Ask Alexa for the best coffee shop nearby and you get one name. Not a list of ten options, not a "next page" button, not a choice. If you're not in that single answer, you simply don't exist as far as the user is concerned.
That's the heart of voice search optimization in 2026. It's no longer about ranking in a list — it's about being cited in an answer that an assistant reads out loud. And that answer increasingly doesn't come from the classic search index, but from a language model that either has or hasn't read your content.
I test this regularly: I ask Siri, Alexa and ChatGPT for the opening hours, prices and locations of random small and medium-sized businesses. About half the time I get a vague answer, an outdated phone number, or "I don't know." That's not a technical problem. It's a content problem.
Voice search works differently: one answer instead of ten links
With a typed query, people scan a results page, compare two or three options and click through. With voice, that's impossible. You ask a question, you get one answer, and you move on with your day. There's no room for second place.
In 2026, virtually all of these assistants run on a language model. Siri works with Apple Intelligence and hands more complex questions to ChatGPT. Alexa+ has been rebuilt around an LLM. Google Assistant and Gemini pull answers from AI Overviews and the Knowledge Graph. As a website owner, it matters little which model sits underneath: they all do the same thing. They look for one factual answer, and they give exactly one.
That changes the question you need to ask. Not "am I on page one?" but "do I have the answer an assistant can read aloud, and does it trust my source enough to name me?"
Voice search optimization starts with the question, not the keyword
People who type use keywords. People who talk use full sentences. "Dentist Nijmegen" becomes "which dentist in Nijmegen is still open tomorrow?" That's exactly why keyword density gets you nowhere here. An assistant doesn't match individual words — it matches meaning.
In practice, that means three things:
- Write headings as questions. "How much does a teeth cleaning cost?" works better than "Teeth cleaning prices". An assistant reads that heading, recognizes the question and lifts the answer underneath it.
- Answer immediately, in 40 to 60 words. Not a paragraph about your passion for dental care first — the price or the opening time, right away. Context can come after.
- Name your entities explicitly. Your business name, the town, the service, the product. Not "we", but "Van Dijk Dental Practice in Nijmegen". Models work with entities, not loose terms.
This is the same logic covered in more depth in AEO for businesses: the practical guide, with one extra requirement: the answer also has to sound good when someone reads it out loud.
Structured data that assistants actually use
A handful of schema types have a direct impact on spoken answers:
- LocalBusiness with a complete address, opening hours, phone number and geo coordinates.
- FAQPage for recurring questions — but only if those questions genuinely appear on the page.
- Speakable to flag which block of text is suitable for reading aloud.
- Product with price and availability, if you sell something.
Without this markup, assistants have to guess. With it, they know exactly where the answer lives. That saves your user a whole lot of uncertainty.
The technical side: can an assistant actually read your site?
A perfect answer is worthless if the bot never reaches your page. And that's where things often go wrong in practice.
First: robots.txt. Many sites accidentally block GPTBot, PerplexityBot, OAI-SearchBot or Google-Extended. It feels like the safe choice — "we don't want AI using our content" — but the result is that you're invisible in exactly those answers your customers hear read out loud.
Second: rendering. If your content only appears via JavaScript, whether an assistant sees it is a coin toss. Far from all AI crawlers wait around for your JS. Server-side rendering isn't a luxury, it's the baseline.
Third: speed. This is one of the few places where Core Web Vitals directly affect AI visibility. A page that takes four seconds to load is out of the running with assistants that want an answer within a second. Core Web Vitals and AI crawlers explains exactly which thresholds apply.
Local voice search: where small businesses win
Most voice queries are local and moment-driven. "Is there a pharmacy open near me?" "Where can I take my car for its annual roadworthiness test (APK)?" These are questions with money behind them — someone needs something sorted today.
Assistants pull that answer from two sources: your website and your Google Business Profile. Make sure the two don't contradict each other. Identical business name, address, phone number and opening hours in both places. Half an hour's difference in opening times is enough to make an assistant hesitate — and then it won't mention you at all.
Reviews play a bigger role here than most people realize. A voice assistant will pick a business with forty recent reviews over one with three. Not because the model analyzes sentiment, but because consistency and volume signal reliability.
What you should definitely avoid
A few mistakes I keep seeing in voice search optimization:
Stuffing spoken queries with keywords. "Best cheap plumber Amsterdam emergency 24 hour" in an H2 reads like a robot wrote it — and an assistant can tell. Write the way people talk.
Superlatives with nothing behind them. "We are the best web designer in the country" never gets cited. No model repeats that without a source. What does work: "We've built 47 websites for small and medium-sized businesses, live in six weeks on average."
Hiding information in PDFs or images. Assistants don't read text inside a brochure PDF or a photo. Put your rates, opening hours and terms on the page as plain text.
Putting key info behind a form. If a user has to log in before they can get their answer, the assistant gives up. Every time.
How to measure whether it's working
Voice search isn't a channel in Google Analytics. There's no "voice" filter you can switch on. So you'll have to measure it indirectly:
- Referrals from AI services. Filter your analytics for referrers such as chatgpt.com, perplexity.ai and gemini.google.com. That isn't voice, but it's the same underlying machine.
- Branded search volume. If more people search for your business name — whether by voice or not — that's a sign you're being mentioned.
- Direct traffic. A rise in "direct" traffic without a campaign can point to people who found you through an assistant and typed your name in afterwards.
- Test it yourself. Ask Siri, Alexa and ChatGPT the same ten questions every month. Track which answers mention you, which sources they cite, and what changes.
Want to know what this kind of work costs and returns? GEO costs and ROI: what does it cost and what does it deliver? sets out realistic ranges.
Frequently asked questions
How do I optimize my website for voice search?
Write headings as natural questions and answer them directly in 40 to 60 words, in the first paragraph under the heading. Add LocalBusiness or FAQPage schema, make sure your site renders server-side, and allow AI crawlers in robots.txt. Also ensure your Google Business Profile matches your website exactly. Those are the five basic steps.
Does voice search optimization work for local businesses too?
It works even better: for local businesses it's the biggest quick win of all. Most voice queries are local and moment-driven — opening hours, availability, "near me". Ensure consistent NAP data (name, address, phone), up-to-date opening hours and enough recent reviews. With assistants, that carries more weight than any technical trick.
Do I need to rewrite my content for AI assistants?
Often, partly. If your content opens with three paragraphs of atmosphere before getting to the answer, it's unusable for voice. The good news: the same fix also helps in Google AI Overviews and in classic featured snippets. So you're not writing for one channel, but for every answer machine at once.
How do I know if my voice search optimization is working?
There's no direct voice search report. Measure indirectly: referrals from chatgpt.com and perplexity.ai, branded search volume, and asking Siri, Alexa and ChatGPT test questions yourself every month. Do they mention you more often, and with the right facts? Then it's working. If not, you're usually missing a direct answer or structured data.