Voice interfaces are becoming a serious traffic source. Google Assistant, Amazon Alexa, Apple Siri, and a growing number of AI-powered tools now read content aloud to users. The question is: whose content gets chosen? The answer, more often than not, comes down to whether your page has told the AI what to read. That is exactly what Speakable schema does.
This post covers what Speakable schema actually is, how to implement it correctly in JSON-LD, where to put it on your site, and what mistakes to avoid. If you have been wondering why your content does not surface in voice responses despite ranking well on Google, this is likely part of the reason.
What Speakable Schema Actually Does
Speakable schema is a structured data type defined by Schema.org that allows you to mark specific sections of a webpage as appropriate for text-to-speech playback. You are essentially pointing an AI assistant or voice-enabled device to the exact passages that summarise your content best.
Think of it as a highlights reel. A full article might be 2,000 words, but a voice assistant cannot read all of that to a user asking a quick question. Speakable schema lets you pre-select the two or three sentences or paragraphs that are most informative, most direct, and most useful to someone listening rather than reading.
Google officially supports Speakable schema as part of its structured data guidelines, primarily for news articles. However, the underlying logic applies far more broadly. AI systems that synthesise and read content, including Perplexity and certain Gemini features, look for signals that indicate which portions of a page are authoritative and quotable. Speakable schema is one of the clearest signals you can send.
The JSON-LD Implementation
Speakable schema is implemented using JSON-LD, which is the format recommended by Google and the one that causes the fewest integration headaches. You reference specific sections of your page using either CSS selectors or XPath expressions.
Here is a basic working example for a news-style article or a service page:
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "WebPage",
"name": "How FlinnSchema Improves AI Visibility",
"speakable": {
"@type": "SpeakableSpecification",
"cssSelector": [".speakable-summary", "h2.intro", "#key-points"]
},
"url": "https://flinnschema.com/what-we-do-differently"
}
</script>In this example, any HTML element with the class speakable-summary, the H2 with class intro, or the element with the ID key-points will be marked as speakable content. The AI assistant reads those sections specifically.
Using XPath Instead of CSS Selectors
If your site structure makes CSS selectors awkward, XPath is an alternative. It looks like this:
"speakable": {
"@type": "SpeakableSpecification",
"xpath": [
"/html/body/article/section[1]/p[1]",
"/html/body/article/section[1]/p[2]"
]
}XPath targets very specific DOM positions. The trade-off is that it becomes fragile if your page structure changes. A CMS update or template tweak can break the reference entirely. For most sites, CSS selectors are easier to maintain because they are tied to class names rather than document position.
Nesting Speakable Inside Article Schema
For blog posts and news articles, you will typically want to nest the Speakable specification inside an Article or NewsArticle schema block rather than a bare WebPage type. This gives the AI more context about the content type:
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "NewsArticle",
"headline": "How Speakable Schema Helps AI Read Your Content",
"author": {
"@type": "Person",
"name": "Your Name"
},
"speakable": {
"@type": "SpeakableSpecification",
"cssSelector": [".article-summary", ".key-takeaway"]
},
"url": "https://example.com/your-article"
}
</script>Nesting it this way also plays well with other schema types on the same page. You are building a richer, more connected structured data layer, which matters for overall AI readability.
Which Pages Should Have Speakable Schema
Google's documentation focuses on news publishers, but that framing is too narrow for most businesses using schema today. The practical answer is: any page where a short, authoritative excerpt would be genuinely useful to someone asking a voice assistant a question.
That includes:
- Blog posts with a clear summary or key takeaway paragraph
- Service pages that answer "what does this company do?"
- FAQ pages where answers are concise and factual
- Product pages with a strong one or two sentence description
- About pages that summarise who you are and what you offer
Avoid marking up huge blocks of text as speakable. The purpose is selectivity. If you mark your entire page as speakable, you have told the AI nothing useful. Pick the one or two passages that genuinely stand alone as answers to questions people ask.
Writing Content That Works When Spoken
Speakable schema points to your content, but if the content itself is not written for speech, it will not land well when read aloud. There is a real craft to writing speakable passages.
Keep Sentences Declarative and Self-Contained
Avoid sentences that rely on visual context. "As shown in the table above" means nothing to someone listening. Instead, write as if the listener has no visual reference. "FlinnSchema adds structured data to your website so AI search engines can read and cite your pages directly" is self-contained. It makes sense without context.
