paths tied to temporary wrapper depth, generated class names, or broad selectors that may collect navigation or every paragraph.
Schema.org also describes ID references, but Google’s feature instructions present cssSelector and xPath as its locator choices. Follow each consumer’s current documentation. For patterns across page types, see JSON LD templates for AEO, applying only properties that fit the visible page.
Validate the rendered page, not just the JSON
Valid JSON does not prove that a selector points to the right passage or that the passage makes sense aloud. Test after the CMS and template render the page. Confirm JSON LD is present, the page type is accurate, each locator selects only the intended visible text, and the audio passage is clear without visual context. Read it aloud for acronyms, long clauses, ambiguous pronouns, and references such as “this figure.”
Google’s general structured data policies require markup to represent relevant page content and warn against hidden or misleading content; valid markup does not guarantee a search appearance. Use the schema validation guide to distinguish technical checks from editorial review.
Build a lightweight editorial and release workflow
Agencies do not need to add selectors to every template by default. Have an editor nominate a headline or summary and confirm it stands alone; a developer maps it to stable markup; QA checks the production URL; and the content owner verifies the wording after updates. Record the URL, page type, selected text, selector, reviewer, and last checked date. Test representative pages from each template, then inspect outliers individually. Retest when a component, class, or article structure changes.
If an article is corrected or substantially rewritten, revisit the passage and remove or update markup when it no longer selects accurate text. This small workflow catches stale or misleading annotations before they spread.
Set expectations and maintain the annotation
Speakable is metadata that may help a supported system identify an audio suitable passage; it is no contract that a system will fetch, read, rank, or cite it. Google describes a beta subject to change and does not guarantee structured data features will appear. Selection remains algorithmic, so an accurate annotation cannot ensure a story is returned. Separate implementation checks valid JSON LD, accurate locator, current visible wording from outcomes the publisher does not control. Do not report deployment as a ranking or citation increase without independent evidence.
Before extending to new page types, countries, languages, or products, check current documentation. For related terminology, see Rankwox’s schema markup explainer.
Key takeaways
- Speakable identifies passages for possible text to speech use; it does not guarantee rankings, citations, or playback.
- Google’s documented beta is narrowly focused on topical news playback, with U.S. and English language conditions.
- Choose a short passage that makes sense aloud, then map it to visible text with one stable locator method.
- Test the production rendered page and review the selector whenever the article or template changes.
Frequently asked questions
Does Speakable schema improve rankings or AI citations?+
Google does not describe Speakable as a ranking boost or promise AI citations. Its documentation covers a beta topical news audio use and says structured data does not guarantee a search appearance. Treat markup as a way to identify a suitable passage, not as a performance claim.
Can any business add Speakable markup to its website?+
Schema.org defines SpeakableSpecification for use with Article or WebPage, but that vocabulary does not mean every page qualifies for a Google feature. Google’s documented use is topical news playback, with stated U.S. and English language conditions. Other consumers may have different behavior, so verify their current documentation before promising a use.
How long should a Speakable passage be?+
Google suggests around 20 30 seconds per section, or roughly two to three sentences, and recommends concise headlines or summaries. Use that as editorial guidance, not as a guaranteed threshold. The passage should be self contained and useful when heard without the page.
Should I use CSS selectors or XPath?+
Google documents either CSS selectors or XPath locators and says not to use both in the same SpeakableSpecification. Choose whichever your team can keep stable, and check that it resolves to the intended visible text on the rendered page.
