Voice assistants and AI answer engines don't read your page the way a person does — they extract a short, self-contained passage and speak it aloud or surface it as the answer. Speakable schema is the structured-data markup that tells these systems exactly which sentences on your page are suitable to be read out. If you want Google Assistant, smart speakers, and voice-driven AI to quote your content instead of a competitor's, you need content that is written to be spoken and, where supported, marked up so machines can find the speakable part instantly.
What speakable content actually is
Speakable content is any passage that fully answers a question on its own, in plain language, in about one or two sentences. When a user asks a voice assistant "what is the return policy for X" or "how long does concrete take to cure," the assistant wants a crisp, quotable answer it can speak without extra context. Pages that bury the answer inside long, winding paragraphs give the assistant nothing clean to lift, so it moves on to a source that does.
Schema.org defines a SpeakableSpecification that points to the sections of a page best suited for text-to-speech, identified by CSS selectors or xpath. It is applied within an Article or WebPage. Support has been limited and largely tied to news, so treat the markup as a bonus — the real, durable win is writing content that is inherently speakable. If you are new to structured data, our guide to JSON-LD and Schema.org covers the syntax you'll build on.
How to write for voice and spoken answers
The techniques that win voice answers overlap heavily with answer engine optimization. The core move is to answer the question directly in the first sentence, then elaborate.
- Lead with the answer. Put a complete, standalone response in the opening line of the relevant section, before any background.
- Use natural question phrasing. Voice queries are conversational and longer than typed ones — "how do I" and "what is the best way to" — so mirror that language in your headings.
- Keep sentences short and unambiguous. A sentence that reads well aloud has one idea, no nested clauses, and no "as mentioned above" references that make no sense out of context.
- Define terms in place. A spoken answer can't rely on a diagram or the paragraph before it, so each quotable sentence must stand alone.
These same habits are what make you quotable across every answer engine, not just voice. Our broader playbook on AEO and GEO explains why writing self-contained answers is the foundation of getting picked by AI.
Marking up the speakable sections
Where you want to add SpeakableSpecification, wrap your quotable answer in an element with a stable CSS class or ID, then reference that selector in your JSON-LD. Keep the marked region genuinely short — the point is to flag one or two sentences, not a whole article. Marking up an entire page as speakable defeats the purpose and produces awkward, over-long spoken output.
Because speakable markup is niche and its support shifts, validate whatever you add and don't rely on it alone. Run it through the tools in our structured data validation guide to confirm the syntax is clean. The markup is the accelerant; the writing is the engine.
Why this matters even without wide markup support
Even if you never add a single SpeakableSpecification, writing speakable content pays off everywhere. Retrieval systems favor short, self-contained passages, which is exactly what a spoken answer is — and that logic drives how ChatGPT chooses which sites to cite. The pages that get read aloud by a smart speaker are usually the same pages that get quoted by text-based AI assistants, because both are hunting for the cleanest answer to the question.
Treat "could an assistant read this sentence aloud and have it make sense" as a quality test for your whole site. If the answer is yes, you are optimized for voice, featured snippets, and AI citation at the same time.
Curious whether AI systems can extract clean, quotable answers from your pages? Run the CheckMy.site scanner to see how bots read your content and structured data — and where your answers are too buried to be spoken or cited.