How to Rank Your Local Business in ChatGPT. The Complete Citation Playbook.

Why Local Businesses Are Structurally Advantaged Here

Ecommerce brands compete for AI citations against the entire internet. A supplement brand trying to get named in "best magnesium supplement" is fighting Healthline, Reddit, dozens of affiliate roundups, and every competitor with a content budget. The corpus is enormous and the competition is national.

Local is different in three ways that all work in your favor.

The corpus is small

There is not that much written about plumbers in Pueblo. There are a handful of directory listings, some reviews, maybe a local news mention, and a couple of Reddit threads. When a model assembles an answer about your category in your city, it is working from a genuinely thin pile of source material. A single well-structured page that directly answers the question can carry disproportionate weight, because there is so little competing for the same slot.

The intent is unambiguous

Nobody asks an AI assistant for a local service provider casually. Location-qualified service queries are near-transactional by definition — the person has a problem and is looking for who to call. The conversion value of a single citation is far higher than a comparable ecommerce mention, because you are being recommended at the moment of decision rather than at the top of a long consideration funnel.

Your competitors are not doing this

National ecommerce brands have been optimizing for AI search for two years. Local service businesses, overwhelmingly, have not. Most are still running the same Google Business Profile setup they built in 2019 and treating AI search as somebody else's problem. That gap is the opportunity, and it is closing, but it has not closed yet.

The Three Sources LLMs Actually Pull Local Answers From

Before you build anything, it helps to understand where the answer is coming from. When a model responds to a location-qualified query, it is drawing on three distinct pools, and they behave very differently.

Source one: parametric memory

This is what the model absorbed during training. For a local business, this is almost always stale, incomplete, or entirely absent. Small businesses generate too little text to be reliably encoded. If a model "knows" your business from training alone, that knowledge is probably a year or more out of date. You cannot influence this directly and you should not try to.

Source two: live retrieval

Most consumer AI products now run a live search when a query has local or time-sensitive intent, then summarize what comes back. This is where the majority of local citations are actually won, and it is the pool you have the most control over. If your pages are crawlable, structured, and directly answer the question being asked, you are eligible. If they are not, you are not.

Source three: structured knowledge and aggregators

Business directories, mapping data, review platforms, and knowledge graph entries. This is the corroboration layer — the thing that tells a model your business is real, operating, and located where it claims to be. Inconsistency here is the single most common reason a legitimate business gets skipped in favor of a competitor.

The practical upshot: layers 1 through 5 of the stack feed pool three. Layers 6 and 7 feed pool two. You need both. Businesses that only do directory work get corroborated but never surfaced. Businesses that only publish content get surfaced but never trusted. For a deeper technical read on how these retrieval mechanics differ across engines, our complete definition of generative engine optimization breaks down the underlying process.

Layers 1 and 2: The Entity Anchor and NAP Consistency

Everything else in this playbook sits on top of these two layers. Get them wrong and the rest of the work underperforms in ways that are almost impossible to diagnose later.

Layer 1: Google Business Profile as the entity anchor

Your Google Business Profile is not just a Google asset. It is the most widely syndicated structured record of your business in existence, and it propagates into datasets that feed well beyond Google's own products. Treat it as your canonical entity definition rather than as a listing.

The fields that matter most for AI citation, in priority order:

• Primary category. This single field does more to determine which queries you are eligible for than anything else on the profile. Choose the most specific category that genuinely describes your main service, not the broadest one. Broad categories dilute.
• Secondary categories. Add every one that is accurate, but stop before you add ones that are merely adjacent. Category padding measurably weakens the primary signal.
• Business description. Write this as an entity definition, not marketing copy. State plainly what you do, who you serve, where you serve them, and what distinguishes you. Models lift this language almost verbatim.
• Services and service areas. Enumerate them explicitly rather than leaving them implied. Each named service is a retrievable attribute.
• Attributes and photos. Underused and disproportionately valuable. Note that the Google Business Profile Q&A section was discontinued through late 2025 and is effectively gone in 2026 — the question-answering role has moved to an AI feature that pulls from your other profile fields and your website, which raises the value of everything else on this list.

Layer 2: NAP consistency across the aggregator layer

NAP stands for name, address, phone. The requirement is boring and absolute: these three data points must match character-for-character everywhere they appear. "Suite 200" and "Ste. 200" are different strings. "&" and "and" are different strings. A model encountering conflicting records has no way to resolve which is correct, and the safe behavior when facing an unresolvable conflict is to recommend a business whose data is unambiguous instead.

You do not need to be listed everywhere. You need to be consistent everywhere you already appear. Auditing and correcting fifteen existing listings beats creating fifty new ones.

Layer 3: Review Corpus Depth and Recency

Reviews function differently in AI search than in traditional local ranking. In the map pack, review count and average rating are ranking inputs. In an AI answer, the review corpus is source material — the model is reading what people actually wrote and using it to characterize you.

That changes what you should optimize for.

