Voice search

Ask a speaker on the kitchen counter "what time does the hardware store close" and you do not get ten links. You get one sentence. Voice search turns a page of results into a single spoken answer, and that changes what it takes to be the source.

How does voice search work?

speech → transcription → query or AI model → one answer → text-to-speech

The device turns speech into text. A search engine or an AI model then answers it, often by searching the web and taking a passage from one page. The answer is read aloud, sometimes naming the source. The person hears one answer; there is no second result to fall back on.

Two kinds of voice search exist now. Classic voice assistants on phones and speakers answer short factual and local questions, often from featured snippets and business listings. AI assistants with voice modes answer longer questions the same way they answer typed ones: by reasoning, searching and writing, then speaking the result.

What is voice search optimization?

Making a page the one a voice answer can be read from. It is mostly the same work as winning a featured snippet or an AI answer, with one extra test: does it still make sense when heard?

• Put the question in a heading, worded the way people say it.
• Answer in two or three sentences directly under it, in plain words. A spoken answer cannot point at a table, a chart or "the list below".
• Name things fully. "It" and "this tool" mean nothing to a listener who did not see the heading.
• Keep local facts current for local questions: hours, address, phone, in the business profile and on the page.

What does the speakable markup do?

Google has one piece of structured data for voice: speakable. It marks the parts of a page suited to text-to-speech. It is in beta and narrow: the Google Assistant uses it to read up to three news articles aloud on smart speakers, for English-language news publishers in the US. Google suggests about 20 to 30 seconds of content per section, roughly two or three sentences.

This page carries it, as every glossary page on attensira.com does: a SpeakableSpecification inside the page's WebPage markup, with two CSS selectors, one for the heading and one for the answer block at the top. The answer block is written to the length Google recommends for a spoken section. We are not a news publisher, so Google's Assistant does not use this markup for our pages, and we do not claim it does. What carries over is the discipline: a short, self-contained answer at the top of every page, which is exactly what a voice answer, a featured snippet and an AI answer all lift.

How is voice different from typed search for a brand?

• Questions are longer and more natural. "What's a good tool to see if ChatGPT mentions my company" rather than "AI visibility tool".
• There is one answer. In a typed result you can be third and still be seen. In a spoken answer, only the source that was read counts.
• The source may not be named. Some voice answers say where they came from, many do not, so brand mentions inside the answer matter.
• AI voice modes blur the line. A spoken question to ChatGPT gets the answer ChatGPT would have typed, so your visibility in typed AI answers is a fair proxy for the spoken ones.

What do people get wrong about voice search?

• Treating it as a separate channel with its own tricks. It runs on the same retrieval as typed search and AI answers.
• Writing for keywords instead of spoken questions. People ask voice assistants in full sentences.
• Expecting speakable markup to work for any site. It is documented for news only.
• Answers that depend on visuals. If the answer needs a table to make sense, it cannot be read aloud.

How does Attensira help?

Attensira tracks the questions buyers ask in their own words, the conversational phrasing voice search produces, and reads how every AI engine on your plan answers them each day. When a competitor's passage is the one being chosen, the crew drafts a short, self-contained answer for your page, for your approval. See prompt tracking for how those questions are picked.