GEO for local businesses: showing up when AI recommends services

Why local is a different GEO problem

When someone asks an AI assistant for "the best emergency plumber in Leeds" or "a family law solicitor near me", the model isn't recalling a fact from training — it's almost always running a live retrieval step. Local queries are time-sensitive and hyper-specific, exactly the kind of thing a model's training data can't answer reliably. So Perplexity, ChatGPT with search, Gemini and Claude with web access fetch fresh results, read a handful of pages, and synthesise an answer from what they find. That means your visibility hinges less on brand fame and more on what's retrievable and corroborated about you right now.

This is good news for small businesses. You don't need to out-spend a national chain on brand awareness to appear. You need to be the most clearly described, consistently referenced, well-reviewed option for a specific service in a specific place. The unit of competition is the query "[service] in [town]", and most local markets have far fewer well-structured competitors than a SaaS category does.

The catch: location adds a verification burden. An AI won't confidently recommend a service it can't pin to a real address, service area and set of corroborating sources. Ambiguity gets you dropped from the shortlist, not ranked lower.

Get your entity unambiguous and consistent everywhere

AI assistants resolve "who is this business" by cross-referencing the same details across many sources. If your name, address and phone number (NAP) differ between your website, Google Business Profile, Bing Places, Apple Business Connect, Yelp, industry directories and your Companies House listing, the model sees conflicting entities and hedges — or picks the competitor it can describe with confidence. Pick one canonical format for your business name, address and phone, and make it byte-for-byte identical across every listing you control.

Claim and fully complete your Google Business Profile, Bing Places and Apple Business Connect. These feed the map and search layers that several assistants draw on, and they establish the structured facts (category, hours, service area, service list) that AIs repeat. Add LocalBusiness schema to your site with the same address, geo coordinates, opening hours, and an areaServed field listing every town or postcode you cover. Schema doesn't get you ranked, but it removes ambiguity and makes the machine-readable facts match the human-readable ones.

Name your service areas explicitly in prose, too. A model matching "near me" needs to see "We serve Chorlton, Didsbury and south Manchester" written out — not inferred from a map pin it may never load.

Win the third-party sources AIs actually cite

Your own website is the weakest possible source for an AI deciding who to recommend, because every business claims to be the best. Assistants lean on third-party corroboration: review platforms, local directories, "best of" roundups, local press, trade-body membership pages, and forum threads like local subreddits or community Facebook groups. When an answer cites sources, these are overwhelmingly what shows up — not the businesses' homepages.

So the practical work is getting accurately described on pages the model trusts and retrieves. Earn listings in the directories and association sites for your trade (FMB for builders, a local chamber of commerce, Checkatrade-style platforms where relevant). Pitch to be included in the "best [service] in [city]" listicles that journalists and bloggers publish — these are retrieval magnets for exactly the queries you want. Where customers naturally discuss your category (a city subreddit, a parenting forum, Nextdoor), being genuinely recommended there is high-value, because models weight community consensus. Do not fabricate or astroturf this; manufactured mentions are easy to spot, brittle, and a reputational risk that outweighs any short-term lift.

Reviews are doing double duty here. Volume, recency and rating all feed the consensus signal, but the text matters most: reviews that mention the specific service and location ("fixed our boiler in Headingley same day") give the model literal phrases to match against a user's query.

Write pages that answer the question, specifically

Most local sites are built to look professional, not to be quoted. To be the sentence an AI lifts into its answer, write content that states plainly what you do, where, for whom, at roughly what cost, and how fast. A page titled "Emergency electrician — Bristol & BS postcodes, 24/7 callout" with a short paragraph confirming response times, the suburbs covered, typical call-out pricing and qualifications gives a model everything it needs to recommend you and justify it.

Build one focused page per core service-plus-area combination rather than a single bloated "Services" page. Each should answer the obvious follow-up questions a customer asks — pricing ranges, availability, what's included, accreditations, guarantees — because those are the dimensions on which an assistant compares options and writes its "why". Specificity is the lever: "qualified, insured, City & Guilds Part P registered, EICR certificates issued same week" is far more recommendable than "experienced and reliable", because it's concrete, checkable, and matches how people phrase their needs.

Keep claims accurate and current. Models increasingly distinguish corroborated specifics from marketing adjectives, and an answer that recommends you will often inherit the exact facts you published — so wrong hours or stale pricing become the AI's mistake, attributed to you.

Track it, because local answers vary by location and change fast

Local AI visibility is harder to monitor than a SaaS keyword for two reasons: answers are personalised to the asker's location, and the live retrieval step means results shift as your sources and reviews change. You can't check it once. Run the realistic queries a customer would type — "best [service] in [town]", "[service] near me", "who should I call for [problem] in [area]" — across ChatGPT, Claude, Gemini and Perplexity, and note whether you appear, in what position relative to competitors, and which sources the assistant cites.

Those citations are your roadmap. If the same three directories or roundups keep getting quoted and you're absent from them, that's the next listing to earn. If a competitor is named and you're not, read the page the model cited and work out what it says about them that isn't said about you. Re-test after each change — new review batches, a fresh listicle mention, an updated service page — to see what actually moved the answer, since the feedback loop here is faster and more measurable than traditional local SEO. A tracker that logs answers and cited sources over time turns this from guesswork into a repeatable loop.