What Speakable Schema Actually Is
Speakable schema is a schema.org vocabulary type that lets you mark specific sections of a webpage as suitable for audio playback by voice assistants. When Google Assistant, Alexa, or an AI search engine reads your page, it doesn't know which sentences are the important ones — it has to guess from context. Speakable markup eliminates that guesswork.
The markup uses either CSS selectors or XPath expressions to point at specific DOM elements and say: these paragraphs are the answer. A voice assistant that supports speakable will prioritize those flagged sections when constructing a spoken response.
Google announced support for speakable in 2019 for news publishers, then quietly expanded the signals it pays attention to as voice and AI-driven search matured. In 2026, with AI Overviews pulling from structured sources and smart speakers handling a growing share of local queries, speakable is no longer a niche publisher tool — it's a practical lever for any business that wants to show up in voice results.
Why Most SMBs Have Never Heard of It (And Why That's an Opportunity)
Run a structured data audit on a typical SMB website. You'll find, at best, LocalBusiness and maybe some FAQ schema. Speakable? Almost never.
There are two reasons for this gap. First, speakable was originally scoped to news and media — Google's documentation launched under "news publisher" guidelines, and the SEO community filed it under "not for us." Second, voice search optimization has historically been treated as a vague future-state thing rather than a tactical priority.
Both assumptions are outdated. Voice and AI search in 2026 aren't niche. When someone asks their phone "what's the best [service] near me" or tells a smart speaker "read me about [topic]," the engine needs a clean, machine-readable signal for what to say. Speakable gives you that signal. Your competitors almost certainly haven't set it up.
The adoption gap is your competitive window. Structured data has a well-documented history: early adopters of FAQ schema, then HowTo schema, captured disproportionate SERP real estate before those markup types became mainstream. Speakable is at an earlier stage of that same curve.
The Technical Anatomy of a Speakable Block
Speakable schema is implemented in JSON-LD (the same format Google recommends for all structured data) and lives in a <script type="application/ld+json"> tag in your page's <head> or <body>.
A minimal working example looks like this:
{
"@context": "https://schema.org/",
"@type": "WebPage",
"name": "How Our Plumbing Service Works",
"speakable": {
"@type": "SpeakableSpecification",
"cssSelector": [".summary-block", ".faq-answer"]
}
}
The cssSelector property points at CSS class names in your HTML. Whatever text lives inside those elements is what the voice assistant will read. You can also use xpath instead of cssSelector if your CMS generates predictable XPath structures.
Three rules for what to flag as speakable:
- Keep it short. Voice assistants read aloud — nobody wants to hear a 400-word paragraph. Flag sections that are 2–4 sentences maximum.
- Make it self-contained. The flagged text should answer a question on its own, without needing surrounding context. "We offer same-day plumbing repairs in Austin, TX. Call before noon for same-day service." is perfect. "As mentioned above, our process involves…" is not.
- Lead with the answer. Voice search is an answer engine. Structure your speakable sections so the first sentence is the direct answer, followed by one sentence of supporting detail.
Speakable, AEO, and GEO: How They Fit Together
Answer Engine Optimization (AEO) is the practice of structuring content so AI systems — not just Google's ten blue links — can surface it as a direct answer. Generative Engine Optimization (GEO) is the parallel effort for large-language-model-based search (Perplexity, ChatGPT Search, Google AI Overviews).
Speakable schema sits at the intersection of both. It's simultaneously a technical signal for voice assistants and a semantic clarity signal for AI-driven engines trying to understand which sentences on your page are the most authoritative and quotable.
Think of it this way:
- FAQ schema tells engines you have a Q&A structure
- HowTo schema tells engines you have a step-by-step process
- Speakable schema tells engines these are the sentences worth quoting out loud
Together, these three schema types create a layered structured data strategy that covers every major AI and voice surface. Implementing all three on a single service page takes about 30 minutes and has no downside.
Where to Use Speakable on an SMB Website
Not every page is a candidate. Speakable performs best on pages that already answer specific questions — which is most of your site, if you've structured it well.
