- Assistants answer "near me" questions by cross referencing several sources, so the shop with consistent facts everywhere beats the shop optimised in one place.
- SOCi's 2026 Local Visibility Index puts business profile accuracy at 68 percent on ChatGPT and on Perplexity, against 100 percent on Gemini, which reads Google Maps directly.
- The same index has ChatGPT recommending 1.2 percent of business locations and Gemini 11 percent, so being mentioned at all is the rare event.
- Google says local ranking rests on relevance, distance and prominence, and that no payment improves it.
- Structured data on your own site is the one surface you fully control and the one most small shops leave empty.
- The work is boring: one canonical set of facts, propagated everywhere, checked quarterly.
Someone three streets away wants a place to get a bike wheel trued this afternoon. Ten years ago they searched. Five years ago they looked at a map. Today a meaningful share of them ask an assistant, read one paragraph naming two or three businesses, and go to one of them. If your shop is not in that paragraph, nothing about the interaction tells you that you were considered and passed over. There is no impression, no position, no report.
AI local search is a genuinely new problem for a business with an address, and most of the advice written about it is recycled search optimisation with new vocabulary. The mechanics are different enough to be worth setting out properly.
Where does an assistant get its facts about you?
From several places at once, and which ones depends on the assistant. That single sentence explains almost everything else.
Gemini is grounded in Google Maps, so what it says about your opening hours is what your Business Profile says. ChatGPT and Perplexity work differently: they cross reference business information across Google Maps, Yelp, Facebook and brand websites, then produce an answer from whatever they reconcile.
That difference has been measured. SOCi's 2026 Local Visibility Index reports business profile accuracy of 68 percent on ChatGPT and 68 percent on Perplexity, against 100 percent on Gemini. The gap is not a difference in model quality. It is what happens when a system consults four sources that disagree with each other about your phone number.
Read that as a merchant and the instruction is unusual. The lever is not making one profile excellent. It is making every copy of your facts say the same thing, because the platforms with the accuracy problem are the ones doing the reconciling.
How often does a shop actually get named?
Rarely, and the numbers are worth sitting with. The same index puts ChatGPT's recommendation rate at 1.2 percent of business locations and Perplexity's at 7.4 percent, with Gemini at 11 percent. Google's traditional local pack, by comparison, shows businesses 35.9 percent of the time.
Two conclusions follow, and they pull in opposite directions. The first is that this is not yet where your customers come from, so treating it as an emergency is a mistake. The second is that the shortlist is short, which means the gap between being on it and being off it is enormous for the businesses that make it.
The reasonable posture is neither panic nor indifference. It is to make sure the cheap, permanent things are right, and then to stop, because there is no lever here worth a monthly retainer.
What does Google say it is actually ranking on?
Three things, stated plainly and unchanged for years. Google's own guidance says local results are based mainly on relevance, distance and popularity, and adds that there is no way to request or pay for a better local ranking.
Relevance is how well your profile matches what was asked, which is a polite way of saying that a profile with thin information cannot match much. Distance is not yours to control. Prominence is how well known the business is, and Google explicitly connects more reviews and positive ratings to better local ranking.
Everything an assistant does downstream inherits from data shaped by those rules, whether it reads Maps directly or reconciles Maps against three other places. So the local basics are not superseded by the new surfaces. They are the input to them.
What each surface is for
| Surface | Read by | You control | How fast a change lands |
|---|---|---|---|
| Google Business Profile | Google Search, Maps, Gemini | Everything, directly | Hours and minutes |
| Apple Maps place card | iPhone users, Siri | Everything, after verification | Days |
| Bing Places | Bing, and assistants using its index | Everything, after verification | Days |
| Your own site markup | Every crawler, no exceptions | Everything, immediately | Next crawl |
| Directories and review sites | The reconciling assistants | Partly, one claim at a time | Weeks, if ever |
The last row is the one that ruins people. A wrong phone number on a directory you have never heard of, published in 2019, is still being read. You cannot fix all of them, and chasing every one is a good way to spend money on a subscription that mostly reports the problem back to you. Fix the ones with traffic, get your own site right, and accept a long tail of noise.
The one surface you fully own
Your website is the only place where nobody has to approve your change and nothing has to be reconciled. It is also, in most small shops, the emptiest.
Google's structured data guidance for a local business names two required properties: the business name and a full postal address, broken into street, locality, region, postal code and country. The recommended list is where the value sits. Opening hours as an OpeningHoursSpecification with day, opening and closing times, and optional valid from and valid through dates for seasonal changes. Geographic coordinates to at least five decimal places. A telephone number. A price range under 100 characters. For food businesses, a fully qualified menu URL. For businesses with distinct departments, nested entries carrying their own hours and phone numbers.
Read that list against a typical small business site and the gaps are obvious. Hours live in an image, or in a sentence that says "open late on Thursdays". The address sits in a footer as one line of text. The phone number is a link and nothing more. Every one of those is a fact a machine cannot extract reliably, on the one page where you had complete control.
The seasonal hours fields are the underused ones. If you close for two weeks in August or open late in December, valid from and valid through express that properly, and a system reading your page in July learns about August. Without them, the only correct answer about your holiday hours exists in a post nobody will parse.
What actually changes an answer?
Consistency first, completeness second, reputation third. That order is unfamiliar to anyone who has read conventional local optimisation advice, and it follows directly from how the reconciling assistants work.
