
“Near me” is not going away, and anyone telling a plumber in Fresno to stop caring about it is selling a different service. People still type it, still say it into phones, and still want a result they can drive to before lunch. What has changed is everything that happens after the first query, and that part decides who gets the call.
Run a few of these searches yourself and the pattern is plain enough to describe. The first question is short. The second and third are long, specific and conversational. The business that survives into the third question is the one that answered the things nobody asked yet.
What a “near me” query is actually asking
Strip the two words and the intent underneath is a shortlist. Someone wants three options they can compare quickly, with enough detail to pick one. A classic search result page gave them ten blue links and let them do the comparing themselves, tab by tab.
An AI answer does the comparing for them. Google AI Overviews, ChatGPT and Perplexity each compress the field into a handful of named places with a sentence or two on why. The shortlist has become somebody else’s editorial decision, and you are either in the sentence or you are not.
There is a second shift hiding in there. A person standing in a car park asking their phone for “a pharmacy near me that’s open now” wants one name and a route. A person at a desk asking the same thing on a Sunday evening is planning Monday, and wants to know about wait times, transfers from another chain and whether the pharmacist speaks Spanish. The two typed identical words. Only the second conversation has room for your detail.
Location used to be a keyword
For years local optimisation meant putting a city name in a title tag and collecting enough citations in directories that the address matched everywhere. Those chores still matter, because consistent name, address and phone details are how machines confirm that two listings describe one business.
The difference is where the work ends. In the old model, ranking third on a map pack still brought clicks. In the new one, an answer that names two places and leaves out the third sends the third no visitors at all. The target moved from position to inclusion.
One search turns into a conversation
Picture someone in Sacramento who asks for a dentist open on Saturday. Google shows an overview. They tap through, and the follow-up questions that carry from AI Overviews into AI Mode arrive with the earlier answer attached, so the second prompt can be “which of those takes Delta Dental and has parking?” without restating anything. Each turn narrows the list.
That changes what a listing has to do. It no longer needs to win a single moment. It needs to hold up through four or five questions, each one drawing on a different fact about the business.
A locksmith shows the problem well. The first answer names three firms. The follow-up asks which of them quotes a price before arriving. If only one of the three states that anywhere a machine can find it, that firm carries the conversation from then on, even if it ranked second on the map.
Longer asks with a pile of needs inside them
People used to trim their searches down to keywords because that worked. Now they write the way they would speak to a friend: a quiet cafe near Lake Merritt with outlets, oatmilk, and room for a group of five.
Google has described AI Mode as built for exactly this kind of request, and the longer, multi-intent queries in Google AI Mode get broken into a set of smaller lookups behind the scenes, one for each need. A cafe page that mentions outlets in a photo caption and group seating in a footnote might satisfy two of the lookups and miss the third.
So the unit of optimisation shifts from the keyword to the attribute. Ask what a person with a messy, multi-part need would want confirmed, then make sure each answer lives somewhere a machine can read it.
A useful exercise takes twenty minutes. Write down five real requests a customer might make, each with three or four needs stacked inside. Then check whether your profile, your site and your reviews each answer every need, in words and not just in a photo.
What the listing has to answer in advance
Google Business Profile is the first place those lookups land, so it deserves a slow review rather than a quick tidy. Check these:
- Primary category, chosen for the main service, with secondary categories added only where they are true
- Service area or address, matching the website to the letter
- Hours, including holiday hours entered before the holiday
- Attributes such as wheelchair access, walk-ins welcome, outdoor seating or EV charging on site
- Products and services entered as actual items with plain descriptions
Seasonal businesses should treat the profile as a living document. A garden centre that changes stock every quarter, or a tax preparer who works different hours in March, needs the profile to say so on the day it changes.
Photos count here too. A picture of the entrance, the parking lot and the waiting room answers questions in advance, and a model that can read image captions or alt text has something concrete to quote.
Reviews are the evidence layer
When an answer says a restaurant is good for groups, that claim came from somewhere, and the likeliest source is the review text. Google reviews, Yelp pages and neighbourhood threads on Nextdoor all supply the sentences a model reaches for when it needs a reason.
Read your own reviews the way a machine would. Look for the specifics customers volunteer: the wait, the staff member’s name, the thing that went wrong and how it got fixed. A review that says “great place” adds almost nothing. One that says “they had a car seat installed in ten minutes and the receptionist found parking for us” adds three usable facts.
Owner replies matter in the same way. A reply that says “thanks for coming in” adds nothing, while one that confirms the Saturday hours, names the service that was performed and apologises for the specific delay puts a second set of facts next to the first.
You cannot script this, and you should not try. What you can do is ask better questions at the point of service, such as “what did you come in for today?”, so customers have something specific to write about.
Your own site still has a job
The website is the source of record when a model needs to check a fact it saw elsewhere. Give each location its own page with the address, phone, hours and a few honest paragraphs about what that branch does. Add LocalBusiness markup from schema.org so the details are labelled, not guessed.
Page speed and a clean mobile layout still matter, since a visitor arriving from an answer card is on a phone and gone within seconds if the phone number is buried.
Look at robots.txt before anything else on a site with a lot of plugins. A blanket rule can lock out the very crawlers whose answers you want to appear in. Resist the urge to publish dozens of near-identical pages that swap one suburb name for another. They read as filler to a person and add no new fact for a machine. A page for the Oakland branch should say something only the Oakland branch can say, such as the street it sits on, the BART station a short walk away and which staff work there. A mechanic’s page can name the cross street, the free two-hour parking behind the shop and the loaner cars on offer, and that detail answers a follow-up that a generic service page cannot.
And keep the facts on the page and in the markup identical to the listing, since a mismatch in opening hours teaches a model that your information cannot be trusted.
What nobody can tell you yet
No one outside Google can say how AI Mode weighs a Business Profile against a review thread against a web page for a given local question, and the weighting probably shifts with the type of business. Dentists, restaurants and locksmiths may all get different treatment.
The part that holds up is narrower. A business that answers its customers’ follow-up questions in writing, on the profile, in reviews and on its own pages, with the same facts each time, gives a model more to work with than one that did a single thing well. Whether that wins the answer, nobody can promise, and the platforms keep changing under it. Test it in your own market with your own searches, and expect to revise.

