GEO for local businesses: getting recommended nearby

A year is a long time in local search. Birdeye's 2026 research measured consumers using AI tools for local business recommendations at 45 percent, up from 6 percent twelve months earlier, a sevenfold jump that outpaces AI adoption in software buying and shopping alike. The same research places AI as the third most-used source for local recommendations, behind only Google and Facebook and already ahead of Yelp and Tripadvisor, platforms that spent two decades building that position. Meanwhile the traditional channel is transforming underneath: 2026 local tracking data shows Google's AI Overviews appearing on a large and growing share of local queries, heavily skewed toward informational questions over simple transactional lookups.

Why local moved fastest is worth understanding, because it shapes the playbook. Local decisions are constraint bundles: open now, takes my insurance, handles my kid's allergy, comes out on a Sunday. Search made people decompose that into filters and tabs. An assistant takes the whole sentence and returns three names. For a time-pressed person with a burst pipe, that interaction is simply better, and the near-me habit it replaces was enormous, running at hundreds of millions of U.S. searches per month, the great majority on mobile.

The shape of Google's shift matters as much as its size. The 2026 tracking data shows AI Overviews appearing far more often on informational local questions, the "how much does a root canal cost" and "how do I choose a roofer after hail" class, than on simple transactional lookups where the map pack still rules. That split is a content assignment: your service pages and FAQs compete for the informational layer where the composed answer now sits, while your profiles and reviews compete for the transactional layer beneath it. Businesses that publish real answers to the questions people ask before they pick a provider get into the conversation one step earlier than the competitors waiting at the map pack.

The strategic frame for owners: the recommendation slot is narrower than a map pack. An AI answer names two or three businesses where a map showed twenty pins. Whoever occupies those named slots collects outsized demand, which is the general dynamic covered in what is AI visibility, applied here to a radius of a few miles.

The technical shift behind local GEO is that AI answers now apply geographic context and local entity data rather than answering placelessly. Ask ChatGPT for a plumber and it uses your location context, searches, and composes from local sources. Google's AI Mode and AI Overviews draw on the business data and reviews behind Google Maps. Perplexity resolves place entities and cites review platforms. The engines stopped treating "near me" as a keyword and started treating it as a filter over structured entity data: who you are, where you are, what you do, when you are open, and what hundreds of customers said about it.

This is good news for small operators, because entity data is the rare GEO input a business fully controls. An engine deciding between plumbers is choosing among database rows before it chooses among brands. Rows with complete attributes, consistent details across sources and rich recent reviews are safe to recommend; rows with conflicting hours and a two-year-old last review are risky, and engines composing a single confident answer avoid risk. The vertical's core discipline is therefore boring and decisive: make your entity data complete, consistent and current everywhere it lives.

How Google specifically assembles local AI answers, and why its Overviews behave differently on informational versus transactional queries, is unpacked in our Google AI Mode explainer.

Build your question list from real intake: the things customers say on the phone, in booking forms and in reviews. Local questions come geo-qualified and constraint-loaded, and each type leans on different data when the engine composes its answer.

Start where the engines start. Google Business Profile: every field completed, services itemized, attributes set, hours accurate including holidays, photos current, and the business description written in plain specific language rather than slogans. Then Bing Places, the listing most owners forget, which matters because the ChatGPT ecosystem grew up on Bing's index; a business invisible in Bing is handicapped in the fastest-growing assistant. Then Apple Maps, and the two or three directories that dominate your category, whether that is Healthgrades, Avvo, Houzz or OpenTable.

Consistency across all of them is a ranking factor in the oldest local SEO sense and a trust factor in the new entity sense. Name, address, phone, hours and service list should match everywhere, because engines cross-reference sources before naming you, and contradictions read as uncertainty. On your own site, LocalBusiness schema with geo coordinates, service area, opening hours and sameAs links tying your profiles together turns your homepage into the canonical record that reconciles the rest.

Crawl access applies to local sites too. OpenAI's search crawler has roughly tripled its crawl since August 2025 according to Botify, and cheap hosting or aggressive firewall plugins on small-business sites block AI crawlers silently and often. Allow OAI-SearchBot and PerplexityBot in robots.txt, and keep menus, price lists and schedules as HTML text rather than PDFs or images, because an engine cannot recommend the Tuesday class it cannot read.

For a national brand, the third-party evidence engines cite is spread across Wikipedia, press and review platforms. For a local business it is overwhelmingly reviews. They are the only large body of independent text about your business, and engines mine them for exactly the constraints in the question: a review that says the crew arrived within an hour on a Sunday answers the urgent-need question; a review that names the gluten-free menu answers the constraint match. Volume, recency and specificity all matter, and specificity is the lever owners underuse.

