Speakable Schema for Voice AI Assistants: Optimizing Content for Spoken Responses

What Is Speakable Schema?

Speakable Schema (formally SpeakableSpecification) is a structured data type defined by schema.org that identifies which sections of a web page are most suitable for audio playback using text-to-speech technology. It tells voice AI assistants -- such as Google Assistant, Apple Siri, Amazon Alexa, and emerging voice interfaces -- which content to read aloud when a user asks a voice query that matches your page.

Without Speakable Schema, a voice assistant that selects your page as a source must decide on its own which portion to read. It might select the opening paragraph, a random section, or attempt to summarize the entire page -- often with poor results. With Speakable Schema, you explicitly guide the assistant to your most concise, self-contained, and informative content.

The markup works by using CSS selectors or XPath expressions to point to specific HTML elements. For example, you can mark your key takeaways section, your H1 heading, and the first paragraph of your article body as the speakable portions. For a broader understanding of structured data in AI SEO, see our JSON-LD basics guide.

Why Voice AI Matters for SEO

Voice AI is a growing channel for content discovery and consumption. Over 4 billion voice assistants are in active use globally, and voice queries account for an increasing share of search interactions. The intersection of voice AI and AI SEO creates specific optimization opportunities.

Key voice AI statistics

• 50%+ of US adults use voice assistants daily
• Voice queries tend to be longer and more conversational than typed queries
• Voice responses are typically single-source -- only one website gets cited per query
• Smart speakers and in-car assistants are growing fastest as voice AI interfaces
• Apple's Safari integration with Claude brings voice AI to the broader web browsing context

The single-source advantage

Unlike text-based AI responses that may cite 2-5 sources, voice responses typically cite only one. When Siri or Google Assistant reads an answer aloud, it selects the single best source. Being that source means capturing the entire voice query -- there is no second position. This makes voice AI optimization a winner-take-all competition.

For a comprehensive guide to optimizing for voice AI beyond Schema markup, see our article on voice AI assistants optimization.

Implementation: The JSON-LD Code

Speakable Schema is implemented as a nested property within a WebPage schema type. Here is the standard implementation:

{
  "@context": "https://schema.org",
  "@type": "WebPage",
  "name": "Your Page Title",
  "speakable": {
    "@type": "SpeakableSpecification",
    "cssSelector": [
      ".key-takeaways",
      "h1",
      ".article-body > section:first-child"
    ]
  }
}

Using CSS selectors

The cssSelector property accepts an array of CSS selectors that point to the HTML elements containing speakable content. The most common selectors:

• .key-takeaways -- Your key takeaways or summary section (highest priority)
• h1 -- The main page heading (provides context for the topic)
• .article-body > section:first-child -- The first section of the article body (typically the definition or introduction)

Using XPath (alternative)

If CSS selectors are insufficient, you can use XPath expressions instead:

{
  "speakable": {
    "@type": "SpeakableSpecification",
    "xpath": [
      "/html/body/article/aside[@class='key-takeaways']",
      "/html/body/article/header/h1"
    ]
  }
}

CSS selectors are recommended over XPath because they are more familiar to developers and less brittle when HTML structure changes.

Choosing What to Mark as Speakable

Not all content works well when read aloud. The sections you mark as speakable must be suitable for audio delivery, which means they must be self-contained, concise, and free of elements that require visual presentation.

Content that works as speakable

• Key takeaways / summary sections -- Concise bullet points that summarize the article
• Definition paragraphs -- Self-contained definitions that answer "What is X?"
• Answer paragraphs in FAQs -- Direct responses to specific questions
• Opening paragraphs -- Introduction paragraphs that provide context
• Conclusion paragraphs -- Summary statements with recommendations

Content to avoid marking as speakable

• Tables -- Column and row structures do not translate to spoken format
• Code blocks -- Programming code is meaningless when read aloud
• Navigation elements -- Breadcrumbs, menus, and internal links
• Image descriptions -- Alt text and captions that reference visuals
• Numbered lists with complex items -- Long list items become confusing when spoken
• Content with many URLs or technical references -- URLs are disruptive when read aloud

Length guidelines

Keep speakable content concise. Google recommends that speakable content should be:

• No more than two to three sentences per speakable section
• A total of approximately 20-30 seconds of audio across all speakable sections
• Written in natural, conversational language that sounds good when read aloud

Writing Content for Spoken Delivery

Content marked as speakable should be written or edited specifically for audio delivery. Several writing principles apply.

Use natural sentence structure

Write sentences that sound natural when read aloud. Short, declarative sentences work best. Avoid complex subordinate clauses, parenthetical asides, and sentences that exceed 25 words.

Not speakable-friendly: "AI SEO, which encompasses GEO (Generative Engine Optimization), AEO (Answer Engine Optimization), and LLMO (Large Language Model Optimization), is the practice of..."

Speakable-friendly: "AI SEO is the practice of optimizing your website for AI-powered search tools like ChatGPT and Gemini. It helps AI models find and cite your content in their responses."

Avoid abbreviations and jargon

Spell out abbreviations on first use. Voice assistants may read "SEO" as "S-E-O" or "see-oh" depending on context. Write for clarity when spoken.

Front-load the answer

The first sentence of a speakable section should contain the complete answer. Additional sentences provide supporting detail but the core answer must come first. This aligns with the BLUF principle and ensures that even if the voice assistant truncates the response, the user gets the essential information.

Read your content aloud

The simplest test: read your speakable sections out loud. If you stumble over phrases, if sentences feel too long, or if meaning is unclear without visual context, revise for spoken delivery.

Speakable Schema and Text-Based AI Citations

While Speakable Schema is designed for voice AI, the sections you mark as speakable overlap significantly with the content that text-based AI models like ChatGPT, Gemini, and Perplexity prefer to cite.

The reason is structural: both voice and text AI models need concise, self-contained, clearly defined content that can be extracted from a page and presented to a user. The key takeaways section you mark as speakable is the same section that text AI models extract as a quotable chunk. The definition paragraph in your first section serves both voice responses and text citations.

This overlap means that implementing Speakable Schema is not just a voice optimization -- it is a forcing function that improves your content structure for all AI citation channels. By identifying your most important content sections with CSS selectors, you also clarify for yourself which parts of your content are truly self-contained and citation-worthy.

Current Support and Future Outlook

Speakable Schema support is currently in beta with limited deployment, but the trajectory points toward broader adoption.

Current status (2026)

• Google: Supports Speakable Schema as a beta feature for news articles in English
• Apple Siri: Does not explicitly process Speakable Schema but uses similar content selection heuristics
• Amazon Alexa: Uses proprietary content selection; Speakable Schema compliance helps indirectly
• Microsoft Copilot: Voice features expanding; structured data awareness growing

Why implement now

Even with limited current support, implementing Speakable Schema now is valuable for three reasons:

1. Early adopter advantage -- When voice AI platforms expand Speakable support, your content is already optimized
2. Content quality improvement -- The process of selecting speakable sections forces better content structure
3. Low implementation cost -- Adding one JSON-LD block with CSS selectors takes minutes per page template