When you type a question, you get a page of options. When you ask out loud, you usually get one answer. That single difference is why voice search matters more than its share of queries suggests: if an assistant reads out a competitor's name, you were never in the running.
What is voice search in 2026?
Voice search in 2026 is any spoken question answered by an AI assistant, usually with one short spoken answer rather than a list of links. It now covers phones, cars, earbuds, watches and speakers, not just the smart speaker on the kitchen worktop.
The assistants themselves have changed underneath us. Google announced in March 2025 that it was upgrading Google Assistant on mobile to Gemini, and Google says it began removing Assistant from phones and paired devices such as watches, earbuds and Android Auto from 3 September 2026. Amazon introduced Alexa+, a rebuilt Alexa powered by generative AI, on 26 February 2025. Apple and OpenAI announced in June 2024 that Siri can hand questions to ChatGPT, with the user's permission. And ChatGPT, Gemini and other assistants all offer voice conversations of their own.
The common thread: a spoken question goes into a large language model, and a short spoken answer comes out. At Question.Marketing we treat that as the spoken layer of answer engine optimisation (AEO), which we cover in more depth in our field guide to AIVS, the spoken recommendation layer.
Why does voice search usually give only one answer?
Because nobody wants to listen to ten links. A screen can show a shortlist; a voice can only read so much before the listener loses track. So the assistant compresses what it found into one answer, often naming one business or one source.
That makes voice a harder test than typed AI search. On screen, being third in a list of options still gets you seen. Spoken aloud, third usually means silent.
What do the assistants have in common?
They prefer sources that answer the question directly, early and in plain language, and they lean on facts they can confirm from more than one place. None of this is new to good SEO, but voice removes the safety net of being fifth on a page.
In practice that means three things. The answer has to exist as a short, self-contained passage. The facts about you have to agree across your site, your listings and the sites that mention you. And for local questions, business listings and reviews carry a lot of weight. The seven steps below follow from those three points.
How do you write an answer that can be spoken?
Open each section with a one or two sentence answer that makes sense with no surrounding context, then expand. Read it aloud. If it sounds like a brochure, rewrite it.
Google's own guidance for speakable content is a useful benchmark even beyond Google: it recommends around 20 to 30 seconds of content per spoken section, roughly two to three sentences, with concise summaries broken into individual sentences so they read clearly through text to speech. That is a good length for any answer you want an assistant to lift.
Avoid answers that depend on a table, an image or "see above". A listener cannot see above.
How do people actually phrase spoken questions?
Spoken questions are longer and more conversational than typed ones. People say "who's the best emergency plumber near me that's open now" rather than typing "plumber Doncaster".
So build pages around the real questions customers ask, in their words. The best sources are the ones you already own: sales calls, enquiry emails, chat logs, reviews and the questions your team answers every week. If the same question keeps coming up, it deserves its own answer on your site, with the question as the heading. That is exactly how this article is structured.
Which schema helps voice search?
Accurate FAQPage, Article, Organization, Person and LocalBusiness schema help assistants understand who you are and what each page answers. Speakable schema can flag the passages best suited to being read aloud, but treat it as a hint, not a strategy.
Be honest about speakable's limits. Google still lists it as a beta feature, used by Google Assistant for topical news queries, and says it works for users in the US with Google Home devices set to English and for publishers that publish in English. It costs little to add and it marks your best spoken passages clearly, which is why we use it on this page. It is not a ranking switch, and it will not make a UK service business the answer on its own.
The schema that matters more is the boring kind: one consistent Organization and Person description, linked to your real profiles, so every system sees the same entity. We cover what schema can and cannot do in does schema markup help ChatGPT understand my business?
Why does being the same entity everywhere matter?
Assistants trust what agrees. Your name, business, address, description and expertise should read the same on your site, Google Business Profile, LinkedIn, directories and podcast bios.
When the facts disagree, a model has to pick one, hedge, or leave you out. When they agree across many independent places, you become the safe answer. This is slow, unglamorous work: pick one sentence that describes what you do, and use it everywhere.
Why do assistants need independent corroboration?
Your own site says what you do. Interviews, guest articles, reviews and citations from sites you do not control are what persuade a model it is true.
This is where press-release-only campaigns fall short. Self-awarded titles pushed through paid newswires can put a name into AI answers for a while, but they can also attract fact-checks and put the word "disputed" next to that name. Distribution is not endorsement. Recognition built on evidence, such as dated work, real clients and independent coverage, lasts longer. It is also why we publish our own dated receipts rather than asking anyone to take our word for it.
How do you win local voice search?
Many spoken questions carry local intent, so start with your Google Business Profile. Complete every field, keep opening hours accurate, choose the right categories, and collect reviews that mention the services you want to be named for.
Then make sure your name, address and phone number match on every directory that mentions you. "Near me" and "open now" questions are exactly the kind people ask out loud in a car or with their hands full, and the assistant has very little to go on except listings, reviews and your site.
How do you measure voice search visibility?
There is no voice search console, so measure by asking. Pick 20 to 50 questions your customers ask, ask them aloud each month on each major assistant, and log who is named, whether you are named, and whether the answer recommends you or just mentions you.
It is manual, but it is the only honest measure of whether you are the answer. Keep the questions fixed so the trend means something, record the date and device, and save the transcript or a screen recording. Our article on how often to check your AI visibility and our AEO measurement guide cover the cadence and the KPIs.
Where should you start this week?
List ten questions your customers ask, write a spoken-length answer to each, publish them on one page with FAQ schema, and run your first monthly assistant check. That is the baseline everything else is measured against.
Then work outwards: align your one-sentence description across every profile, tidy your Business Profile, and go after one independent mention a month. Our voice search FAQ shows the ten-question format in practice, and the voice search hub collects everything we publish on the subject.
Quick answers
What is voice search optimisation?
Voice search optimisation is making your content easy for voice assistants to find, read aloud and attribute, usually by giving short, direct answers to the questions people ask out loud. At Question.Marketing we treat it as the spoken layer of answer engine optimisation (AEO).
Which voice assistants matter in 2026?
Gemini, which Google is rolling out in place of Google Assistant on phones and paired devices; Siri, which can pass questions to ChatGPT with your permission; Alexa+, Amazon's generative AI Alexa; and the voice modes in ChatGPT and other AI assistants.
What is speakable schema?
Speakable is structured data that marks the parts of a page best suited to being read aloud. Google lists it as a beta feature for news content, available to US users with Google Home devices set to English, so treat it as a hint rather than a guarantee.
How do I measure voice search visibility?
Ask a fixed set of 20 to 50 customer questions aloud each month on each major assistant, and log the date, the device, who is named and whether you are recommended. Keep the questions the same so the trend means something.
Sources
- Google Search Central, Speakable (BETA) structured data: https://developers.google.com/search/docs/appearance/structured-data/speakable
- Google, "The Assistant experience on mobile is upgrading to Gemini" (14 March 2025): https://blog.google/products-and-platforms/products/gemini/google-assistant-gemini-mobile/
- Google Assistant Community, "Transitioning from Google Assistant to Gemini on Mobile" (6 August 2026): https://support.google.com/assistant/thread/457649886
- Amazon, "Introducing Alexa+, the next generation of Alexa" (26 February 2025, updated 21 July 2026): https://www.aboutamazon.com/news/devices/new-alexa-generative-artificial-intelligence
- OpenAI, "OpenAI and Apple announce partnership" (10 June 2024): https://openai.com/index/openai-and-apple-announce-partnership/