Voice Search Optimisation: How to Be the Answer That Gets Read Out

Typed search gives people ten options.

Voice search usually gives them one.

That single reply is either your answer or someone else's. There's no second place read aloud.

So voice search optimisation is really about one thing: writing the answer a machine can say out loud without embarrassing itself.

Below are the seven steps I use, from finding spoken questions to preparing for Google's live voice conversations.

What Changes When the Answer Is Spoken

Reading and listening are different jobs.

A reader can skim a table, glance back at a figure or click a link. A listener gets one pass, in order, at the speed of speech.

That creates four constraints:

Google has also changed how voice works in Search.

Its help page for Search Live describes back-and-forth voice conversations with Search in the Google app, with spoken responses and web links shown on screen. Google says it is offered wherever AI Mode is available.

That shifts the goal slightly.

You want to be the source the spoken answer draws on, and the link the person taps when they want more.

Step 1: Find the Questions People Say Out Loud

People speak in full questions. They type in fragments.

"plumber leeds" becomes "who's a good plumber near me that can come today".

So start by collecting the spoken versions:

• Search Console: filter queries with a regex on question words, such as ^(who|what|when|where|why|how|can|does|is)\b, and sort by impressions.
• People Also Ask: note the exact phrasing Google shows for your main topics.
• Phone calls and emails: the questions customers ask your team are the questions they ask their phone.
• Autocomplete: type the first words of a question and record the suggestions.

Search Console does not label which queries were spoken. Long, conversational questions are the closest proxy you have.

Group the questions by the page that should answer them. One page can own a whole cluster.

Pro tip: Read your question list out loud. If a phrase sounds like something nobody would say, it is a typed query, not a spoken one. Keep both, but write the answer for the spoken version.

Step 2: Write a 20 to 30 Second Answer Under Each Question

This is the core of the work.

Google's speakable documentation suggests sections of about 20 to 30 seconds of audio, roughly two or three sentences, for content meant to be read aloud.

That's a useful target for any spoken answer.

The format I use:

1. A heading phrased as the question. "How long does a boiler service take?"
2. A first sentence that answers it outright, naming the subject.
3. One or two sentences of the detail that matters most.
4. Then the full explanation for people reading the page.

Here's the difference:

Hard to read aloud: "Timescales vary depending on a range of factors including make, model, age and condition (see table below)."

Easy to read aloud: "A standard boiler service usually takes 30 to 60 minutes. Older boilers or ones with faults found during the check can take longer."

The figures are illustrative. The shape is what counts.

The second version works whether it is read on a screen, spoken by a phone or quoted in an AI answer.

This is the same answer-first structure behind answer engine optimisation, squeezed to spoken length.

Step 3: Run the Read-Aloud Test

Now read every answer out loud. Or paste it into a text-to-speech tool and listen.

You'll catch problems no editor spots on screen:

• Symbols. "£1.2k-£3k pa" is unreadable. Write "between twelve hundred and three thousand pounds a year" in the spoken answer, or at least "£1,200 to £3,000 a year".
• Abbreviations. "T&Cs", "approx." and "e.g." trip up speech engines and listeners alike.
• Brackets. Asides in brackets break the flow when spoken.
• "See below" and "as shown". There is no "below" in audio.
• Long lists in one sentence. More than three items in a row is hard to follow by ear.

Fix the answer paragraph only. The detail underneath can keep its tables and figures for readers.

My SERP snippet preview tool is handy here too. If the answer reads badly as a short snippet, it will sound worse.

Step 4: Win the Typed Answer That Voice Reads From

A spoken reply still has to come from somewhere, and on Google that somewhere is the search results.

Google's AI Mode help page says it splits a question into subtopics and searches each one. Search Live runs inside that same experience.

So the passage that wins the typed answer for a question is in the strongest position to be spoken, too.

That means:

• Targeting questions where your page already ranks on page one
• Checking which page holds the snippet now, and why
• Matching the format Google shows: paragraph, list or table
• Making sure no nosnippet rule blocks your best section

I cover the full process in how to get featured snippets.

One caution for voice: if the current snippet is a table, the spoken version will often be clumsy. A crisp paragraph answer above your table gives the system a better option to read.

Step 5: Fix the Local Facts Assistants Read Back

Many spoken questions are practical and local, asked by someone on the move.

"What time does the garden centre close?"

"Call the nearest locksmith."

"Is the pharmacy on the high street open on Sunday?"

Those answers rarely come from a blog post. They come from business listings.

So check:

• Business Profile: hours, holiday hours, phone, address, categories and services all complete
• Your website: the same name, address, phone and hours, in plain text, on the contact page and footer
• Directories: no old phone numbers or closed branches still listed

Wrong opening hours are worse than missing ones. Someone drives over and finds the door locked.

My guide to optimising your Google Business Profile walks through every field, and local SEO covers the citations and reviews around it.

Step 6: Prepare Pages for Live Voice Conversations

Search Live and AI Mode change voice from one question to a conversation.

Google's AI Mode help page says people can ask by voice and keep going with follow-up questions. Search Live adds spoken replies with links on screen.

That has two consequences.

First, follow-ups matter. Someone asks about a loft conversion, then the cost, then planning rules, then who does it nearby. Each turn is a new chance for your page to be the source.

Second, the click is a decision. When someone taps a link mid-conversation, they want the specific thing they just heard about. If your page opens on a hero banner and three paragraphs of introduction, they leave.

So for each important page:

• Put the most-asked answer in the first screen
• Use headings that match the follow-up questions in your Step 1 list
• Make the next action obvious: call, book, get a price

I go deeper on conversational search in my guide to Google AI Mode SEO.

Step 7: Use Speakable Only If You Fit the Beta

Speakable schema marks the sections of a page that are best suited to being read aloud.

It sounds made for voice search. In practice it's narrow.

Google's documentation says speakable is in beta, works for users in the US with Google Home devices set to English, and is used by Google Assistant to answer topical news queries on smart speakers.

So my rule:

If you do use it, Google says to avoid marking up content that sounds confusing when read, such as datelines, photo captions and source attributions.

For everyone else, the answer paragraphs from Step 2 do the job without any markup. My answer engine optimisation service uses schema only where it describes the page accurately, and this is a good example of why.

Mistakes That Keep You Out of Spoken Answers

These come up again and again in audits:

• Writing for keywords, not questions. "Boiler service Leeds prices" is not how anyone speaks.
• Answers that depend on the page. "As mentioned above" means nothing when heard alone.
• Figures in a format speech can't handle. Ranges with symbols, abbreviations and footnotes.
• Out-of-date hours and phone numbers. The commonest voice failure for local firms.
• Expecting a voice traffic report. No mainstream tool reports spoken queries separately, so judge progress by the question clusters you target.

Make Your Answer the One People Hear

Voice search optimisation isn't a separate strategy. It's answer-first content held to a stricter standard: short, spoken, accurate and local where it counts.

Your first action: pull ten question queries from Search Console, write a two-sentence answer for each, and read them aloud.

You'll hear what needs fixing straight away.

If you'd like help restructuring pages so they become the reply, my work on making your pages the answer covers featured snippets, People Also Ask and spoken results together.