Voice search optimization is search engine optimization for spoken questions: making your pages and business listings the answer an assistant or a spoken Google search picks and reads aloud. Spoken queries go through the same index and ranking systems as typed ones, so the work is direct answers, a complete Business Profile and fast mobile pages.

Voice was sold for years on a statistic that never held up, so this guide starts there. Then it covers what voice search means in 2026 (Gemini, Siri AI, Alexa+ and Search Live), how one answer gets chosen, what structured data still does, and the changes that actually help. If you want the whole discipline first, read what SEO is; the step-by-step process is in how to improve SEO.

The "50% of searches by 2020" prediction, traced

You have probably seen the claim that half of all searches would be voice searches by 2020, usually credited to comScore. It started as something narrower. A Fast Company article published on September 17, 2014 reported that about 10% of Baidu's queries were already spoken, and quoted Andrew Ng, then Baidu's chief scientist, forecasting that within five years at least half of all searches would be made by speech or images, if progress continued at its current rate.

Three details fell away as the number travelled:

  • It was about Baidu. Ng was describing China's largest search engine, at a time when many people in China were coming online for the first time, mostly on phones.
  • It counted images too. The forecast was speech and images combined, not voice alone.
  • It had a condition, and no comScore. It depended on 2014's rate of progress continuing. When Econsultancy traced the statistic in 2018, it found the origin in Ng's interview and could not find any comScore source for it.

No search engine has published data showing that it came true, and Google does not separate voice at all in the data it gives site owners: the Search Console Performance report groups data by query, page, country, device, search appearance and date, and its search types are web (text-based or multimodal), image, video and news. Nobody can tell you your voice share, so a plan built on one starts from a guess.

Claim you will still hearWhat the primary sources say
Half of all searches will be voiceA 2014 forecast about Baidu, for speech and images combined, with a condition attached
Voice search has its own ranking factorsGoogle says there are no additional requirements or special optimizations for AI Overviews or AI Mode, where spoken questions in Search Live are answered
You need a page for every long-tail phrasingGoogle's AI optimization guide says its systems understand synonyms, and that pages made for every variation to manipulate rankings breach its scaled content abuse policy
Schema markup gets you read aloudSpeakable is a beta for English news on Google Home devices in the US, and FAQ rich results stopped appearing on May 7, 2026

Tip

When a proposal for "voice search optimization" opens with a share of voice searches, ask for the primary source. If it leads back to 2014, or to nothing, judge the rest of the proposal accordingly.

What voice search means in 2026

The voice assistants of the 2010s were command systems. Google itself describes the old home assistant as something you had to address in rigid, specific commands, a transactional tool. Since 2024 Google, Apple and Amazon have each rebuilt their assistants on large language models that keep context between questions, handle follow-ups and fetch live information. Each change below is dated from the vendor's own announcement.

AssistantMilestone, from the vendorHow it answers, per the vendor
Gemini on phonesMarch 14, 2025: Google begins moving mobile users from Google Assistant to Gemini and says the classic Assistant will no longer be accessible on most mobile devices later in 2025Free-flowing, multimodal conversation in Gemini Live, plus Assistant tasks such as timers
Gemini for HomeOctober 1, 2025: Gemini replaces Google Assistant on smart speakers and displays, through early accessKeeps context across follow-ups and uses real-time information
Gemini in cars with Google built-inApril 30, 2026: rollout starts in English in the US, for new and existing carsFinds places, such as a lunch stop on the route, with information from Google Maps
Siri with Apple IntelligenceOctober 28, 2024: iOS 18.1, with the device and Siri set to US EnglishMore natural language, typed or spoken, with context kept from one request to the next
Siri AIIntroduced June 8, 2026; beta in English since September 14, 2026, not at first in the EU on iPhone, iPad and Apple WatchGoes out to the web for up-to-date answers, and searches the user's messages, emails and photos
Alexa+Announced February 26, 2025; available to everyone in the US, free with PrimeLarge language models from Amazon Nova and Anthropic, with partner services such as OpenTable, Vagaro and Thumbtack

Two things follow for a website. First, many spoken requests never touch a web page: Alexa+ makes reservations through OpenTable, Amazon's own example has it arranging an oven repair through Thumbtack, Siri AI answers from the user's own email, and Gemini in a car picks a restaurant from Maps data. If you take bookings, the platforms you are listed on matter as much as your site. Second, when a question does need the web, the assistant generates its answer from the sources it retrieves, so the page that states the answer most clearly is the one that gets used. Our guide to AI assistants in 2026 compares what each one can do, and the post on AI smart home devices covers speakers and displays.

