Voice search optimization

TL;DR

Voice search rewards pages that answer a spoken question in a single spoken sentence. The three moves that matter: rewrite every FAQ answer to stand alone under 29 words, add speakable schema on question blocks, and only build an Alexa Skill or Google Action when the brand has a real habit loop. Everything else is theater.

The surface, in one paragraph

Voice search is the class of queries a person speaks to Siri, Alexa, Google Assistant, or Bixby, then hears an answer read back. The query is longer, more conversational, and shaped like a question. The assistant picks one answer, reads it, and moves on. There is no page two. The winning page has three properties: it uses the words the person spoke, it contains a single answer that reads well aloud, and it is machine-readable enough that the assistant can find that answer without guessing. Most brand sites fail all three. That is why the incumbents in every voice-answered category are Wikipedia, government sites, and a small cluster of publishers who invested in question structure.

Where this fits in Search Everywhere Optimization

VSO does not stand alone. It sits on top of five adjacent surfaces and inherits most of its ranking signals from them.

SEO is upstream. Google Assistant reads featured snippets. If the page does not rank in the top three for the underlying text query, it does not get spoken. Every VSO project starts by winning the text version of the query first.

AEO is a mirror image. Answer Engine Optimization wants a direct answer at the top of the page for Claude, ChatGPT, and Perplexity. VSO wants the same answer read aloud by Siri or Alexa. The same block of copy serves both. I never separate them.

GEO overlaps on structure. Generative Engine Optimization rewards the same TL;DR pattern, the same FAQ discipline, and the same entity clarity. Ship the structure once, harvest three surfaces.

LSO drives the local slice. Alexa and Google Assistant answer "where is the nearest" queries from local pack data. Google Business Profile hygiene, category selection, and review recency drive this. VSO for a local business is downstream of LSO.

E-E-A-T gates the whole thing. Assistants avoid reading answers from pages with weak author signals. Named authors, credentials in schema, and a real About page are prerequisites, not decoration.

The five levers

1. Natural-question FAQ discipline

Rewrite every FAQ so the question is the way a person actually says it aloud, and the answer is a single spoken sentence that stands alone. "What time does the store open?" beats "Store hours." "How much does a mattress cost?" beats "Pricing information." I keep answers under 29 words because that is the median length of a spoken snippet response Google reads. I test every answer by reading it aloud. If I stumble, it fails.

2. Speakable schema on the right blocks

Speakable is still a Google beta with limited rollout, but it costs nothing and it fires more often than the skeptics claim. I add SpeakableSpecification to the FAQ block and the article summary block using CSS selectors that map to the actual DOM. I never wrap the whole article. Assistants reject long speakable blocks. Bing had a parallel spec for Cortana that quietly retired, but the structure still reads clean to other parsers.

3. Entity clarity on the answer

The assistant needs to know what the page is about. That means schema.org type declared correctly (LocalBusiness, MedicalClinic, Product, HowTo, Recipe), sameAs pointing to Wikidata and Wikipedia where they exist, and a homepage that names the entity in the first sentence. When a page is about "Frederick Sona," the assistant needs to disambiguate from every other Frederick and every other Sona. KGO work makes VSO work.

4. Long-tail question capture at the URL level

I build question pages with the question as the URL slug: /how-much-does-a-crown-cost, /how-long-does-lasik-take, /when-is-a-roof-replacement-necessary. Each page: H1 matches the question, first paragraph is a 29-word answer, then a longer explainer, then FAQPage schema. This is the format that shows up in Google's People Also Ask, in featured snippets, and in the assistants' spoken responses.

5. Action skills only when the habit loop is real

Alexa Skills and Google Actions live or die on repeat use. If the brand does not have a use case a person invokes weekly (a recipe, a workout, a briefing, a meditation, a reorder, a status check), the skill dies within a quarter and drags the brand's assistant discovery down with it. I refuse skill projects where the habit loop is imaginary. When the loop is real, the skill compounds: repeat invocations, hands-free reorders, published skill directory listings that surface in Alexa search.

First 30 / 60 / 90 days

Days 1 to 30: audit and inventory

Query pull from Google Search Console filtered for interrogatives (how, what, when, where, why, who, does, is, can). Bing Webmaster Tools for the same. Answer The Public and AlsoAsked for the questions the tools miss.

Page-by-page FAQ audit. Which pages have FAQPage schema. Which do not. Which answers stand alone under 29 words when read aloud. Which are buried under three paragraphs of throat-clearing.

Speakable audit. Any existing speakable spec, and where the selectors point. Rich Results Test on the top ten question pages.

Entity audit. Wikidata QID if it exists. sameAs coverage on the homepage schema. Knowledge Panel state.

Deliverable at day 30: a prioritized list of the fifty highest-value question queries the brand should answer, a rewrite scope for the FAQ block on the top twenty pages, and a decision on whether an Alexa Skill or Google Action is worth building.

Days 31 to 60: rewrite and ship

FAQ rewrites shipped on the top twenty pages. Speakable schema added on the FAQ and article summary blocks. Rich Results Test passes for each.

Twenty new question pages built with question-as-slug URLs, question-as-H1, 29-word direct answer in the first paragraph, deeper explainer below.

Entity work: homepage schema updated with sameAs pointing to Wikidata, LinkedIn, Crunchbase, and any authoritative directory. About page rewritten to name the entity and its category in the first sentence.