Consistency. Write down your canonical facts once: legal name, trading name, address exactly as you want it rendered, phone, hours, categories. Then make every surface match that document, including the punctuation. Suite 4 and Unit 4 are two different businesses to a matcher. This is dull, it takes an afternoon, and it is the single highest return action available.
Completeness. A profile with three attributes filled can only be relevant to three kinds of question. Every service you offer, payment method you accept, accessibility feature you have and category you fit is a query you become eligible for. Google's own relevance advice reduces to this.
Reputation. Reviews carry more weight here than in general search, because prominence is one of the three named factors and because reconciling assistants lean on review sites for anything qualitative. What people say matters as much as the star average, since the qualitative claim in an answer has to come from somewhere. If you have never read your own reviews as data, our piece on what review analysis actually surfaces and what it misses is a practical starting point.
Which profiles are worth claiming, and in what order?
Three, and the order is set by how many people each one reaches through your door rather than by how modern it feels.
Google Business Profile comes first and it is not close. It feeds Search, it feeds Maps, and it is the direct grounding for Gemini, which is the reason Gemini's accuracy figure sits at 100 percent while the reconciling assistants sit at 68. One profile, edited by you, read verbatim by the platform most of your customers still start on. If you only ever do one thing from this article, make this profile complete rather than merely present.
Apple Maps comes second if a meaningful share of your customers carry an iPhone, which for most consumer facing businesses in wealthy markets means it comes second. It is a separate claim with separate verification, and the place card it produces is what appears when somebody asks Siri or taps a pin. Businesses that skip it usually do so because they assume Apple pulls from Google, and it does not.
Bing Places comes third, and its importance is entirely indirect. Almost nobody chooses Bing. But the index behind it is what several assistants search when they go looking for a page about you, so a verified listing there is a structured, consistent record sitting in exactly the pool the reconciling systems read. It takes twenty minutes and you will never think about it again.
Everything after those three has sharply diminishing returns. Industry directories matter where your customers actually browse them, which for some trades is a real thing and for most retail is not. The test is whether you have ever received an enquiry from one.
What quietly breaks a listing
Four things, and none of them announce themselves. This is the part where a shop that did the work two years ago finds out it stopped being true.
Hours drift. You changed your Sunday opening in spring and updated the one profile you think about. The other surfaces still say the old time, and a reconciling assistant now sees a disagreement it will resolve by weight of repetition rather than by recency. Wrong hours are the single most common error in local data and the one that costs a visit outright.
A category change you did not make. Platforms restructure their category trees, and a business that fitted a category cleanly can be moved into a broader parent without notice. The symptom is a slow decline in appearances for the specific thing you do, which is almost impossible to attribute unless you check the category itself.
Duplicate listings. These appear from a previous owner, a data aggregator, or your own second attempt at claiming a profile years ago. Two records for one address is the exact condition that lowers a reconciling assistant's confidence, and duplicates are much harder to remove than to create. Search your business name and address on each platform once a year specifically to look for a second copy of yourself.
A phone number change. The worst of the four, because the old number propagates into records you will never find and a customer reaching a dead line does not call back. If you change a number, keep the old one forwarding for a year and treat the propagation as a project rather than an edit.
None of these need software to catch. They need a calendar entry and half an hour, which is why the quarterly check below is the whole maintenance plan.
How do you check any of this?
By asking, in the way a customer would, and writing down what comes back.
Pick five questions a real customer might type. Not your business name, which proves nothing. Questions like the best place near a landmark for whatever you sell, or somewhere open on a Sunday for it. Run each one on the assistants your customers plausibly use, and record three things: whether you appear, what facts about you the answer states, and which businesses appear instead.
The second column is the useful one. A wrong closing time in an answer tells you exactly which source is stale, because you can go and find the surface still carrying it. Perplexity makes this easiest since it shows its sources beside the answer, so the audit is close to free there.
Do it quarterly. Not weekly. The variance in these answers is high enough that a weekly check mostly measures noise, and the underlying data moves slowly.
What this shares with everything else on the machine readable web
The same principle keeps reappearing across surfaces that look unrelated. Facts a machine can extract get used. Facts expressed as prose, images or design do not. A shop's opening hours in a graphic, a product's material in a paragraph, a return policy on a page with no markup: all invisible in the same way and for the same reason.
We went through the general version of this in a piece on what actually moves your citation rate in AI answers, and the local case is the strictest instance of it, because the facts are checkable and the assistant will simply prefer whichever version it can verify against several sources. If your business runs on appointments, the same data discipline carries into the booking flow, which we covered in what actually gets a booked customer to turn up.
For the part you own outright, the site itself, the requirement is unglamorous: server rendered pages with correct markup and facts that match your profiles. That is a property of how the site is built rather than something you can bolt on later, which is the case we make for a site whose structure and data stay under your control.
A caution about the industry growing around this
A category of service has appeared promising to get you recommended by assistants, and it is mostly selling the local optimisation that already existed with a new label on the invoice. The genuinely new work in this article is one afternoon of writing down your facts and propagating them, plus a quarterly check. There is no repeatable lever that justifies an ongoing fee, because the systems reconcile public records and the records are already yours to fix.
Be especially wary of anyone offering to place you in answers. The mechanism they would need does not exist publicly, and Google states outright that ranking cannot be bought. What can be bought is a report telling you what you would have found by asking five questions yourself. If you want a rule for the whole category: any local visibility service that cannot show you which specific source carries a wrong fact about your business is selling you a dashboard rather than a fix, because naming the stale record is the entire job.