Does Speakable Schema Help Content Get Cited by LLMs?

What is speakable schema?

Speakable is a schema.org property, speakable, with a SpeakableSpecification type. It points to the parts of a page best suited to being read aloud, using CSS selectors or XPath. Google's documentation describes it as a beta feature. Google Assistant uses it to answer topical news queries on smart speakers. When a user asks for news about a topic, the Assistant returns up to three articles and reads the marked sections aloud, attributing the source. It then sends the article link to the user's phone.

The documented limits are narrow:

• Content: news articles, from publishers eligible for Google's news results.
• Devices: Google Home devices set to English.
• Language: publishers that publish in English.
• Length: Google recommends about 20 to 30 seconds of audio per section, roughly two or three sentences.

Google has since begun replacing Assistant with Gemini on its Home speakers. Its speakable documentation does not say whether Gemini uses the markup.

Does SpeakableSpecification schema improve AI visibility?

Not in any way that has been shown. Consider each route by which it could:

• Google's AI Overviews and AI Mode. Google says its AI features need no special markup or files, only that a page is indexed and eligible for a snippet. Speakable is not mentioned as an input.
• ChatGPT, Claude, Perplexity and Gemini fetching pages. SearchVIU tested these systems in October 2025 and found that when they fetched a page directly, they extracted the visible text and ignored JSON-LD entirely. A speakable block inside JSON-LD would never be read.
• Search indexes that feed AI answers. Bing and Google parse structured data for their own features. Microsoft has said schema helps its language models understand content. But neither has said speakable affects which passages are cited.
• Measured outcomes. Ahrefs compared 1,885 pages that added structured data with matched pages that did not, from August 2025 to March 2026. ChatGPT and AI Mode citations moved by about 2%, which it judged indistinguishable from zero. AI Overview citations fell slightly more for the pages that added markup. The study pooled several schema types, not speakable alone. It is still the best controlled evidence available.

How schema problems can block citations, rather than earn them, is covered in how indexing, canonical and schema issues prevent LLM citations.

Why do some analyses find schema pages cited more?

Several analyses report that pages with FAQ, speakable or other schema are cited more often. These are usually correlations, and they mislead for a simple reason. Sites that add FAQ or speakable markup tend to be sites that already write in question-and-answer form, with short answers under clear headings. They also tend to be larger, better maintained and technically sound. Any of those traits could explain the citations. The markup travels with them.

To separate cause from correlation, a study has to compare similar pages with and without the markup, over the same period, as Ahrefs did. When that is done, the effect largely disappears. If someone cites a figure such as "FAQ schema pages get 35% more citations", ask whether the comparison controlled for page format and site quality.

What about FAQPage and QAPage?

The same logic applies, with some differences in Google's support:

• FAQPage. Google restricted FAQ rich results to government and health sites in August 2023 and stopped showing them for all sites in May 2026. The markup remains a valid schema.org type and does no harm, but it no longer produces a visible feature in Google.
• QAPage. Google's QAPage markup is meant for pages where one question is answered by users, such as forums and community answer sites. It is not meant for a company's own FAQ. Using it on a page the business wrote misrepresents the content.

For both, the visible question-and-answer structure is what AI systems can read and quote.

What actually helps, then?

Write the kind of passages speakable was designed for, and make sure they are in the visible HTML:

• Put a two- or three-sentence answer directly under each question heading. It should make sense read aloud, without the rest of the page.
• Name the subject in that answer rather than relying on "it" or "this".
• State the key fact first: the number, the date, the verdict.
• Keep the answers current, and remove outdated ones rather than leaving them.
• Use headings phrased as the questions people ask, so a passage can be matched to a query.

Search systems split pages into passages and retrieve the ones that best answer a question. A passage that stands alone is easier to retrieve and quote accurately, as covered in how content chunking affects what LLMs retrieve and recommend.

When is speakable still worth adding?

• You publish news in English and want Google's voice features to be able to read your articles aloud.
• Your CMS can add it automatically for the summary or standfirst of each article, at no ongoing cost.
• You keep the selectors accurate. If a redesign changes class names, speakable markup can end up pointing at nothing, or at the wrong text.

For most businesses, the markup is neither harmful nor useful. Time spent on it is better spent on the visible answers themselves.

How can you test whether markup matters on your site?

1. Choose two groups of similar pages, matched on topic, traffic and format.
2. Add the markup to one group only, and change nothing else.
3. Track citations and mentions for the questions those pages answer, across several assistants, for at least two months.
4. Compare the change in each group, not the before-and-after for one group alone, since AI answers shift for many reasons.

If you cannot hold other changes constant, the test will not tell you much.

What might an illustrative case look like?

Consider an illustrative health information publisher that added speakable and FAQPage markup to 200 articles after reading that schema raised citation rates. Three months later, citations for those articles were up about 10%. The team then noticed that a control set of 200 unmarked articles had risen by a similar amount, because a broader content refresh had happened at the same time. The markup had not made a measurable difference. The team kept the markup, since it did no harm. It turned its attention to rewriting each article's opening answer into two plain sentences, which it could test the same way.

What do people ask about speakable and AI citations?

Does Gemini use speakable markup?

Google has not said so. Its speakable documentation describes Google Assistant on Home devices, not Gemini.

Can speakable hurt my site?

No, if it points to accurate text on the page. Markup that points to missing or misleading content is a problem, as with any structured data.

Is speakable only for news?

Google's feature is for news articles. The schema.org property can be used on other pages, but Google does not document any use for them.

What should I prioritise instead?

Clear, visible answers under question headings, accurate facts, and pages that search engines can index.