Voice Search

Voice search lets users query Google by speaking instead of typing, through the microphone in the Google app on Android and iOS, the assistant on phones and smart speakers, and built-in assistants in cars and televisions.¹ For SEO the difference is the answer. On a screen device a spoken query returns a standard results page while the assistant also replies aloud, but on a screenless device such as a smart speaker only the spoken reply is delivered, so the answer Google picks is the only one the user hears. That source used to be a single result read aloud; as Gemini replaces Google Assistant, it is increasingly an answer composed from several sources.

How does Google generate spoken answers?

Where the legacy Google Assistant model still operates, Google sources spoken answers predominantly from featured snippets, reading the snippet text aloud when one is present; where no snippet exists, the answer typically comes from a top-ranking organic result. The selection criteria are the same in either case: a passage that directly answers the query, reads naturally as a spoken sentence, and does not depend on visual formatting to make sense.

Gemini is replacing Google Assistant on different timelines: on phones and tablets, and on paired devices such as Wear OS watches, compatible earbuds and projected Android Auto, Google began removing access on 4 September 2026, a process it said might take a few weeks to reach everyone, while speakers, smart displays, televisions and cars with Google built-in transition separately.² Gemini for Home, which began shipping on Google’s speakers in June 2026, reasons through multi-step questions and holds conversational context rather than reading a single answer.¹

As Gemini replaces Assistant on these devices, spoken answers are increasingly generated conversationally, synthesising across sources like an AI Overview. Either way, a voice user receives an answer and either accepts it or rephrases, with no scrolling past position one, so being the source Google draws on is what matters.

Query characteristics

Voice queries differ from typed queries in three consistent ways.

Length. Spoken queries are longer and more complete than typed ones. Users speak naturally rather than abbreviating for a keyboard, so queries tend to include the full question.

Phrasing. Voice queries are almost always phrased as questions: “what”, “how”, “where”, “when”, “who”. Typed queries often drop the question word entirely (“best running shoes 2026” versus “what are the best running shoes to buy?”).

Local intent. A significant proportion of voice queries have local intent: “where is the nearest…”, “what time does… open”, “directions to…”. Voice search on mobile is closely tied to navigation and immediate-need queries. This makes it a meaningful channel for any business with a physical location.

Optimisation

Target the featured snippet

The featured snippet is the primary target for voice search visibility. To target it:

• Answer the question directly in the opening sentence or paragraph, before elaborating
• Keep the direct answer concise: one or two sentences that state the answer plainly
• Structure the page so the question appears as a heading and the answer follows immediately beneath it
• Use numbered lists for step-by-step answers, which Google can read as a sequence

Write for spoken delivery

Content written for voice needs to read naturally as a sentence. Bullet points, parenthetical asides, and tables do not translate to spoken answers. A paragraph that begins “To change a tyre, first ensure the car is on level ground” reads aloud sensibly. A table comparing products does not.

Avoid nominalisations and passive constructions. “Google reads the featured snippet aloud” is clearer spoken than “the featured snippet is read aloud by Google’s voice assistant”.

Local voice search

For businesses with physical locations, local voice queries are a significant opportunity. Optimise for them by:

• Keeping the Google Business Profile complete and accurate: address, opening hours, phone number, and primary category
• Including consistent name, address, and phone number in LocalBusiness schema
• Writing content that addresses “near me” and “open now” intent directly, such as a page section covering hours, parking, or directions

Schema markup

Google says it uses structured data “to understand the content of the page”,³ and documents no role for it in choosing a spoken answer other than Speakable, below. The types usually discussed for voice:

• FAQPage: structured question-and-answer content; no longer produces FAQ rich results in Google Search, and no platform documents reading it for spoken or AI answers
• HowTo: step-by-step instructions
• LocalBusiness: address, hours, and contact details
• SpeakableSpecification: marks specific sections as suitable for audio playback; aimed at news content for audio readouts (its documentation still references Google Assistant), and still marked as beta by Google, with narrow adoption⁴

Page speed

Page speed contributes to ranking, and featured snippet eligibility correlates with it. A page that loads slowly is less likely to rank in the positions where featured snippets are won, and Google can verify fast pages are reliably retrievable. Core Web Vitals, which Google’s ranking systems use, are the relevant benchmark.⁵

Voice search and AI Overviews

Voice search and AI-generated answers are converging. Google AI Mode and AI Overviews handle many of the same conversational, question-format queries that voice search has always dominated. A page optimised for voice, with a direct answer first, clear structure, and credible sourcing, is also well-positioned for AI citation.

The key difference is scope. AI Overviews can synthesise from multiple sources, while a voice answer is read from one. A page does not need to be the only good source on a topic to appear in an AI Overview, and where a single answer is still read aloud, holding the featured snippet is what earns it.

Measurement

Google Search Console does not separately report voice search queries. Clicks and impressions from voice searches appear alongside typed searches in the Performance report, with no voice-specific attribution.

You can approximate voice query exposure by filtering for queries that are:

• Five or more words
• Phrased as questions (beginning with “what”, “how”, “where”, “when”, “who”, “can”)
• Combined with featured snippet position data

This does not give exact voice traffic figures, but it identifies the pages and queries where voice search visibility is plausible.

Footnotes

1. The Google Home Speaker, with Gemini for Home — Google ↩ ↩²

2. Google Assistant to be discontinued on Android starting September 4 — Search Engine Land ↩

3. Introduction to structured data markup in Google Search — Google Search Central ↩

4. Speakable (Article, WebPage) structured data (BETA) — Google Search Central ↩

5. Understanding Core Web Vitals and Google search results — Google Search Central ↩