Front-Load the Key Information
Radio journalists write with the most important fact first. Do the same. A voice assistant might only read the first sentence of your speakable section before moving on. If that sentence is a preamble rather than a point, you have wasted the slot. Get to the point in the first eight words.
Avoid Jargon and Unexplained Acronyms
Acronyms that are obvious to readers can confuse listeners. If you write "JSON-LD schema" in a speakable section, consider whether a voice assistant's target audience will know what that means. You may want to write it out or contextualise it. "Structured data code called JSON-LD" is more listener-friendly than "JSON-LD".
Use Simple, Active Sentences
Passive constructions sound strange when read aloud. "The schema is implemented by adding a script tag" is harder to process aurally than "You add a script tag to implement the schema." Active voice, present tense, direct address. These improve both listenability and AI comprehension.
Validating Your Speakable Schema
Once you have added Speakable schema, test it. Google's Rich Results Test does not currently surface a specific Speakable result in its preview pane, but it will confirm that the markup is valid and free of errors. The Schema Markup Validator at validator.schema.org is also worth running to check that your JSON-LD is correctly structured.
Check that:
- The CSS selectors you referenced actually exist in the page HTML
- The selectors return the content you intended, not empty elements
- The schema is in the
<head>or early<body>of the page, not buried at the bottom - There are no syntax errors in the JSON (missing commas and unclosed brackets are common)
If you are not confident validating schema yourself, our free AI visibility audit checks for structured data issues including missing or broken Speakable markup across your key pages.
Speakable Schema and Broader AI Visibility
Speakable schema does not exist in isolation. It works best as part of a fuller structured data strategy. A page that has Article schema, Author schema, and Speakable schema together is significantly more legible to AI systems than a page with just one of those elements.
The reason is that AI systems are probabilistic. They are looking for confidence signals before quoting a source. Multiple structured data types on the same page, each pointing to accurate and useful information, increase the overall confidence score that an AI assigns to your content. Speakable schema is one layer; Author schema is another; Review schema adds social proof. Together they build a picture of a trustworthy, well-organised source.
At FlinnSchema, we see this pattern repeatedly in client results. Pages with layered, accurate structured data consistently outperform bare pages in AI citation rates, even when the bare pages rank higher in traditional Google search. AI search and traditional search reward different signals, and Speakable is one of the signals that traditional SEO tools simply do not track. For a broader look at how these two worlds differ, the post on how AI visibility differs from traditional SEO is worth reading.
Common Mistakes to Avoid
A few errors come up repeatedly when sites first implement Speakable schema.
Marking up too much. If your speakable CSS selector targets the entire article body, you have not helped anyone. Be selective. Aim for 50 to 150 words of speakable content per page, covering your clearest and most direct points.
CSS selectors that do not match. If your page uses dynamically generated class names (common in some React or Vue setups), your selectors may not resolve correctly. Test with a static build or use IDs instead, which tend to be more stable.
Speakable content that is not actually on the page. The text you mark as speakable must exist as readable text in the DOM. If it is loaded dynamically via JavaScript after page render, some AI crawlers may not see it at all. Keep speakable content in static HTML where possible.
Forgetting to update selectors after a redesign. A site redesign often changes class names and element structures. Speakable selectors are easy to forget. Add a post-launch schema audit to your redesign checklist. Our automations and implementation service can handle this automatically so your schema stays current after any changes.
Frequently Asked Questions
Is Speakable schema only for news publishers?
Google's official documentation targets news publishers, but the schema type itself is applicable to any webpage. Service businesses, e-commerce brands, and blog publishers can all benefit from marking up concise, authoritative content as speakable. The schema signals to AI systems which parts of your page are worth reading aloud, regardless of industry.
Will adding Speakable schema affect my Google rankings?
Speakable schema is not a direct ranking factor for traditional Google search results. It primarily affects how your content is surfaced in voice search and AI assistant responses. That said, having well-structured markup across your site is generally associated with better crawlability and AI visibility, which are increasingly important as AI search grows.
Can I use Speakable schema on product pages?
Yes. Product pages often have a short description or key benefit statement that is ideal for speakable markup. If someone asks a voice assistant about a product you sell, you want your concise, accurate description to be what gets read back. Combine it with Product schema for the best results. Keep the speakable section to two or three sentences at most.
How do I know if my Speakable schema is working?
You can validate the markup using Google's Rich Results Test and the Schema Markup Validator at validator.schema.org. For real-world testing, try asking Google Assistant or a similar tool a question that your speakable content answers directly, and see whether your page is cited. Track AI referral traffic in Google Analytics to spot patterns over time.