High-priority pages for speakable markup:
- FAQ pages and FAQ sections — Each answer block is a natural speakable unit. Flag the first 2 sentences of each answer.
- Service pages — The opening summary paragraph that explains what you do, who you serve, and where you operate is ideal speakable content.
- About/Contact pages — "We're a family-owned HVAC company serving the Denver metro since 2008. Our office is open Monday through Saturday, 7am–6pm." That's exactly what voice search surfaces for "who is [business name]."
- Blog posts and guides — The TL;DR or key-points section. Flag the summary, not the full body.
- Location pages — For multi-location businesses, the opening paragraph of each location page is a prime speakable target.
What to avoid flagging: Navigation text, calls-to-action ("Click here to book"), legal disclaimers, or any sentence that only makes sense in visual context ("See the chart below").
Speakable Schema and Local Voice Search
Voice queries skew heavily local. "Plumber open near me," "best Italian restaurant downtown," "what time does [business] close" — these are the queries smart speakers and voice assistants field dozens of times a day in every market.
If your LocalBusiness schema already covers your name, address, phone, and hours, you've handled the structured data baseline. Speakable fills the gap that LocalBusiness doesn't: it tells the assistant what to say about you when the query is informational, not just directional.
A real example: a pest control company in Phoenix marks up its service page with a speakable block that reads: "We handle termite inspections and treatments for Phoenix homeowners, with same-week appointments available year-round. Our licensed technicians are state-certified and carry full liability insurance."
When someone asks "what pest control companies in Phoenix offer same-week service," that speakable block is exactly what a voice assistant needs to construct a spoken answer — and it positions that company ahead of competitors whose pages have the same information buried in unstructured prose.
Common Implementation Mistakes
1. Flagging too much text. If your speakable selector captures a 600-word section, you haven't given the assistant a useful answer — you've given it a wall of text. Be surgical.
2. Using speakable on thin or duplicate content. Speakable amplifies whatever it points at. If the content is generic ("We offer high-quality services at competitive prices"), marking it speakable won't help — and it signals low quality to the engine.
3. Inconsistency between speakable content and structured facts. If your speakable text says "open Monday through Friday" but your LocalBusiness schema says you're open weekends too, that's a conflict. Engines notice. Keep your facts consistent across all structured data types.
4. Forgetting to validate. Use Google's Rich Results Test and the Schema Markup Validator after every implementation. A syntax error in your JSON-LD silently nullifies the entire block.
Does Google Actually Use Speakable in 2026?
The honest answer: Google hasn't published granular confirmation of how heavily speakable influences Google Assistant responses, and it's never been a ranking factor for traditional web search. What we do know:
- Google's structured data documentation still actively references speakable and describes it as relevant for voice and audio interfaces.
- Third-party testing consistently shows that pages with speakable markup appear in spoken responses at higher rates than equivalent pages without it.
- AI-driven search engines (including Google's AI Overviews) have demonstrated a preference for pages with dense, unambiguous structured data signals — and speakable is part of that signal set.
The practical stance: There is no credible downside to implementing speakable correctly. The upside — even if it's incremental in 2026 — compounds as voice and AI search traffic grows. It takes one hour to implement on a typical SMB website. That math works.
The Bigger Picture: Structured Data as a Competitive Moat
The businesses winning AI and voice search in 2026 aren't necessarily the ones with the best content. They're the ones whose content is the most machine-readable. Structured data is the translation layer between what you've written and what an AI engine can confidently surface as an answer.
Speakable schema is one piece of that layer — alongside LocalBusiness, FAQ, HowTo, Product, and Review schema. Each type covers a different surface and a different query type. None of them are hard to implement once you understand what they do. Together, they turn a well-written page into a page that machines can confidently cite.
The window for early-mover advantage on speakable is open right now. Most SMBs haven't touched it. The ones that implement it cleanly over the next six months will be the ones their voice assistant recommends when a customer asks the right question.
