Voice search optimisation is the practice of structuring your site so a voice assistant reads it out as the single spoken answer to a question. Because Siri, Alexa and Google Assistant return one result rather than a page of links, the work is a competition for one slot: question-led content, answers of forty to sixty words, the schema assistants read, and clean local signals. This guide covers the whole programme, from the Google Search Console audit to measured snippet wins.
Why voice search is a competition for one slot
Ask a phone for an electrician and you get one answer, not ten links. That single constraint defines the channel: the assistant picks one result and speaks it, so second place is silent. Assistants also strip most of your branding, so the listener hears the answer and only sometimes the source, which means the answer itself has to carry the sale. Optimising for voice is therefore narrower and more specific than general SEO advice admits, and the payoff is concentrated. You either own the spoken slot or you are invisible for that query.
How spoken queries differ from typed ones
A typed query reads “emergency electrician leeds”. The spoken version is “who can fix a tripped fuse box near me tonight”. Spoken queries run longer, open with question words, and carry local and immediate intent far more often. Phrases such as “near me”, “open now” and “how much does it cost to” dominate. If your pages target only the short head terms, an assistant has nothing that matches how the question was actually asked, so it reaches for a competitor whose page mirrors the phrasing.
Find the questions people actually ask
You do not need to guess the questions. Open the Performance report in Google Search Console, add a Query filter, choose Custom (regex) and enter ^(who|what|how|why|when|where|which|can|should|does|is). Set the date range to the last twelve months so seasonal questions surface, then export the table to a spreadsheet. Every row is a question Google already ties to your site, with impressions, clicks, average position and CTR attached. Add two working columns: the page currently ranking, taken from the Pages tab, and a plain yes or no for whether that page answers the question directly. On most sites the honest answer is no for more than half the rows.
Two other sources round out the list. People Also Ask boxes show the follow-up questions Google already associates with a topic, so expand them a few levels deep. Your inbox and phone log hold the rest, because the questions your team answers every week are the ones buyers speak into their phones, and sales and support staff can usually list twenty in ten minutes. If the site is large, a crawl in Screaming Frog or Sitebulb tells you at scale which pages already carry question-led headings and which are missing schema, so you audit structure in an hour rather than clicking through by hand.
Score down to the twenty questions worth the work
An export of two hundred questions is not a plan, so score each candidate on three points and keep the top twenty. Value: the query relates to something you sell or support, so cost and comparison questions outrank trivia. Winnability: the current average position sits between four and fifteen, close enough for a restructure to move it. Beatability: search the query, read the snippet showing now, and only pick fights where the incumbent answer is vague, dated or buried mid-paragraph. Put the twenty survivors in a tracker with the target page, today’s position and today’s snippet owner. That sheet is the whole scorecard, so date it and keep it honest.
Write answers an assistant can read aloud
For each question, put the question in a heading in natural spoken wording, then answer it fully in the first forty to sixty words underneath. That length is roughly what an assistant reads out in eight to ten seconds, and it matches the size of the featured snippets voice answers are usually pulled from. Order it answer first, caveat second, detail third. If your opening sentence is “there are several factors to consider”, you have already lost the slot.
A quick editorial test: read the answer out loud. If it sounds like a person helping a customer, it is right; if it sounds like a brochure, rewrite it until it would not sound odd coming from a kitchen speaker. Where a page covers several related questions, group them on that page rather than spinning up thin single-question URLs, because consolidated pages hold their authority better.
Add speakable, FAQ and local schema
Three schema markup types earn their place here. The speakable property uses a SpeakableSpecification with a cssSelector that points assistants at the blocks best suited to being read aloud; assistant support is still uneven, but the markup costs minutes and states your intent. FAQPage schema fits any genuine run of question and answer pairs, and its text must match the visible copy word for word, because a mismatch quietly disqualifies the markup. LocalBusiness schema carrying your telephone, address, geo coordinates and openingHoursSpecification feeds direct spoken answers such as “what time does this shop close”.
Validate every page in the Rich Results Test and fix errors the same day. Then request recrawling of all twenty URLs through URL Inspection in Search Console, so the changes are picked up in days rather than weeks.
Do not neglect the local layer
A large share of voice queries are local, and for those the assistant often answers from your Google Business Profile rather than your website. Keep the primary category exact, the opening hours current including bank holidays, and the phone number correct, because “is this place open now” and “call the nearest plumber” are resolved entirely from profile data. A polished website cannot rescue a neglected profile. Speed matters too: run the key pages through PageSpeed Insights and clear the Core Web Vitals thresholds on mobile, since assistants favour pages that load quickly on a phone.
Measure it like a campaign
Movement arrives unevenly, so give the test eight weeks before you judge it. Each week, update the tracker with average position and CTR from Search Console, and note who owns the snippet, checked in a private browsing window or with a rank tracker such as Ahrefs or Semrush that records snippet ownership. A fair expectation is that a quarter of the twenty move meaningfully inside the first month and half by week eight; the stragglers usually need a better answer, not more schema. When a question is won, retire it and promote the next candidate from the export, so the programme keeps rolling instead of ending. To connect the work to money rather than rankings, tie the won queries to enquiries in GA4, and if several people need to see progress, a simple Looker Studio dashboard fed from Search Console keeps the snippet-win trend in one place.
Key takeaways
- Voice search returns one spoken answer, so the work is a competition for a single slot, not a comfortable ranking range.
- Spoken queries are longer, question-led and local, so target the phrasing people actually say, not just typed head terms.
- Pull your question list from the Google Search Console Performance report with a regex filter over twelve months, then score down to twenty by value, winnability and beatability.
- Answer each question in the first forty to sixty words, answer first and caveat second, in language that sounds right read aloud.
- Add speakable, FAQPage and LocalBusiness schema markup, validate in the Rich Results Test, and request recrawls through URL Inspection.
- Keep the Google Business Profile accurate and clear Core Web Vitals on mobile, because local spoken answers come from the profile and from fast pages.
- Measure like a campaign over eight weeks, track snippet ownership, then retire won questions and promote the next candidate.
If you would rather this ran as a managed programme end to end, our Voice Search Optimisation service handles the audit, the rewrites, the schema and the measurement for you, and questions are welcome at digitalstrategy@globalcharter.com.
Frequently asked questions
Is voice search optimisation different from normal SEO?
Not really a separate discipline. Voice optimisation is featured-snippet and Google Business Profile work done with a listener in mind. The same question-led headings and forty to sixty word answers that win a spoken slot also win text snippets and read better for humans, so the work compounds with the rest of your search programme rather than competing with it.
How long before voice search work shows results?
Give it eight weeks before judging. On a typical rollout of twenty questions, expect around a quarter to move meaningfully in the first month and roughly half by week eight, with the slower ones usually needing a sharper answer rather than more markup. Requesting recrawls of the changed URLs through URL Inspection shortens the initial wait from weeks to days.
Do I need speakable schema for voice search to work?
It helps, but it is not the whole job. Assistant support for the speakable property is still uneven, so the answer structure, the forty to sixty word length and an accurate Google Business Profile do most of the lifting. Add speakable and FAQPage schema anyway, because they cost minutes, state your intent clearly and back up the on-page work.