When an Assistant Recommends a Shop Nearby

Somebody standing three streets away asks their phone where to buy a replacement watch strap. Nothing about that question mentions your shop, and everything about the answer depends on records you may not have looked at since you opened.

The instinct is to treat this as a new problem needing a new tactic. It is mostly an old problem with a new consumer of the data. The assistant summarising three options for that person is drawing on business listings, map records and the pages of the businesses themselves. Which means a shop that has never thought about search suddenly benefits from an afternoon of unglamorous data hygiene, and a shop that already ranks locally is largely already doing the work.

What decides whether a business gets named at all?

Three things, according to the people running the largest of these systems. Google's own explanation of local ranking names relevance, distance and prominence: how well the profile matches what was asked, how far the business is from where the search happened, and how well known the business is.

Those three are not equally within your control, which is the useful part. Distance is fixed by your address. Prominence accumulates slowly from mentions, links and reviews. Relevance is the one you can change this afternoon, because relevance is decided by what your listing says and how precisely it says it.

The same page carries a sentence worth quoting to anyone who offers to sell you a local ranking: there is no way to request or pay for better local placement. Google says it keeps it that way to make the system as fair as it can be. Every service that promises to move you up the local results is either selling you the underlying data work under a grander name, or selling you nothing.

Why do assistants get local facts wrong so often?

Because they are reconciling several copies of your business that disagree, and none of the copies announces itself as the authoritative one. Your hours exist on a map listing, on your own site, on a directory you forgot about and in a review platform's cached record.

When those agree, an assistant answers confidently and correctly. When they disagree, it picks one, and the one it picks is not necessarily yours. This is the mechanism behind the single most common complaint small businesses have about AI answers, which is being told they are closed when they are open. Nothing invented the wrong hours. Something old was still lying around.

The fix is a reconciliation exercise rather than a tool. List every place your business facts appear, which for most shops is between four and ten locations, then make them identical. Name exactly as it appears on your signage, address in one format, one phone number, and hours including the days you are unusual. We went through the correction mechanics in more detail in how to correct what AI answers say about your business, and the principle there holds: fix the source, not the answer.

Where your facts live, and who controls each copy

This table is the audit worth running once a year. It is assembled from the documentation of each system rather than from one place, because no single provider describes the others.

The second row surprises people. Apple runs its own system, announced on 11 January 2023, and Apple describes Business Connect as a free tool for claiming a location place card that appears across Maps, Messages, Wallet and Siri. Claiming your listing on one map does nothing for the other. A business that has done the Google work and never heard of the Apple one is publishing to half its potential audience with unverified data.

What should you put in structured data?

Less than you fear. Google's local business structured data documentation requires exactly two properties: a name and an address as a structured postal address with street, locality, region, postal code and country.

Everything else is recommended rather than required, and the recommendations are the practical ones: telephone, url, geographic coordinates to five or more decimal places, opening hours, a price range, and for food businesses the cuisine served. If your site collects reviews itself, aggregate rating and review properties belong there too.

The opening hours format deserves attention because it is where most implementations go wrong. Hours are expressed as objects with a day of week, an opening time and a closing time in hours, minutes and seconds. There is provision for the awkward cases: hours that run past midnight, a business open around the clock expressed as 00:00 to 23:59, and seasonal hours bounded by valid from and valid through dates. That last pair is how a shop with summer hours states the fact once rather than remembering to edit it twice a year.

The documentation also covers businesses with distinct parts through a department property, with a naming convention of the store name followed by the department name. A shop with a café inside it is two things to a searcher, and saying so is more accurate than choosing.

The category trap

Choosing a broad category feels safe and does the opposite. A business listed under a general retail category competes with everything, matches nothing precisely, and gets recommended for queries it cannot serve while missing the ones it can.

Narrow beats broad because relevance is a matching problem. Somebody asking for a watch strap is served better by a watch repairer than by a jewellery shop, and the system can only know that if the listing says watch repair. The guidance is to use the fewest categories that describe the core business, which means one primary category chosen precisely, plus additional ones only where they describe something you genuinely do.

There is a related trap in naming. Your listing name should match your sign, not your ambitions. Adding a location or a keyword to the name is a rules violation rather than a clever tactic, and the downside is a suspended listing rather than a slightly lower position.

Do reviews still matter when a machine is summarising them?

More, not less, because a summary compresses many reviews into a sentence and that sentence is what most people will ever see. A pattern in your reviews becomes the pattern in the summary.

This changes which reviews matter. A single furious one star review among forty positive ones has always been survivable. What is newly costly is a repeated theme, because repetition is exactly what a summarising system detects. Five people mentioning parking, over two years, becomes a clause about parking in the description of your business.