Spoken queries in Google Search and Search Live

In Google Search itself, voice is an input method. Tapping the microphone in the Google app runs an ordinary Google search with your spoken words, as Google's help page on voice search describes. More than half of the examples on that page are quick facts: the time in London, whether you need a jacket, 16 ounces in pounds. No website should build content to win those. The rest are where websites and businesses come in: the closest coffee shop, attractions for a trip, a table for two on Wednesday night.

Spoken questions are longer and phrased as sentences, and Google has been adapting to that for years. When it introduced BERT in October 2019, Google said it still sometimes got complex or conversational queries wrong, which is why people typed "keyword-ese", and that BERT helped it understand longer queries where small words like "for" and "to" change the meaning. The practical upshot is that you do not need to match a spoken question word for word; you need to answer it.

The bigger change is Search Live. Since September 24, 2025, people in the US have been able to hold a spoken conversation with AI Mode in English, without signing up for Labs, and share the phone's camera for context. On March 26, 2026 it expanded to every language and location where AI Mode is available, more than 200 countries and territories. You tap Live under the search bar in the Google app, ask out loud and get an audio answer, then ask follow-ups or open the web links shown with it. The links are the part a website can win, and because Search Live runs inside AI Mode, they come from the same process as a typed AI Mode answer; our post on AI Overviews and AI Mode explains that selection in detail.

How one answer gets chosen

A results page shows ten links; a spoken reply is one answer, with links only if there is a screen. That makes the selection matter more, but it is the same selection. Google says its generative AI features are rooted in its core Search ranking and quality systems: the model issues several related searches at once (query fan-out), retrieves pages from the Search index and writes a response grounded in them, with links to the pages that support it. To be one of those pages, a page must be indexed and eligible to be shown with a snippet, and your site must not be excluded from generative AI features in Search Console.

A laptop screen lists ten grey result bars. One bar is lifted out and becomes an orange answer card above a smart speaker, while a phone below shows two links.
Fig. 1 A screen shows ten results and a speaker reads one: voice raises the stakes of ranking first without changing how ranking works.

Featured snippets play the same part in classic results. Google's help page says they appear when it thinks the answer can be found in a short piece of a website, and that they are helpful when you talk to your device. You cannot mark a page as a featured snippet: Google's systems decide whether a page would make a good one. What you control is whether a page may be used at all. A nosnippet rule blocks featured snippets along with regular snippets, and a low max-snippet value makes a featured snippet less likely, so neither belongs on a page you want read aloud.

Google matches meaning rather than exact wording, which is what semantic SEO builds on, and the whole pipeline from crawling to ranking is explained in how Google Search works. For voice, the conclusion is short: the page that ranks as the best answer to a typed question is the page most likely to be spoken.

Local voice searches: "near me" and your Business Profile

Local questions are where voice pays most directly, because the person asking usually wants to go somewhere or call someone. Google says local results are based mainly on relevance, distance and prominence: how well a Business Profile matches the search, how far the business is from the searcher (when they do not share a location, Google uses what it knows about it), and how well known the business is, including links to it and its reviews. There is no way to request or pay for a better local ranking.

A phone with sound waves sits inside dashed distance rings on a simple map. Grey pins surround it, and one orange pin carries a small card with a clock, a phone handset, a star and a car.
Fig. 2 Relevance, distance and prominence choose the business; the details in its profile answer the follow-up question.

Spoken follow-ups make the profile's details count. In cars with Google built-in, Gemini finds restaurants along a route with information from Google Maps, then answers follow-ups such as what the parking is like or whether there are vegetarian options. Several of those details are ones the owner can supply in the Business Profile. Google's local ranking page asks for:

  1. A verified profile, so Google knows you are authorized to represent the business.
  2. The right category, which tells Google what kind of business you are.
  3. Hours that are current, including special hours.
  4. Attributes such as parking or Wi-Fi, the details a follow-up question asks about.
  5. Replies to reviews, photos and, for retailers in eligible countries, in-store products.

Google's AI optimization guide adds that Business Profiles help a business appear in AI responses as well as in other results. If many of your customers use iPhones, claim your place on Apple Maps through Apple Business as well. Ranking in local results, citations and reviews are covered in our local SEO guide, setting up the profile in Google Business Profile optimization, and a plan for owners short on time in SEO for small business.

What structured data does for voice in 2026

Most voice advice is out of date on markup. Speakable, the markup Google documents for reading content aloud, is still a beta. It lets Google Assistant answer topical news questions on smart speakers by reading marked sections of up to three articles aloud and sending the full article link to the user's phone. It works only for users in the US with Google Home devices set to English, and for publishers of English-language content. The page still describes Google Assistant and does not mention Gemini for Home, which has been replacing Assistant on Google's speakers since October 2025.

