GEO for Local BusinessesGetting Recommended in Location-Based AI Answers

Summary

AI assistants answer local questions with a shortlist of three to five businesses, assembled from map profiles, review platforms, directories, community threads and local press rather than from your website alone. GEO for local businesses therefore means making your entity unambiguous across those sources, earning recent third-party corroboration, and writing service-area copy specific enough to quote. This playbook covers where location-based answers come from, the five signals that decide who gets named, a 30-day plan, and how to measure visibility city by city instead of brand-wide.

What GEO Means for a Local Business

Generative Engine Optimization (GEO) is the practice of getting your brand named, described accurately and cited when someone asks an AI assistant a question. For a local business, that question is almost always some version of “what is the best plumber in Lyon”, “where should I get dinner near Shoreditch”, or “which dental clinic in Austin takes new patients”.

The difference from classic local SEO is structural, not cosmetic. A search results page gives you a map pack with three listings, an organic list below it, and a user who scrolls. An AI answer gives you a paragraph naming a handful of businesses, often with a sentence of justification for each. There is no page two. Either you are in the shortlist or you are invisible for that query.

That compresses the funnel dramatically. In classic local search, being the seventh result still earned some traffic. In an AI answer, being the seventh most credible option earns nothing, because the assistant stopped at five. The upside is that shortlists are far more winnable than a national keyword: the competitive set for “best physiotherapist in Nantes” is small, and most of it has done nothing deliberate about AI visibility.

Where Location-Based AI Answers Actually Come From

Local queries are the category where assistants lean hardest on live retrieval rather than model memory. Opening hours change, businesses close, prices move. Models are trained on data that goes stale quickly, so almost every location-based question triggers a search step before the answer is written.

That search step is where the outcome is decided. The assistant fetches a small number of pages, reads them, and composes an answer from what it found. If your business is not represented in those pages, no amount of on-site optimisation will put you in the response.

In practice, the sources that feed location-based answers cluster into a predictable set:

• Map and business profile data: Google Business Profile, Apple Business Connect and their aggregators, which supply the canonical name, category, address and hours.
• Review platforms: Google reviews, Yelp, Trustpilot, Tripadvisor, plus vertical equivalents like Doctolib, Booking or Houzz depending on your sector.
• Vertical and municipal directories: chamber of commerce listings, professional association registries, industry-specific marketplaces.
• Community discussion: Reddit threads, local Facebook groups that are publicly indexed, forum posts where residents recommend businesses by name.
• Local editorial: city magazines, regional newspapers, neighbourhood blogs and “best of” roundups, which assistants treat as relatively high-trust curation.
• Your own website: which matters most for confirming details and answering the specific question, and least for getting you into the consideration set in the first place.

Notice the ordering. Your website is on the list, but it is the last source to change the outcome, not the first. Local GEO is mostly work you do off your own domain.

The Five Signals That Decide Local AI Recommendations

Across the local queries we see tracked, the same five signals separate the businesses that get named from the ones that do not.

1. Entity consistency. An assistant has to be confident that the “Maison Duval” on Yelp, the “Maison Duval Paris” on Google and the “maisonduval.fr” website are one business. Divergent names, old addresses, three different phone numbers and mismatched categories create ambiguity, and ambiguous entities get dropped rather than guessed at.

2. Review recency and volume. Not just the star rating. A business with 40 reviews from this year reads as active; one with 300 reviews that stop in 2023 reads as possibly closed. Assistants routinely hedge on businesses whose most recent evidence is old.

3. Third-party corroboration. A claim made only on your own site is a marketing claim. The same claim repeated in a local paper, a directory description and a customer review becomes a fact the model will restate. This is the single biggest lever for most local brands, and the one they spend the least time on.

4. Service-area specificity. Pages that say “serving the greater metropolitan area” give a model nothing to work with. Pages that name neighbourhoods, postcodes and travel times give it text it can match to the question and quote back.

5. Machine-readable structure. LocalBusiness schema with address, geo coordinates, opening hours and service area is not a ranking trick. It is the difference between a model inferring your details from prose and reading them from a field.

A 30-Day GEO Playbook for Local and Multi-Location Brands

Here is a sequence that works for a single location and scales to fifty. It is deliberately front-loaded with the cleanup work, because corroboration built on an inconsistent entity does not compound.

• Days 1-5: Audit the entity. Pull every public listing of your business and record the exact name, address, phone, category and hours. Fix every divergence to a single canonical version. For multi-location brands, do this as a spreadsheet with one row per location before touching anything.
• Days 6-10: Fix the profile layer. Complete every field on your Google Business Profile and its equivalents: services, attributes, service area, photos with recent dates. Empty fields are missing answers.
• Days 11-15: Build the answer pages. One page per location, and one page per service crossed with your main city if the volume justifies it. Each page should answer the literal questions people ask: what it costs, how long it takes, whether you take walk-ins, which neighbourhoods you cover.
• Days 16-20: Add structure. LocalBusiness or the relevant subtype schema on every location page, with sameAs pointing at your verified profiles. Add FAQPage markup for the questions you just answered in prose.
• Days 21-25: Trigger corroboration. Ask recent customers for reviews that mention the specific service and the neighbourhood, not just “great service”. Pitch one local publication with something genuinely newsworthy. Claim listings in the two or three vertical directories that matter in your sector.
• Days 26-30: Establish the baseline. Write out the 20 to 40 questions a real customer would ask an assistant before choosing a business like yours, run them across the engines your market actually uses, and record who gets named. This is the number you will improve against.

How to Measure Local AI Visibility City by City

The most common measurement mistake in local GEO is tracking the brand as a single entity. “Are we visible in ChatGPT” is not an answerable question for a business with eight branches, because the answer is different in every city and the average hides the branches that are failing.

Local visibility has to be measured per location and per question. That means a prompt set built from the geography, not from your keyword list: the city name, the neighbourhood names, the “near me” phrasings, and the qualifiers that real customers attach. Open Sunday, English-speaking, wheelchair accessible, accepts new patients.

Four numbers are worth reporting per location:

• Presence rate: the share of tracked local prompts where your business is named at all. This is the headline metric and the one that moves first.
• Position in the shortlist: being named third of five is meaningfully different from being named first, and the gap shows up in click-through to your profile.
• Accuracy: whether the assistant states your hours, services and location correctly. A wrong closing time is a lost customer even when you won the recommendation.
• Source mix: which pages the engines actually cited to build the answer. This tells you where to invest next, and it is usually not your website.

Mistakes That Keep Local Businesses Out of AI Answers

Most local businesses lose AI visibility to unforced errors rather than to competitors outspending them. The recurring ones:

• Treating the website as the whole strategy. Publishing more location pages while your Yelp listing shows a five-year-old address is optimising the least important source.
• Letting old listings survive after a move or rebrand. A stale duplicate profile does not just fail to help: it actively creates the ambiguity that gets you dropped.
• Chasing review volume without recency. Twenty reviews a year forever beats 200 reviews three years ago.
• Writing service-area copy that names no places. “The greater Manchester area” is unmatchable. “Didsbury, Chorlton and Withington” is quotable.
• Ignoring the community layer. A single Reddit thread recommending three competitors can outweigh a year of your content, and it is often answerable with an honest reply from the owner.
• Measuring once. Local answers change as reviews arrive and listings update. A quarterly snapshot cannot tell you whether a drop is real or noise.

Local GEO rewards diligence more than budget. The competitive sets are small, the signals are fixable, and most of the businesses you are up against have not looked at an AI answer about their own city yet. The ones that start measuring now will spend the next two years defending a position rather than trying to win one back.