Google's guidance lists responding to reviews among the actions that help, which is unusually direct advice from a system that rarely confirms what helps. Replies also give the summariser text from you, written in your words, attached to the complaint. Our approach to replying to negative reviews with AI drafting covers how to do that without sounding automated, which matters when the reader may be a person deciding and a machine summarising at the same time.

Being named is not the same as being booked

The listing systems have quietly grown past description into action, and that is where the commercial difference sits. A place card is no longer only a set of facts. It can carry a button.

Apple's description of Business Connect includes action buttons for ordering or reservations, with named partners handling the transaction behind them, alongside Showcases for seasonal items and promotions. Google's structured data documentation points the same way, noting that local business markup can support reservations or placing orders directly in search results through its Maps Booking API.

For a restaurant, a salon or anything with a diary, that changes the arithmetic of the afternoon spent on this. You are not improving a description that somebody might act on later. You are adding the step where they act. For a shop without bookings the equivalent is narrower but real: a promotion surfaced on a place card reaches somebody who never visited your site.

The caution is that every action button is a commitment. A reservation control that leads to a system nobody checks is worse than no control, in the same way that a phone number nobody answers is worse than no phone number. Add the action only where the process behind it exists.

What if you have no shopfront, or several?

Both cases have a specific answer, and both are handled badly by generic advice written for a single high street location.

If you travel to customers rather than receiving them, the listing systems expect you to state a service area instead of, or alongside, an address. Getting this wrong in either direction costs you. A service business that publishes a home address invites visitors it does not want and may breach the rules it agreed to. A business that publishes neither an address nor a service area gives the distance factor nothing to work with, and distance is one of the three things deciding whether it gets named at all.

If you have several locations, the rule is one profile per location rather than one per business, and the structured data on your site should describe each place rather than averaging them. Google's markup documentation handles the related case of a business with distinct parts through its department property, using a naming convention of the store name followed by the department name. The same instinct applies across sites: each thing that a person could physically visit deserves its own complete record with its own hours.

The failure mode for multi location businesses is a single set of hours applied everywhere, which is correct for none of them the moment one branch closes early. The failure mode for service businesses is an address that a map cannot place, usually because it was typed as a single line rather than entered in the structured fields the form provides.

How would you check what an assistant says about you?

By asking it, from a device that is not signed into your own accounts, in the words a customer would use. This takes ten minutes and almost nobody does it before deciding that AI answers are either a threat or irrelevant.

Ask three kinds of question. A direct one, naming your business, which tests whether the facts are right. A categorical one, naming what you sell and a place, which tests whether you are in the set at all. And a comparative one, asking which of several nearby options suits some specific need, which tests what the summariser thinks distinguishes you.

Write down what comes back, including the wrong parts, then trace each error to a record rather than arguing with the output. Wrong hours point at a listing. A wrong description of what you sell usually points at a category or at your own site saying something vague. Absence from a categorical answer usually points at prominence, which is the slow factor, and at the category choice, which is the fast one.

Repeat it quarterly. The records change underneath you, particularly after any platform migration, and the only way to find out is to look.

What does this actually get you?

Not a flood of traffic. The honest expectation is a small number of very well qualified arrivals, because somebody asking an assistant for a nearby business has both intent and proximity already.

It is also defensive. The cost of getting this wrong is measured in people who were told you were closed, went elsewhere, and never mentioned it. That is invisible in your analytics and it is the reason the work feels optional right up until you discover the scale of it.

Measuring is harder than it used to be, since a referral from an assistant may arrive with no useful source information or none at all. The practical approach is to ask. A single question at the till or in your order form, about how somebody found you, produces data no analytics package currently gives you. We looked at what the measurable part looks like in which assistants actually send referral traffic, and the summary is that the visible share understates the real one.

An afternoon's work, in order

Start with the map listings, both of them, because they are free and they are where the majority of local answers still originate. Verify, then fill every field the form offers, then look specifically at hours for bank holidays and for whatever day your business does something unusual.

Then your own site. Add the structured data, which needs a name and an address to be valid and is better with the recommended fields. Put your hours there in the structured format rather than only in a sentence on the contact page, because a sentence is a guess to any machine reading it. If you are building the site rather than editing one, an ecommerce website builder that emits the business schema alongside the pages saves you the retrofit entirely.

Then the cleanup. Search your own business name and note every listing that appears, including the ones you did not create. Old entries from directories you signed up to once are the usual culprits behind an assistant confidently reporting a phone number you gave up in 2019.

Finally, set a reminder for the two weeks of the year when your hours change. Almost every wrong answer about a small business traces back to a special day nobody updated, and it is the single highest return calendar entry a local business can make.

The summary

Local answers from an assistant are built on ordinary business data, held in several places, controlled by you in most of them and by nobody in the rest. The factors that decide whether you are named are published, the two big map systems are free, and the structured data requirement on your own site is two fields. There is no lever to pull and nothing to buy. There is a list to go through, and most shops have never been through it.