Speakable Schema: The 99.4% Gap Hiding in Every Site's AEO Rank

Of the 13,385 sites we have scored on AEO Rank, 13,274 fail the Speakable Schema criterion outright. That is 99.4% of the public web we have audited. The median score on the criterion is zero out of ten. The mean is 0.1. It is, by a wide margin, the most ignored signal in the entire AEO Rank framework, and that makes it one of the cheapest wins still on the table for any team that wants to surface in voice assistants and AI answer panels.

What Speakable Schema actually is

Speakable Schema is a schema.org annotation - formally SpeakableSpecification - that marks specific sentences or sections of a page as suitable for text-to-speech rendering. Google’s voice features and Assistant-class products read these annotated sections aloud when a user asks a query the page can answer. More importantly for our purposes, AI assistants that ingest structured data use Speakable as a strong “this is the quotable summary” signal, not unlike how description works for OpenGraph but at the paragraph level inside the article body.

A valid Speakable annotation lives inside an Article, NewsArticle, or WebPage JSON-LD block. It contains either a CSS selector or an XPath expression pointing at the sentences you want read aloud. Most well-built sites already have a hero answer paragraph or a “short answer” block at the top of each article - the engineering work is wiring a Speakable selector to that block. It is rarely more than a few lines of code per template.

The corpus numbers, and why they are this bad

Across all 13,385 unique domains in our corpus, only 111 sites score above zero on Speakable Schema. Inside that small cohort, scores are still mostly low: the criterion has a 10-point ceiling but the corpus mean is 0.1, which means even the sites that have Speakable annotations often have them on a single template rather than across the article body, FAQ, and recipe pages where AI engines look first.

Criterion	Pillar	Median (0-10)	Mean	Critical-fail count	Critical-fail share
Speakable Schema	Technical Foundation	0	0.1	13,274	99.4%
Content Licensing & AI Permissions	AI Discovery	0	0.7	12,594	94.3%
llms.txt File	AI Discovery	0	1.8	10,982	82.2%
Owned Data Density	Trust & Authority	0	1.2	10,316	77.5%
RSS/Atom Feed	AI Discovery	0	2.1	10,311	77.5%

Why is the gap this wide? Two reasons. First, Speakable Schema is genuinely under-publicized. Schema.org documents it, Google documents it for news publishers, but it is rarely on the front page of an SEO or AEO checklist. Second, most CMS templates do not ship it by default. WordPress, Webflow, Astro, and most headless setups expose JSON-LD for Article and FAQPage out of the box, but not for Speakable. Teams that have never been told to look for it never add it.

Why AI engines weight Speakable harder than its visibility suggests

When ChatGPT, Claude, or Perplexity needs to cite a page, it has to pick a span of text to quote. The shorter and more self-contained that span is, the higher the chance of citation - long, meandering paragraphs are riskier to quote and less useful for the user. Speakable Schema explicitly hands the engine a pre-selected, author-approved span. It is the difference between asking an engine to guess what to quote from a 3,000-word article and handing it the two sentences you want quoted.

We have observed this empirically across the small handful of top-tier sites in the corpus. Among the 32 domains scoring 70 or higher, Speakable Schema presence correlates with sustained scores above 75 — the criterion is one of the cheapest pillar moves available to a site already in the 60s, and the corpus data shows top-tier sites have almost universally adopted it where the rest of the corpus has not.

How to ship Speakable Schema in under an hour

The implementation is mechanical. In your Article or WebPage JSON-LD, add a speakable property containing one or more SpeakableSpecification objects. Each object holds a cssSelector or xpath array. Point those selectors at your hero answer paragraph, your “short answer” block if you have one, and your FAQ answers. Keep the spans short - one to three sentences. Verify the markup with Google’s Rich Results Test. Re-audit the page to confirm the AEO Rank Speakable criterion moves off zero.

The only mistake we see in the wild is teams adding Speakable selectors that point at containers (like the entire <article> element) rather than at specific sentences. That defeats the purpose. The whole value of Speakable is that you are telling the engine “this sentence, not the others.” If you are pointing at the whole article you are telling it nothing.

How We Tested

The 99.4% failure rate is computed directly from the AEO Rank engine output across all 13,385 unique domains scored as of 2026-05-29. We define a “critical fail” on Speakable Schema as a criterion score of zero, which the engine assigns when no speakable property is present in any JSON-LD block on the page or when the property is present but malformed (invalid selectors, missing required fields).

The corpus is built from the aeo_audit_versions table joined to the canonical aeo_domains row using the most recent published audit per domain. We do not filter by engine - the criterion is computed identically across the instant, ChatGPT, and Claude engines because it is a structural HTML/JSON-LD check, not an LLM judgment.

The corpus comparison cohort (32 sites scoring 70 or higher) was identified by ordering aeo_published_audits by overall_score descending and taking every row with overall_score >= 70. Speakable adoption inside that cohort was checked by parsing the most-recent published HTML on each domain for any JSON-LD block containing a speakable property.

The cheapest 10 points on AEO Rank

If you score below 50 on AEO Rank today and you are not running Speakable Schema, you are leaving the easiest available points on the table. The engineering work is small, the documentation is mature, and the corpus tells us 99.4% of your competitors are not doing it either - so the relative gain is even larger than the raw point gain. Read the AEO Rank methodology for how Speakable fits into the Technical Foundation pillar, or run a site audit to see the exact selector recommendations for your templates.