Voice search optimization

How spoken queries differ from typed ones

People type in fragments and speak in sentences. Someone at a keyboard searches for emergency plumber leeds. The same person talking to a phone asks who can fix a burst pipe near me right now. The intent is identical, the wording is not, and content written only for the typed version often has no passage that matches the spoken one.

Three differences matter in practice. Spoken queries are longer, usually by several words. They are far more often phrased as complete questions, starting with who, what, where, how or can. And they carry more local and immediate intent, because a large share of voice use happens on a phone while doing something else.

There is a fourth difference that changes the strategy entirely. A results page shows ten options. An assistant reading aloud gives one. There is no second place to be in a spoken answer, which is why voice work concentrates on being the clearest single answer rather than on being present in a list.

• Longer queries, phrased as full questions rather than keyword fragments
• Natural spoken grammar, including filler words a typed search would drop
• Heavy local and immediate intent, especially on mobile
• One answer read aloud, so partial visibility is not a useful outcome

Writing content that survives being read aloud

Read your current service page out loud and the problem usually announces itself. Long sentences with three clauses lose the listener. Bracketed asides make no sense spoken. Bullet lists collapse into a run-on. Copy that scans well on a screen can be unusable in the ear.

So we write differently for these passages. Short sentences. One idea each. The subject named rather than referred to as it or this, because a listener joining mid-answer has no antecedent to attach it to. Numbers and times stated plainly. No formatting that carries meaning the voice cannot convey.

The self-contained test is the one that matters. Take any answer passage out of the page, read it to someone who has not seen the page, and check whether it makes sense on its own. If it needs the heading above it or the paragraph before it, an assistant reading it aloud will produce something confusing, and the passage will usually not be chosen at all.

• Short sentences carrying one idea each
• Named subjects rather than pronouns, so passages stand alone
• Plain statements of times, prices, distances and conditions
• No meaning that depends on layout, tables or nested lists
• Every answer passage tested by reading it aloud in isolation

Question-led headings and the answer beneath them

The structural pattern is the same one that wins question boxes and snippets, because it is the same extraction problem. The heading carries the question in the words a person would speak. The passage directly under it answers that question completely in two or three sentences. Everything else follows afterwards.

We phrase headings the spoken way rather than the keyword way. Not emergency plumbing costs, but how much does an emergency plumber cost. Not opening hours, but what time do you open on Sunday. It feels less tidy than a keyword heading and it matches how the query actually arrives.

Where the question has a conditional answer, we state the condition inside the answer rather than leaving it for later. An answer that says the call-out charge is one figure on weekdays and another at weekends is usable spoken. An answer that says charges vary, see the table below, is not.

The local and business data layer

A large share of spoken queries resolve to a business fact rather than a page: are you open, where are you, how far is it, do you do this thing, what is the number. Assistants pull these from business listings and maps data far more often than from your website copy. If the listing is wrong or incomplete, better writing on the site fixes nothing.

So the data work runs alongside the content work. Categories and services set correctly so the business is considered for the right questions. Hours accurate, including holiday hours, which are a common failure point. Address and phone consistent across the listing, the site and the major directories. Service areas defined honestly rather than covering half the country.

Attributes matter more than they look. Wheelchair access, parking, whether you take card, whether you offer a particular service. These fill the answer slots for questions that are otherwise unanswerable, and most profiles leave them blank.

• Accurate categories, services and attributes on the business profile
• Hours kept current, including holiday and seasonal changes
• Name, address and phone identical across the profile, the site and directories
• Service areas defined to match where you genuinely work
• LocalBusiness and FAQ markup supporting the facts on the page

What can and cannot be measured

This is where we differ from most people selling this service. There is no reporting interface that tells you how often an assistant read your business aloud. Voice queries are not separated out in Search Console. The assistants do not publish source data. Anyone showing you a voice search ranking report is showing you an inference or an invention.

What we can measure honestly is a set of proxies that move for the same reasons. Impressions and clicks on long question-shaped queries in Search Console, which we can filter for. Presence in question boxes and snippets for those spoken phrasings, since assistants often read from the same extracted passages. Calls, direction requests and profile interactions from the business listing, which is where a lot of resolved voice queries land.

And we spot check by asking. Running your priority questions through the common assistants on a fixed schedule and logging what comes back is crude, but it is real observation rather than a modelled number. We report it as what it is: a sample, not a metric.

• No platform reports voice share, so we do not pretend to measure it directly
• Question-shaped query performance in Search Console as the closest available proxy
• Snippet and question box presence on the same phrasings
• Calls, direction requests and profile actions from the business listing
• A logged manual sample of assistant answers, described as a sample

Who this is worth doing for

Voice work pays best for businesses whose customers ask short, factual, urgent questions. Trades, restaurants, clinics, garages, salons, anything with a physical location and immediate need. For these, the answer is usually a single business and the query volume is real.

It pays least for considered purchases with long research cycles. Nobody chooses an enterprise software vendor by asking a speaker. For those businesses, the underlying content improvements still help, but we would be honest that they help through snippets and AI answers rather than through spoken results, and we would price the work accordingly.

The overlap is the reason we rarely sell this alone. Conversational question headings, self-contained answers and clean business data are the same foundations that support question boxes, snippets and AI answers. Voice is one output of that work, not a separate project with its own content.