What Speakable actually is
Speakable is a JSON-LD specification that names specific CSS selectors on a page as answer-ready passages. When a downstream system needs a short, quotable answer — a voice assistant reading a search result aloud, an AI Overview picking a passage to render, an LLM retrieval step picking a snippet to ground on — it can read the Speakable specification, pull the text from the named selectors, and treat that text as the extract-ready answer.
The mechanism is intentionally simple. It does not rewrite content. It does not change how the page renders. It hands the extraction system a pointer to the passage the page author already marked as answer-worthy, so the system does not have to guess which paragraph out of the full page is the actual answer.
{
"@context": "https://schema.org",
"@type": "WebPage",
"speakable": {
"@type": "SpeakableSpecification",
"cssSelector": [".tldr-block", ".faq-a"]
}
}
Why it disproportionately helps healthcare
Two features of healthcare research work in Speakable's favour. First, patient queries are overwhelmingly short-answer queries — "is IVF painful", "how long does LASIK recovery take", "what causes melasma", "does hair transplant hurt". Each of those has a short, self-contained answer of one to three sentences. Second, the answer surfaces where those queries are resolved — voice assistants, AI Overview panels, chatbot answers, smart-speaker responses — all favour passages that can be lifted whole rather than paragraphs that only make sense in longer context.
Speakable is the mechanism that says "this is the paragraph, extract it as-is". A healthcare brand with 40 well-marked Speakable passages across its FAQ blocks and TL;DR sections is materially easier for a voice assistant, an AI Overview or an LLM retrieval step to extract cleanly than a brand whose answer-worthy paragraphs are buried inside 800-word explainers with no markup at all.
The extraction impact is not the only benefit. Because Speakable makes the extracted passage more predictable, the passage the system quotes is more likely to be exactly the passage the brand wanted quoted. That matters in a regulated industry where a misquoted claim can trigger an ASCI or NMC flag. Speakable is, in that sense, both a citation-earning discipline and a compliance-hygiene discipline — the same short, careful passage that earns the citation is also the passage the brand is most confident to have quoted.
Where to deploy Speakable, and where not to
Speakable belongs on passages that are already answer-ready — short, self-contained, compliance-checked. It does not belong on long-form prose, on promotional paragraphs, or on passages that only make sense inside surrounding narrative. The four page patterns where Speakable pays back its deploy cost inside 60 days:
- Treatment-explainer money pages with FAQ sections — the highest-ROI single application.
- TL;DR blocks at the top of long-form pillars — a Speakable-marked TL;DR often becomes the AI Overview citation on the query the pillar covers.
- Condition explainers with a "quick answer" paragraph and a longer detailed section — Speakable pointed at the quick-answer selector.
- Doctor bio pages with a one-line "specialty and focus" summary — Speakable pointed at the summary line for named-expert voice-answer citation.
Where not to deploy Speakable: on generic paragraph tags, on marketing pitch prose, on passages that carry any unqualified claim ("100% success rate", "best in India") because those will be quoted verbatim and the compliance liability is direct.
Common Speakable failure modes
- Selectors pointed at generic tags.
cssSelector: ["p"] is worse than no Speakable at all; the extraction system has to guess which paragraph, and often guesses wrong.
- Speakable-marked passages that are too long. 800 words of Speakable content is not extraction-ready. Working ceiling is under 300 characters per passage.
- Speakable pointed at content that is not on the visible page. Structured-data spam risk. The selector must actually point at visible content the user sees.
- Speakable deployed once on the homepage only. Speakable is a per-page schema. Deploy it via the component that renders FAQ and TL;DR blocks, so every page inherits it automatically.
- Speakable content that fails compliance verbatim quote-out. Every Speakable-marked passage must survive being read aloud out of context. Anything less is a liability, not a feature.
This page is Speakable-marked, deliberately
The Speakable specification at the top of this page points at the lead paragraph, the H2 headings, and the FAQ selectors. That is not an accident. A primer on Speakable that fails to deploy Speakable itself is not credible. Every ICG money page and insight page carries Speakable pointed at its answer-ready selectors, tied to the component that renders those passages, so every new page ships with the markup already wired.