Semantic coverage beats raw count

A business with 90 reviews that collectively mention response time, pricing transparency, cleanliness, specific services, and specific neighborhoods is far more citable than a business with 400 reviews that all say "great service, highly recommend." The first corpus lets a model answer a specific question. The second one does not.

When you request reviews, ask an open question rather than a closed one — "how did it go?" produces richer text than "please leave us a review." Keep the request neutral: Google overhauled its review policy in April 2026, and staff review quotas, asking customers to name specific employees, on-premises review kiosks, and incentivized reviews are now explicit violations enforced automatically. Prompting for detail is fine. Directing what the review should say is not.

Recency is a live signal

A cluster of recent reviews signals an operating business. A profile whose most recent review is fourteen months old reads as possibly defunct, and models are conservative about recommending businesses that may have closed. A steady trickle beats a burst followed by silence.

Your responses are indexed too

Owner responses are part of the corpus. A thoughtful, specific response to a negative review does more for your citation profile than the negative review costs you, because it demonstrates operational competence in text a model can read. Respond to everything, and respond with substance rather than templates.

Layer 4: Schema and the sameAs Graph

This is the layer where most local businesses have literally nothing, and it is the cheapest one to fix. A single block of JSON-LD in your site's head section, written once, does the work.

The job of schema here is not to make you rank. It is to make you unambiguous — to state in machine-readable form that this website, this Google Business Profile, this Facebook page, and this Yelp listing are all the same entity. Without that statement, they are four unconnected records that a model has to guess about.

The minimum viable LocalBusiness block

Use the most specific schema type that fits — Plumber, HVACBusiness, Dentist, RoofingContractor and dozens of others exist as subtypes of LocalBusiness. Specificity helps. Populate at minimum: name, image, address as a full PostalAddress, telephone, url, geo coordinates, openingHoursSpecification, areaServed, priceRange, and a hasOfferCatalog enumerating your services.

The sameAs array is the part people skip

Every profile you control should be listed in the sameAs array: Google Business Profile, Facebook, Yelp, Instagram, LinkedIn, BBB, industry association directories, Angi, Nextdoor, and any local chamber listing. This array is the connective tissue of your entity graph. It is the difference between a model seeing scattered records and seeing one corroborated business.

For the full implementation detail across every schema type that matters in AI retrieval, including the ones beyond LocalBusiness, our complete schema markup stack guide covers the whole build. If your entity is established enough to warrant it, the Wikipedia and Wikidata authority playbook covers the next tier of entity corroboration.

Layer 5: Local Press, Chambers, and Associations

Layers 1 through 4 are things you say about yourself. Layer 5 is the first layer where somebody else says it, and that shift matters enormously to how a model weighs the information.

What actually counts

• Local news mentions. A metro paper, a neighborhood blog, a regional business journal. These carry weight disproportionate to their traffic because they are topically and geographically aligned with the query.
• Chamber of commerce membership. Cheap, boring, and a genuinely strong corroboration signal because chamber directories are structured, local, and hard to fake.
• Trade and licensing associations. Industry body directories carry authority signals specific to your category. If your trade has a certifying body, be listed with it.
• Sponsorships with a web presence. The youth sports team sponsorship only helps if the league publishes a sponsor page that names you and links to you.
• Supplier and manufacturer locator pages. If you are a certified installer or authorized dealer for anything, get on the manufacturer's find-a-pro page. These are high-authority, high-specificity local records.

How to actually get local press

Not by pitching a story about your business. Local outlets do not care that you exist. They care about things happening in the community. Be the source rather than the subject — offer expert commentary on a seasonal issue, publish original local data, or attach yourself to a genuine community event. The mention that comes from being quoted as an expert is worth more than the one that comes from a press release, because it embeds you in a topical context.

Layer 6: Reddit and Forum Presence in Your Metro

Nearly every metro has an active subreddit, and those subreddits are full of exactly the query type you want to win: "can anyone recommend a good electrician?" Those threads get retrieved by AI systems constantly, because they are recent, specific, and read as authentic human recommendation rather than marketing.

This is also the layer where businesses most often destroy their own credibility. Handle it carefully.

What works

• Genuine participation by a named human. The owner or a senior tech, using a real identity, answering questions in the community over months without pitching.
• Answering technical questions with no ask attached. Somebody posts a photo of a leak and asks what it is. You tell them what it is and roughly what it costs to fix, and you do not pitch. Do that thirty times and you become the person the community references.
• Transparent disclosure. If you do mention your business, say plainly that it is your business. Communities forgive disclosed self-interest and punish concealed self-interest severely.
• Being recommended by others. The highest-value outcome, and it is a byproduct of doing good work and being visible, not something you can shortcut.

What backfires

Sockpuppet accounts recommending yourself. Buying mentions. Dropping a link into a thread with no history in the community. Beyond the ethical problem, these get detected and removed, and a removed comment contributes nothing. Worse, a public accusation of astroturfing becomes indexed content that describes your business negatively — and that text is retrievable too.

