Check this page with an assistantOpens a chat asking it to summarise this article and name the evidence behind each claim.
Claude opens with the prompt on your clipboard: Anthropic does not support prefilled prompts on the web, and we would rather copy it than ship a button that drops it.
The Query Changed Shape Before the Advice Did
Local search used to be a category plus a place: "plumber Bristol". Conversational local search is a situation plus constraints: "my boiler is leaking, who can come out this evening and won't charge a call-out fee". The second query has no keyword to optimise for and three conditions to satisfy, and satisfying them is the entire game.
Google's Ask Maps makes this explicit inside Maps itself, letting people describe what they want in natural language and having Gemini assemble recommendations. The canonical example that has circulated since launch is asking for a quiet coffee shop good for working, which is a request no keyword field on any profile contains. The system has to infer it from attributes, from photos, and from what reviewers actually wrote.
The practical consequence is that proximity has lost some of its monopoly. A business that is open now, does the specific thing asked for, and has reviews describing that thing can be surfaced ahead of one that is closer but silent on all three. That is good news for well-run small businesses and bad news for anyone whose local strategy has been to exist near the customer and hope.
Constraints Are Answered From Fields, Not Prose
Every constraint in a conversational query maps to something structured: hours answer "open now", service listings answer "do you do X", attributes answer accessibility and amenity questions, and review text answers everything subjective. A blank field is not neutral. It reads as an absence of the thing.
| What the customer asks | What answers it | Common failure |
|---|---|---|
| Open now, open late, open Sunday | Regular and special hours | Holiday hours never set, so the record is wrong |
| Do you do this specific job? | Individually listed services | One broad category instead of the twelve jobs you do |
| Step-free, parking, dog-friendly, wifi | Attributes | Left blank because they felt minor |
| Good for working, good for kids, quiet | Review language and photos | Reviews that say "great place" and nothing else |
| Do you serve my area? | Service area definition and location pages | A vague radius, covered in multi-location SEO |
| What does it cost? | Website pricing that a machine can read | Pricing only available by phone |
Work down that table field by field. It is genuinely a morning of data entry rather than a strategy engagement, which is why it stays undone: nothing about it feels like marketing. The businesses that win constrained queries are usually just the ones whose records are complete, and completeness is a chore rather than an insight.
Reviews as the Constraint Evidence
Subjective constraints have nowhere to come from except what other people wrote. "Quiet", "good for working", "patient with nervous patients", "turned up when they said" are not fields on any profile. They exist in reviews or they do not exist at all.
This reframes review generation from a rating-average exercise into a vocabulary exercise. A five-star review saying "great service" contributes almost nothing to a constrained query. A four-star review saying "came out at 8pm on a Sunday for a leaking boiler and quoted before starting" contributes to at least three different queries. When you ask for reviews, asking people what specifically they needed and whether you delivered produces text that does work, and it is no harder than asking for a rating.
The weighting here is supported by the broader citation research: analysis of AI citations across major assistants has repeatedly found reviews and third-party social proof accounting for a larger share of cited sources than businesses' own pages. Your site describes you; reviews corroborate you, and corroboration is what these systems reach for. The mechanics of asking without breaking platform rules are in how to get more Google reviews.
The Website Half People Skip
The profile is not the whole record. Assistants cross-check what a business claims against what its website says, and inconsistency between the two is a reason to reach for a competitor whose story holds together. Name, address, hours, services, and service area should read identically in both places.
Three specific pieces of website work carry disproportionate weight for constrained queries. First, a page per service rather than a single services page listing twelve things in a bulleted row, because a page can answer a constraint in depth and a bullet cannot; the structure for that is in service page SEO. Second, hours and contact details in visible HTML rather than inside an image or a widget that renders late. Third, structured data that labels what is already visible on the page, following the approach in schema markup for small businesses.
Resist the temptation to add markup describing things the page does not say. Structured data that contradicts visible content is a quality problem rather than a shortcut, and the systems reading it are cross-checking against the rendered page anyway. The rule that holds everywhere in this work applies here too: label what is true, do not annotate what you wish were true.
Testing Whether Any of It Worked
Write down the ten questions a customer would actually ask, constraints included. Run each several times in fresh sessions across the surfaces that matter to you, and log whether you were named and who else was. Repeat monthly with the same questions.
Do this from a clean session and, where possible, without your usual account signed in. Your own phone knows your business, has been to your address, and has searched your name a hundred times, so it is the least representative device you own. The full method, including how many runs are enough and how to log results so the trend means something, is in share of model.
Then connect it to whether anything arrived. Local discovery increasingly resolves without a website visit, so calls, direction requests, and messages are the outcomes worth counting rather than sessions alone. Where visits do happen, make sure they are being attributed rather than dumped into Direct, which is the subject of fixing AI traffic attribution in GA4. And keep the fundamentals current underneath all of it using the local SEO checklist.
Questions People Ask About Local AI Search
- What is Ask Maps?
Ask Maps is Google's conversational search inside Google Maps, powered by Gemini, which lets people ask for local businesses in natural language rather than by keyword. Instead of searching 'coffee shop', someone can ask for a quiet coffee shop that is good for working, and receive AI-generated recommendations that weigh the stated constraints alongside location.
- Does the nearest business still win local search?
Less reliably than it used to. Conversational queries carry constraints, and a business that satisfies the constraints can be surfaced ahead of one that is physically closer but does not. Proximity remains a strong factor, but it now competes against how completely your hours, services, attributes, and reviews answer the specific thing that was asked.
- How do I optimise a Google Business Profile for AI search?
Complete it in a way that answers constraints rather than describes a category. Every service listed individually, hours that are genuinely accurate including exceptions, attributes filled in rather than left blank, current photos, and a description written in the language customers use. The profile is being read as a structured record of what is true about you, so blank fields read as absences rather than omissions.
- Do reviews matter more for AI local search?
Reviews appear to matter a great deal for AI local search, both as a quality signal and as a source of specific language. Research into what AI systems cite has consistently found reviews and social proof accounting for a large share of citations. Reviews that describe the experience in detail supply exactly the constraint-matching evidence a conversational query needs.
- How do I test whether my business appears in AI local results?
Ask the questions your customers would ask, in fresh sessions, several times each, on the surfaces you care about. Log whether you were named and which competitors were. It is the same repeated-sampling method used for any AI visibility measurement, and the discipline matters because a single run of a non-deterministic system tells you nothing.