Several other features that voice guides used to recommend are gone, according to Google's documentation changelog and its developer notice:

FeatureWhat it didStatus as of September 2026
Speakable markupMarked article sections for text-to-speech on Google HomeBeta; US users with devices in English, English-language news publishers
FAQ rich resultsQuestions and answers shown under a search resultLimited to government and health sites in September 2023; no longer shown from May 7, 2026; documentation removed in June 2026
HowTo rich resultsSteps shown in search resultsNo longer shown; documentation removed in September 2023
Guided recipesGoogle Assistant reading recipe steps aloudRemoved from the recipe documentation in March 2024, because the Assistant feature was removed
Conversational ActionsCustom voice apps for Google AssistantShut down on June 13, 2023

Important

FAQPage markup no longer produces anything in Google Search. Keep a genuine FAQ section if it helps readers, but do not expect the markup to get you a rich result or a spoken answer.

What still holds is Google's general line: structured data is not required for generative AI search, there is no special schema.org markup for it, and it remains worth using because it makes pages eligible for rich results. Mark up what a page actually is, such as an organization, an article or a product, and keep it consistent with what the page shows.

If you publish news in English for a US audience, speakable costs little. Google recommends marking key points rather than whole articles, about 20 to 30 seconds of content per section (two or three sentences), and leaving out datelines, photo captions and source attributions. Point the markup at those sections with CSS selectors or XPath, not both:

{
  "@context": "https://schema.org",
  "@type": "WebPage",
  "name": "Council approves the new cycle lane",
  "speakable": {
    "@type": "SpeakableSpecification",
    "cssSelector": [".headline", ".summary"]
  },
  "url": "https://www.example.com/news/cycle-lane"
}

How to optimize for voice search: what actually helps

Everything that follows is ordinary SEO, pointed at questions. It serves typed search, AI Mode, Search Live and assistants at the same time, which is why it is worth doing even though nobody can measure voice on its own.

Answer the question in the first sentence

Put the answer directly under the heading that asks the question, in one or two plain sentences, then give the detail. A reader can stop there, and an assistant can read it without the rest of the page. Google's advice for text meant to be spoken, written for speakable, is a good yardstick for any page: two or three sentences, 20 to 30 seconds of speech, that make sense on their own. For a repair company, under "Do you repair dishwashers on weekends?", the first sentence should say yes or no and when, not open with how complex modern dishwashers are.

A web page drawn as grey bars has a heading and an orange first line under it. A dashed arrow lifts that line into a speech bubble above a smart speaker; the rest of the page stays grey.
Fig. 3 The sentence under the heading is the one that gets read aloud, so it has to make sense on its own.

Write headings the way people ask

Spoken questions are full sentences, so write section headings in your customers' words: "How much does a boiler service cost?", "Do you deliver on Sundays?". One question per section keeps each answer easy to lift out. Google's own guidance favours pages organized by paragraphs and sections with clear headings, and warns against the old voice tactic of a separate page for every phrasing: its systems understand synonyms, and pages created to cover every variation mainly to manipulate rankings breach its scaled content abuse policy. Titles, headings and the rest of the page's elements are covered in our on-page SEO guide.

Keep your Business Profile complete

For a local business, a spoken question about nearby places is answered from Maps and Business Profile data, as the car example above shows, so the profile deserves as much care as the website. Verify it, keep hours and attributes current, and answer reviews, as in the local section above.

Make the page fast on a phone

The person who taps a link under a spoken answer is on a phone. Google indexes and ranks the mobile version of your content (mobile-first indexing) and recommends responsive design. Its Core Web Vitals targets are Largest Contentful Paint within 2.5 seconds, Interaction to Next Paint under 200 milliseconds and Cumulative Layout Shift under 0.1. Our guide to mobile optimization covers the fixes.

Keep the SEO fundamentals in order

A page that is not crawled and indexed cannot be anyone's answer. Technical SEO covers crawling, rendering and indexing; our 14-point SEO checklist orders the basics by impact; SEO marketing covers budgets and measurement as a channel; and SEO vs SEM explains where paid search fits.

How to measure voice search traffic

No report separates voice for you, so measure the kinds of questions voice produces. In the Search Console Performance report, add a query filter, choose Custom (regex) and use a pattern that matches question openers. The filter uses RE2 syntax, matches anywhere in the query unless you anchor it, and ignores case by default:

^(who|what|when|where|why|how|which|can|do|does|is|are|should)\s

A second filter such as near me|nearby|open now isolates local intent. Compare clicks and impressions for those groups over time, alongside the Generative AI performance report that Google's AI optimization guide points to for AI Mode and AI Overviews. A rising number of question queries you rank for is a better signal than any voice share, because it measures the thing voice rewards: answers.

If you want help turning this into work on your own site, our SEO service includes local search (Google Business Profile, consistent name, address and phone, and location and service-area pages) and tracks how often AI assistants cite you when they answer questions in your field.