If a skill or action is in scope: interaction model designed, sample utterances documented, prototype in the Alexa Developer Console or Actions on Google Console. No launch yet.

Deliverable at day 60: twenty rewritten FAQ blocks, twenty new question pages, tightened entity schema, and either a shipped skill prototype or a written decision not to build one.

Days 61 to 90: measure and expand

GSC query analysis. Which question queries picked up impressions. Which position-one rankings landed the featured snippet. Which pages started attracting long-tail interrogative traffic.

People Also Ask capture check. Which of the fifty target questions the brand now occupies inside PAA.

If a skill launched: invocation count, retention curve, review count. If invocations are below 200 a month by day 90, the skill is a maintenance burden. Kill it.

Expansion planning. Which topical clusters produced the highest question-query yield. Where the next fifty questions come from.

Deliverable at day 90: a working question-answer engine, measured PAA capture, a defensible speakable footprint, and a plan for months four through twelve.

Tools I use

Nothing exotic. The stack is:

• Google Search Console. Query filter for interrogatives. Position and impression data on the question queries the brand actually earns.
• Bing Webmaster Tools. Query data Google does not surface. Historical Cortana signals where they still appear.
• Google Rich Results Test. Speakable and FAQPage validation on every page before ship.
• Schema.org validator. Broader structured data check when Rich Results Test is silent.
• Answer The Public. Question mining by prefix.
• AlsoAsked. People Also Ask trees for topical planning.
• Alexa Developer Console. Skill build, interaction model, testing simulator.
• Actions on Google Console. Google Action equivalent, Dialogflow when the conversation is deep.
• ScreamingFrog SEO Spider. Site-wide schema extraction and FAQ audit.
• Ahrefs or Semrush. Question keyword volume and competitor snippet ownership.

What kills the program

Answers buried under throat-clearing. The first paragraph explains context, the second qualifies, the third eventually answers. The assistant never reaches paragraph three.

Corporate-voice FAQ. "Our services encompass a broad portfolio of solutions" is not how a person speaks or hears. Rewrite as if speaking to a friend on the phone.

Speakable wrapped around the whole page. Assistants reject long speakable blocks. The spec is for a single answer, not an article.

An Alexa Skill without a habit loop. A "brand skill" that plays a jingle and reads the FAQ dies inside a month. Only ship a skill where a real weekly-use case exists.

Weak author signals. Anonymous content, no About page, no author schema. The assistants deprioritize pages that fail E-E-A-T checks.

Ignoring the text query first. If the page does not rank in the top three for the underlying text query, no amount of speakable schema helps. Win text first.

Chasing Bixby. Bixby's answer share is small enough that dedicated Bixby work rarely pays. The one exception: brands with heavy Samsung device penetration and a repeat-use case worth a Bixby Capsule. Otherwise, general VSO work is enough.

KPIs that matter

The report I send every month has six numbers. None of them are "voice searches captured," because that metric is not reliably measurable across the four assistants.

• Question-query impressions in GSC. Filtered for interrogative words. Monthly trend.
• Featured snippet coverage on question queries. Count of question queries where the brand owns position zero.
• People Also Ask capture. Count of PAA slots the brand occupies across the fifty target questions.
• FAQPage rich result impressions. Reported directly by GSC.
• Assistant referral traffic. Small, but trackable via GA4 referrer analysis (assistant.google.com, alexa.amazon.com, some device-specific referrers).
• Skill invocations, if a skill is live. Alexa Developer Console dashboard, weekly retention curve.

I do not report on Alexa's "brand mention" metrics because they conflate paid and organic. I do not report on Cortana metrics because the surface is functionally gone. I report what I can measure and what actually moves the business.

FAQ

Does voice search still matter in 2026?

Yes. Assistant queries have plateaued in growth but the base is enormous: over a billion Alexa devices, hundreds of millions of Siri and Google Assistant queries a day. VSO also pulls double duty for AEO, since the same natural-question structure feeds AI answer engines.

Is speakable schema still supported?

Google keeps speakable as a beta and lets US publishers use it for Assistant news readouts. Broad support is limited. I still add it on FAQ and news pages because it costs nothing and pays off when it fires.

Do I need an Alexa Skill or Google Action?

Only if the brand has a repeat-use case: recipes, workouts, meditations, briefings, checkouts. Skills without a habit loop die within a quarter. Assess the use case before shipping.

What is the single highest-leverage VSO move?

Rewriting the FAQ so each answer is a single spoken sentence under 29 words that stands alone. Assistants prefer answers that sound like an answer, not a paragraph they have to summarize.

How do I track voice search results?

Directly, poorly. Indirectly, well. I monitor featured snippet coverage on question queries in Google Search Console, People Also Ask capture, and long-tail question impressions. Bing Webmaster shows Cortana signals for what remains of that surface.

How does VSO interact with AEO?

They share a spine. AEO wants a direct answer near the top of the page. VSO wants a direct answer that reads well aloud. The same rewrite serves both surfaces. I work them together, never in isolation.

Related reading

• Feed and Discover optimization (FEO): how Google Discover and Apple News handle question-shaped stories.
• Knowledge graph optimization (KGO): the entity work that gates every assistant answer.
• E-E-A-T playbook: author signals and editorial standards that gate assistant trust.
• Visual search optimization (VxSO): the image-side sibling of the same discovery system.
• Video search optimization (VDO): how transcripts and chapters feed assistants and search.