The operational habit is simple: ask every satisfied customer for a review, and ask them to mention the service and the neighborhood. Never script the text, but the prompt "it helps if you say what we did and where" is honest and transforms the corpus. Respond to reviews, including critical ones, with substance, because responses are part of the crawlable record and engines quote resolutions. And spread reviews across platforms in proportion to your category: Google reviews for everyone, plus Yelp for restaurants and home services, plus the vertical directories where your buyers check credentials.

Community threads are the second corpus. City and neighborhood subreddits, local Facebook groups and Nextdoor threads get cited when engines want opinion, and 5W Research puts Reddit near 12 percent of all U.S. ChatGPT citations. When your business gets recommended organically in those threads, that text works for you for years. Participate as yourself, disclosed, answering questions in your specialty without pitching; locals and engines both punish astroturf.

Most local websites are brochures: a homepage, a services blur, a contact form. The GEO-ready structure is one page per service per area, each answering one question completely. A page titled "Emergency plumbing in Mesa: response times and honest pricing" that opens with the actual price range, states the average response time, lists the neighborhoods covered and ends with a short FAQ gives an engine everything it needs to name you for the urgent-need question. The Princeton GEO study found statistics, quotations and cited sources lift generative visibility by up to 40 percent, and a local business has all three in its filing cabinet: jobs completed, years in the area, response averages, and customers willing to be quoted by name.

Two cautions. First, do not spin up hundred-city template pages with a find-and-replace town name; that is the local dialect of keyword stuffing, which the same Princeton research measured below baseline, and thin duplicate pages poison the trust you are building. Write pages only for areas you genuinely serve, with details only someone who works there would know. Second, publish prices or at least ranges. Price questions dominate local asks, engines prefer sources that answer them, and the competitor who publishes a range gets named in the answer that quotes one. Your FAQ blocks should come from the phone log rather than a keyword tool: the questions people ask your front desk, in their own words, are the questions they now type into assistants, and a page that answers them verbatim is retrievable in exactly the register the query arrives in. The structural craft transfers directly from GEO vs SEO.

Everything above compounds, and complicates, at ten locations. The entity rule becomes a data-management rule: every location needs its own complete profile set, its own LocalBusiness schema with distinct coordinates and hours, and its own service-plus-area pages with details that prove local knowledge, because a shared template with the city name swapped fails the thin-page test in every market at once. Reviews must accrue to the right location entity, since an engine recommending your downtown branch on the strength of reviews for the suburban one will eventually contradict itself in front of a customer.

Measurement scales the same way: the question list belongs per location, or at least per market, because AI answers apply geographic context and a brand that wins in one metro can be invisible two towns over. The franchise advantage is that the loop, once designed for one location, is a template: the question patterns, the page structure, the review ask and the profile checklist all replicate, and the per-market mention rates tell headquarters exactly which locations need which fix rather than funding another round of generic brand content.

Here is the playbook compressed into a single week for a single location, using a physical therapy clinic as the stand-in. Monday: write the question list from the front desk's actual phone log, fifteen questions like "physical therapist near the university who works with runners and takes Aetna," then run the baseline across engines and store the answers. Tuesday: entity pass. Complete every Google Business Profile field, claim Bing Places, fix the hours mismatch between the site footer and the profiles, add LocalBusiness schema with sameAs links. Wednesday: review pass. Message the last twenty discharged patients with an honest ask, respond to the three unanswered reviews, and note which review phrases match the question list.

Thursday: publish one page, the highest-value lost question from Monday's baseline. If the lost question was about runners and insurance, the page states which insurers you take, what an initial running assessment costs and includes two named patient quotes, then goes to Google and Bing for indexing the same afternoon. Friday: read the citations on the remaining lost questions and start the one off-domain move they point to, often a directory profile or a local guide that keeps getting quoted. That is the whole week, repeatable monthly, and every move traces back to a stored answer rather than a hunch.

AI answers vary run to run, SparkToro measured under a 1 percent chance that two identical ChatGPT runs return the same brand list, and local answers add the volatility of hours, seasons and fresh reviews. So measurement means repetition: a fixed list of 10 to 20 geo-qualified, unbranded questions, run on a schedule across the engines your customers use, with raw answers stored and mention rates trended over weeks. Exclude branded questions from the score; if the question names your business, the answer will too, and the number becomes flattery.

Then the loop: for each question that consistently names competitors, read the citations, decide whether the gap is entity data, reviews or a missing page, fix that one thing, and recheck. Reachroller automates the cycle end to end: scheduled runs, mention rates with per-question receipts, the citations behind every answer, and a publish-ready fix page generated for the questions you lose, slug and schema included. Starter is $29 per month for 400 credits and 25 tracked questions. The honest caveat: it is a young product, ChatGPT tracking is live today and the remaining engines are rolling out. Start with the free check, three days and 50 credits, and see whose names come back when your town asks for what you do.