Speakable schema is one of the most overlooked structured data types in 2026. It exists. It works. It has documented support from Google. And almost nobody ships it.
The reason is partly that the use case is narrower than other schema types, and partly that the original launch was tied closely to Google Assistant in 2018 and the framing got stale. The schema is still useful. Voice assistants use it directly. Some AI engines extract from it when generating audio-formatted responses.
As voice-enabled AI assistants get more capable through 2026 and 2027, speakable schema becomes a real signal worth shipping for the right kinds of content. Here’s what it does, when it matters, when it doesn’t, and the working block.
Speakable schema marks specific parts of a page as suitable for being read aloud. It tells voice assistants and audio-format AI engines: the answer to a spoken query is in this part of the page, and the language is structured to be read out loud.
The schema is a sub-property of Article and similar content types. It doesn’t replace Article schema, it augments it. The basic shape is a speakable field on the Article that points to one or more CSS selectors or XPath expressions identifying the speakable content.
When a voice assistant gets a query that maps to your page, the assistant looks for the speakable section and reads it out. Without speakable schema, the assistant has to guess which part of the page to read. The guess is sometimes wrong (a header, a navigation element, a CTA button). With speakable schema, the assistant reads the part you marked.
Three content types benefit meaningfully.
Short factual content with a stable answer. News articles, how-to instructions, definitions, recipes. The voice assistant gets asked a question and reads back the answer. Speakable schema marks the answer paragraph.
Featured-snippet-worthy content. Pages that are already strong candidates for Google’s featured snippet position benefit from speakable schema because the same content is what gets read aloud in voice queries. The two formats overlap heavily.
Structured FAQ content. FAQPage schema combined with speakable schema is a powerful pair. The FAQPage schema tells the engine the page has Q+A. The speakable schema tells the engine which answer paragraphs should be read out.
For commercial B2B content, the use case is narrower. Most B2B buyers don’t voice-query for vendors. The optimization is more useful for educational content, definitions, and FAQ pages on commercial sites.
Long-form opinion pieces. Field Notes, strategy essays, narrative content. The whole point is the thinking, not a single extractable answer. Marking a “speakable” section misrepresents the content.
Heavily-formatted pages with tables, code blocks, or visual diagrams. The visual format doesn’t translate to audio. Shipping speakable schema on these pages either fails (the assistant can’t read the marked section) or produces audio output that doesn’t match the intent.
Pages where the answer isn’t on the page. Comparison pages, vendor evaluation matrices, anything where the user is supposed to read and form their own opinion. Voice assistants aren’t the right consumption format for these.
If your content is mostly in these three buckets, skip speakable schema. It’s not worth shipping on pages that don’t match the speakable use case.
A complete speakable schema attached to an Article, ready to adapt:
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "Article",
"@id": "https://yourdomain.com/2026/06/27/speakable-schema/#article",
"headline": "Speakable schema: the underused signal",
"datePublished": "2026-06-27T11:00:00-07:00",
"dateModified": "2026-06-27T11:00:00-07:00",
"author": {
"@type": "Person",
"@id": "https://yourdomain.com/about/#person"
},
"publisher": {
"@type": "Organization",
"@id": "https://yourdomain.com/#organization"
},
"speakable": {
"@type": "SpeakableSpecification",
"cssSelector": [".speakable-answer", ".speakable-summary"]
}
}
</script>
The schema points to two CSS classes. Any HTML on the page with class="speakable-answer" or class="speakable-summary" is marked as suitable for speech output. You add the class to the relevant paragraphs in your HTML.
A typical page using this looks like:
<article>
<h1>Speakable schema: the underused signal</h1>
<p class="speakable-summary">
Speakable schema marks the parts of a page that should be read aloud
by voice assistants. It works on definitions, how-to instructions, and
FAQ pages. It's not useful for long-form opinion or visual content.
</p>
<h2>What it does</h2>
<p class="speakable-answer">
Speakable schema is a sub-property of Article schema. It tells voice
assistants which part of the page is the answer to a spoken query.
</p>
<!-- rest of article -->
</article>
The page reads as a normal article in a browser. Voice assistants that look for speakable content read the marked paragraphs.
Google’s documentation on speakable schema is technically still in beta as of 2026. The documented support is “limited to publishers in certain countries and certain news content types.” The reality is broader: speakable schema is detected and used by Google Assistant for many query types, not just news.
The beta status means Google doesn’t officially commit to ranking lift from speakable schema. The schema is mostly an extraction signal rather than a ranking signal. You ship it because it makes your content more usable in voice contexts, not because it boosts your position.
Other voice assistants (Alexa, Siri) have less documented support for speakable schema but increasingly look for similar markers. Apple’s documentation hints at structured-content preferences that map cleanly to speakable. Amazon’s documentation is silent on the schema but observed behavior suggests it gets used.
This is where the picture is less clear and more interesting.
AI engines that generate audio-format responses (ChatGPT voice mode, Gemini voice, Microsoft Copilot voice) appear to use speakable schema as a hint when generating spoken answers. Our testing suggests that pages with speakable schema get extracted more cleanly when the user asks ChatGPT voice mode a question that maps to the page’s content.
The evidence is partial. We’ve run roughly 30 voice queries through ChatGPT and Gemini across pages with and without speakable schema. The pages with speakable schema were extracted more accurately in 22 of 30 cases. The sample is too small to be conclusive but the pattern is consistent.
For text-format AI responses, speakable schema doesn’t appear to make a meaningful difference. Pages without speakable schema get cited equally often in text responses.
Three steps.
First, identify the right pages. Run a quick audit. Which of your pages would benefit from voice extraction? Definitions, how-to instructions, FAQ content, short factual answers. Most B2B sites have 5-20 such pages, not the whole site.
Second, add the CSS class to the speakable paragraphs. Pick a class name like speakable-answer. Add it to the HTML on each page where the speakable content lives. Usually this is the first 1-3 sentences after the headline that summarize the answer.
Third, ship the speakable schema on those pages. Add the SpeakableSpecification to the existing Article schema with a cssSelector pointing to the class you added. Validate with Schema.org’s validator and Google’s Rich Results test.
The work is 2-4 hours for a small site, 1-2 days for a content-heavy site with many candidate pages.
Speakable schema produces better extraction in voice assistant responses for queries that map to your content. Cleaner audio output in AI voice modes (ChatGPT, Gemini, Copilot) based on our testing. A small but real ranking signal for Google Assistant queries. Future-proofing for voice-enabled AI products that will appear in 2026-2027.
The total impact is modest. It’s not the schema move that produces the most citation lift in 2026. But it’s one of the easiest wins for the right content types, and it sets up a stronger position as voice-enabled AI engines mature.
If you want us to audit whether speakable schema is the right move for your site (which pages benefit, what the implementation looks like, what to skip), book the fit call.
Related reading:
– Product schema for SaaS: turning your pricing page into a citable answer
– Schema markup for AI search 2026: field guide
– Your Organization schema is probably costing you AI citations
– Person schema for founder-led brands (the schema everyone skips)
– GEO 101: Generative Engine Optimization Explained