The unit of optimisation in voice search is not the page. It is the single spoken answer an assistant reads aloud. This article shows how to structure pages so that answer is yours.
Context: The Retailer Whose 'Near Me' Traffic Fell While Question Queries Rose
A mid-sized South African homeware brand, running a Shopify store with a physical showroom in Cape Town, noticed something odd in Search Console. Queries beginning with “how”, “what” and “where” were climbing steadily. At the same time, traffic from searches containing “near me” had thinned, even though showroom footfall and phone orders had not dropped. Google Analytics showed no corresponding referral source. The team assumed voice search was stealing clicks, but had no way to prove it or act on it.
This is the practical reality of voice search optimisation and AI in 2025: the assistant answers the question, the user never visits the page, and your analytics records a zero-click. The opportunity is not to win a “voice channel” - it is to become the answer that gets read aloud, and to measure the downstream behaviour that follows. That requires a different structural discipline than classic SEO.
What Voice Search Optimization Means Now That Assistants Answer Instead of List
Voice search has gone through two distinct eras, and most advice online conflates them. In the first era, a spoken query was simply a search typed with your voice. Google returned a list of blue links and, often, a featured snippet. Optimising for voice meant winning the snippet: a concise paragraph, a list, or a table that answered the query directly.
In the second era, AI assistants (Google Assistant, now being replaced by Gemini on mobile; Siri; Alexa; ChatGPT Voice) do not return a list. They synthesise an answer from multiple sources, then read it aloud. They may not cite you at all, or may mention your brand only if entity confidence is high. The optimisation target shifts from “rank for the query” to “be the passage the assistant selects, trusts, and speaks.”
Speakable schema is one of the few explicit signals you can send. It tells assistants which parts of a page are suitable for text-to-speech. Adoption is still low, which makes it a differentiator for early movers.
How a Voice Assistant Decides What to Read Aloud
Assistants do not rank pages the way Google Search does. They retrieve candidate passages, score them for relevance and confidence, then compose a spoken response. The factors that matter are:
- Conversational query fan-out: a single spoken question is expanded into several sub-queries. A question like “what are the best load-shedding-friendly fridges under R10,000” triggers separate retrieval for price, power consumption, and availability.
- Passage retrieval: the assistant looks for self-contained text blocks that answer each sub-query. Long, meandering pages lose to structured sections with clear headings.
- Entity confidence: the assistant needs to know who you are - your business name, location, product category - from structured data, consistent citations, and your Google Business Profile.
- Freshness and verification: hours, stock status, and prices are checked against local signals. Stale data is skipped.
- Structured data cues: FAQPage, HowTo, and speakable markup give the assistant machine-readable answers.
The Two Eras Side by Side: Snippet Capture Versus Answer Synthesis
| Dimension | Era One: Spoken Query + Snippet | Era Two: AI Assistant Synthesis |
|---|---|---|
| Primary target | Featured snippet ownership | Passage selection and spoken citation |
| Content structure | 40-60 word paragraph or list | Question-shaped H2, direct answer block, entity cues |
| Structured data | Optional, rarely used | FAQPage, HowTo, speakable JSON-LD |
| Measurement | Snippet impressions in Search Console | Zero-click query growth, branded residual lift, assistant referral logs |
| Failure mode | Losing snippet to a competitor | Being retrieved but not cited, or cited without a link |
The overlap is real: both eras reward concise, well-structured answers. But era two adds entity verification and passage granularity. You cannot win era two by simply writing shorter paragraphs; you need to build the page so each answer stands alone.
Rewriting Page Structure Around the Spoken Answer Block
The core structural unit is the spoken answer block: a self-contained passage that directly answers a question, written to be read aloud without visual context. It should begin with the answer, not preamble. For example, instead of “When it comes to choosing a fridge for load-shedding, there are many factors…”, write: “A load-shedding-friendly fridge should maintain temperature for at least four hours without power. Look for inverter compressor models with a stated holdover time.”
Place this block immediately under a question-shaped heading. Keep the block to one short paragraph. Follow it with supporting detail, but ensure the block itself is complete.
Writing Headings as Questions People Actually Speak
Use the phrasing people use when talking to an assistant. “How long does a fridge stay cold without power?” is better than “Load-shedding performance”. Pull these questions from Search Console's query report, your site search logs, and sales call transcripts. Avoid keyword-stuffed headings like “Load Shedding Fridge | Best Load Shedding Fridge”.
The 40-Word Answer Rule and Why Length Matters Aloud
A spoken answer needs to fit comfortably in one breath - roughly 40 words. Beyond that, assistants often truncate or paraphrase, diluting your brand. Aim for 35-45 words. If the answer requires more, split into two blocks: a direct answer and an optional elaboration block. Test by reading aloud at normal pace; if you run out of breath, it's too long.
Warning: Do not sacrifice accuracy for brevity. If a legal or safety detail is required, include it in a second sentence, but keep the core answer under 40 words.
Entities, Not Keywords: What the Assistant Needs to Name Your Business
Assistants rely on knowledge graphs. To be named in a spoken answer, your business must be an entity with consistent attributes across the web: name, address, phone, category, and product lines. Use schema.org Organization and LocalBusiness markup. Ensure your Google Business Profile matches your website exactly. Build citations on relevant directories and industry sites. When an assistant answers “where can I buy X in Cape Town”, it draws on entity confidence, not keyword density.
