Voice search never became the majority of local queries the early predictions claimed. What it became is the format people use at the moment they are ready to act: in the car, hands covered in flour, walking with the phone in a pocket. Low volume, high intent, unusually unforgiving about profile detail.
How a spoken query differs from a typed one
People type in fragments and speak in sentences. "dentist manchester" typed becomes "which dentists near me are open on a Saturday morning" spoken. Longer, grammatical, and usually containing a constraint that a keyword list would never capture.
Three differences matter for the work:
- Questions instead of nouns - spoken queries begin with who, what, where, when and can I far more often than typed ones do.
- Implicit location - "near me" is often not even said. The device knows where it is and applies that silently, which makes proximity even more decisive than usual.
- Immediate constraints - open now, still serving, does it have parking, do they take card. These map onto profile fields, not onto content.
The last point is the whole practical lesson. Optimising for voice is mostly a data completeness exercise on the profile, with a smaller content component behind it.
The single-answer problem
A screen shows ten results. A speaker reads one, occasionally offering a second and third if the user asks. There is no page two, no scrolling, no chance to be noticed at position six because the listing looked interesting.
Being in the pack is not enough here; being the first thing selected is. And selection is influenced by things unrelated to content quality - how close the business is, whether it is currently open, whether the assistant has enough confidence in the data to commit to reading it out.
There is a limit worth stating plainly. No amount of optimisation wins the "nearest one that is open" query for a business that is genuinely further away. What is winnable is every query where the constraint is something other than distance.
Profile fields that voice assistants read aloud
When an assistant answers a local question, the sentence it produces is assembled from a small set of fields. Worth knowing exactly which ones, because these are the ones to get right first.
| Field | How it gets used |
|---|---|
| Business name | Read aloud verbatim. This is why stuffed names are a liability. |
| Primary category | Determines whether the business is considered for the question at all. |
| Hours, including special hours | Produces "open now", "closes at six", "opens again tomorrow at nine". |
| Address and service area | Used for the distance statement and for driving directions. |
| Phone number | The action. A wrong number here fails silently and permanently. |
| Attributes | Answers parking, accessibility, payment methods, outdoor seating, appointment requirements. |
| Rating and review count | Often read as justification: "rated four point six from over two hundred reviews". |
The business name deserves its own note. "Smith Plumbing - Emergency Plumber Leeds 24 Hour" violates Google's naming guidelines and is a suspension risk, but it also fails on its own terms the moment an assistant reads it out. Hearing that sentence spoken makes a business sound like spam, which is the opposite of what a name is for.
Attributes are the most neglected item on that list and the cheapest to fix: checkboxes that take twenty minutes per location and answer the constraints people speak aloud.
Question-shaped content that gets quoted
Some voice queries cannot be answered from the profile: "how much does a boiler service cost in Bristol", "do I need to be home for a survey". Those get answered from web content, and the passage that gets read has recognisable characteristics. It is short, it directly restates the question, and it gives the answer before the context.
Collect the questions from the phone, not from a tool
Ask the client what people ring up and ask before booking. That list is better than any keyword export, because it is already in spoken language and it is already about their business. Twenty questions is plenty to start.
Answer each one in about forty words, answer first
Use the question as a heading, written the way it is spoken. Then answer it immediately. "A standard boiler service costs between X and Y in most of Bristol and takes about an hour." Detail can follow underneath; the first sentence has to stand alone if it is read out with nothing after it.
Put the same questions in the profile's Q&A section
Owners can post and answer their own questions, and most never do. It is the one place where an answer sits directly on the listing rather than a click away, and it fills the gap before a member of the public posts something you would rather not have there.
Hours, parking and the boring details that win
The unglamorous fields decide more voice queries than anything else, and they are the ones that rot. A short maintenance list, in order of how often it causes a problem:
- Special hours for holidays - set them in advance for the whole year. A profile that says open on a bank holiday when the shop is shut produces an angry review and, worse, a customer who does not come back.
- More hours - separate fields exist for kitchen, delivery, drive-through, senior hours and similar. A restaurant whose kitchen closes two hours before the bar needs both stated.
- Parking attributes - free lot, street, paid, accessible. This is a genuinely common spoken question and it takes a minute to answer permanently.
- Accessibility - entrance, toilet, seating, car park. People ask assistants about this specifically because asking a stranger is tiring.
- Payment and booking attributes - card, contactless, appointment required, walk-ins welcome.
- Service area - for businesses without a shopfront, an accurate area rather than an optimistic one. Overstating it dilutes relevance for the places you actually serve.
Across a portfolio the failure is never one missing field, it is drift: forty locations where hours were correct at onboarding and are wrong at nine of them now. A multi-location dashboard like Ampli5 Pulse is useful here mainly for making that visible, so hours and attributes get checked in one pass rather than location by location.