Where AI Voice Layers (Gemini, Siri, Alexa, ChatGPT Voice) Differ in What They Read
Not all assistants read the same sources. Gemini on mobile leans heavily on Google Search and Knowledge Graph, and is replacing Assistant on Android. Siri uses Apple's own index plus partner data. Alexa prioritises Amazon and Bing. ChatGPT Voice, when browsing, may pull from Bing or its own training data. The practical implication: optimise for the intersection (structured, entity-rich, fresh content) and prioritise the assistant your audience actually uses. For most South African brands, Gemini/Google and Alexa (for product queries) matter most.
Structured Data for Spoken Surfaces: FAQPage, HowTo and Speakable
Three schema types do most of the work for spoken answers. FAQPage marks question-answer pairs; HowTo describes step-by-step processes; speakable indicates which parts of a page are suitable for text-to-speech. Here is a minimal speakable + FAQ example for a product page:
{
"@context": "https://schema.org",
"@type": "WebPage",
"speakable": {
"@type": "SpeakableSpecification",
"cssSelector": ["#answer-how-long", "#answer-holdover"]
},
"mainEntity": {
"@type": "FAQPage",
"mainEntity": [{
"@type": "Question",
"name": "How long does a fridge stay cold without power?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Most modern fridges maintain safe temperatures for 4 to 6 hours if kept closed. Inverter models with better insulation can reach 8 hours."
}
}]
}
}Implement once and validate with Google's Rich Results Test. For South African stores, budget around R8,000-R15,000 for a developer to add speakable and FAQ markup across a 50-page catalogue if you do not have in-house skills.
Local Voice Signals: Google Business Profile, Hours and 'Near Me' Queries
When someone asks “what hardware store is open near me now”, the assistant verifies hours and location from your Google Business Profile (GBP). Ensure your GBP hours are accurate, including holiday hours. List service areas if you deliver. Respond to reviews, as assistants may use review sentiment as a trust signal. For multi-location brands, each location needs its own GBP and LocalBusiness schema with consistent NAP data. A common mistake is a single GBP for multiple branches, which confuses assistants and suppresses near-me answers.
Quick win: Add a “near me” FAQ block to your contact page that names your suburbs and says “we are open [hours]”. This gives assistants a direct answer to read.
Multilingual and South African Spoken-Query Realities
South Africa's spoken-query landscape is shaped by mobile-first usage and multilingual users. Assistants support isiZulu, Afrikaans, and other languages to varying degrees. A user may ask in English but expect answers that reference local terms (e.g. “load-shedding”, “bakkie”, “robot”). Include local vocabulary naturally in your answer blocks. For example, a query like “where to buy a bakkie canopy in Pretoria” should find an answer block that uses “bakkie canopy” and “Pretoria”. Avoid machine-translating entire pages; instead, create targeted FAQ blocks in the languages your audience actually queries in. As an estimate, multilingual voice queries are growing but still a minority of total search volume in SA - treat as an emerging segment.
Measuring Voice Without a Voice Report: A Practical GA4 and Search Console Approach
GA4 does not have a “voice” channel. Voice-originated sessions may appear as direct, referral from google.com, or not at all (zero-click). To infer voice impact, build a measurement layer:
- Search Console question queries: filter queries containing who, what, when, where, why, how, is, can, does. Track impressions and average position for these over time.
- Zero-click behaviour: monitor overall clicks vs impressions for question queries. Rising impressions with flat clicks suggests answer capture.
- Featured snippet ownership: use Search Console's “featured snippet” filter (under Search Appearance) to see which queries you own.
- Residual lift: track branded search volume and direct traffic in GA4. If voice answers mention your brand, you may see a lift in branded queries and direct sessions without a clear source.
- Call and chat logs: tag calls and chats that mention “I asked my phone” or “Google said”.
Combine these into a custom report. Prebo Digital's custom reporting approach builds these layers into a single dashboard so you can see leading indicators before revenue shifts.
Leading Indicators Table: What to Track and What It Tells You
| Indicator | Where to find it | What it signals |
|---|---|---|
| Question query impressions | Search Console Performance | Growing conversational demand |
| Snippet ownership count | Search Console Search Appearance | Answer capture in era one |
| Zero-click ratio on question queries | Calculated from Search Console | Assistants answering without click |
| Branded search volume | Search Console, GA4 | Residual brand mention in answers |
| Direct traffic to answer pages | GA4 | Users typing URL after hearing it |
| GBP calls and direction requests | Google Business Profile | Local voice action |
Where Voice Plugs Into Amazon: Spoken 'Buy' Intent and the Marketplace Angle
Alexa's ecosystem ties spoken purchase intent directly to Amazon listings. When a user says “Alexa, buy [product]”, the assistant selects from Amazon's catalogue based on title, bullet points, and reviews. Optimising for this means treating your Amazon listing as a spoken-answer block: a clear product title, a concise benefit statement, and structured attributes. Ensure your A+ content answers common questions. If you sell on Amazon, review your listing's readability aloud. For brands using Amazon Marketing, this is a natural extension of listing optimisation. Learn more about Amazon marketing services.
What to Expect in the First 90 Days
Voice search optimisation is not an overnight switch. In the first 30 days, focus on restructuring top question pages into answer blocks and adding FAQ schema. By day 60, expand to speakable markup and local signal cleanup. By day 90, you should see movement in leading indicators: more question query impressions, increased snippet ownership, and possibly a lift in branded search. Revenue impact may lag by a quarter. Treat this as a structural improvement to your content and data, not a campaign. If you need a structured approach, explore our SEO and AI SEO services.
Remember: the goal is not to chase a “voice channel” but to ensure that when an assistant answers a question in your category, your brand is the one it names and trusts.