The mechanics of building forum presence for citation purposes are covered in more depth in our Reddit strategy for AI citations guide, which applies to local businesses with only minor adjustments.

Layer 7: Owned Content That Answers Real Questions

The top layer is the one you fully control, and it is where the retrieval pool gets fed directly. But local service content has to be built differently than the location-page template most agencies still sell.

Stop building thin city pages

The old playbook — one near-identical page per city with the town name swapped in — does not work for AI retrieval. There is nothing in those pages to retrieve. They contain no answer to any question. A model scanning them finds a service description and a place name and nothing worth citing.

Build answer pages instead

The content that gets retrieved answers the question a person actually typed. For a local service business, that means:

• Cost pages with real numbers. "How much does a water heater replacement cost in [metro]" with an actual range, the variables that move it, and what is included. Specific pricing is the single most retrievable local content type because it directly answers the most common local query and almost nobody publishes it honestly.
• Problem-diagnosis pages. "Why is my furnace making a clicking noise" — genuinely useful, genuinely retrievable, and it positions you as the expert before any transaction.
• Regulation and code pages. Permit requirements, local code specifics, seasonal regulations. Highly specific, rarely covered well, and unambiguously local.
• Timing and seasonality pages. When to schedule, what fails in which season, what your local climate does to equipment.
• Comparison pages. Repair versus replace, this system versus that one, DIY versus hiring out — with an honest answer including when the customer should not hire you.

Format for extraction

Lead every page with a direct answer in the first hundred words. Use question-shaped headings. Keep answers self-contained so a paragraph pulled out of context still makes sense. Include a specific number wherever you honestly can. These are the same principles that govern getting featured in Google AI Overviews, and they transfer cleanly to conversational AI.

The 10-Prompt Local Citation Diagnostic

You cannot improve what you have not measured, and there is no dashboard for this. The workaround is structured sampling: run a fixed set of prompts on a fixed schedule and log what comes back.

Run all ten of these in ChatGPT, Perplexity, Claude, and Google AI Mode. Use a fresh session each time with memory and personalization disabled, or the model will feed you back your own history and you will get a falsely optimistic reading.

What to log

For each prompt on each engine, record four things: were you named, what position in the list, was the description accurate, and what sources were cited. Prompt 10 is the most operationally useful of the set — it tells you which specific URLs the model pulled, which is the closest thing to a keyword report that exists in AI search.

Run the full grid quarterly. Ten prompts across four engines is forty data points and about ninety minutes of work. If you would rather automate the sampling, several AI visibility tracking tools now run this monitoring continuously, though manual sampling is entirely adequate to start.

Multi-Location vs Single-Location: What Changes

Everything above assumes one location. Multiple locations change the architecture in ways that are easy to get wrong.

The core principle: a location only earns a page if it has a real address and real staff. Service-area businesses without a physical presence in a town should build answer content about that area rather than a fake location page. Models are increasingly good at detecting location pages with no corroborating entity behind them, and the downside is a credibility hit across your whole domain, not just that page.

The 30-Day Local AI Citation Sprint

Seven layers is a lot to look at. Here is the order that actually gets it built, structured so each week produces something usable rather than leaving everything half-finished.

Week 1 — Baseline and cleanup

• Run the full 10-prompt diagnostic across four engines and log every result. This is your before picture and you only get one chance to capture it.
• Audit every existing listing for NAP consistency. Search your business name, your old phone number, and your old address. Correct every conflict you find.
• Verify AI crawlers are not blocked in robots.txt or by your firewall.

Week 2 — Entity foundation

• Rebuild the Google Business Profile completely: primary category, secondary categories, description rewritten as an entity definition, every service enumerated, every attribute set, and real photos replacing any stock imagery.
• Write and validate the LocalBusiness schema block including a complete sameAs array.
• Claim or correct listings on the three or four platforms that matter most in your category.

Week 3 — Corroboration

• Join the chamber and any relevant trade association, and confirm your directory entry is live and correct.
• Get on every manufacturer or supplier find-a-pro page you qualify for.
• Launch the review request process with specificity prompts, and respond to every existing review that has no response.
• Start participating in your metro subreddit as a named human. No pitching for at least the first month.

Week 4 — Content and re-measure

• Publish three answer pages: one cost page with real numbers, one problem-diagnosis page, one local regulation or seasonality page.
• Confirm each is indexed and reachable.
• Re-run the diagnostic. Do not expect much movement in thirty days — you are establishing the baseline trend, not the result.

What Does Not Work and Wastes Budget

An honest list, including a few things sold specifically to local businesses right now.

• Mass directory submission packages. Two hundred listings on dead directories nobody reads and no model retrieves. Fifteen accurate listings on platforms that matter beat two hundred on platforms that do not.
• "AI SEO" retainers with no measurement. If a vendor cannot show you a diagnostic baseline and a re-measure, they have no way to demonstrate they did anything. Ask what they will measure before you ask what they will do.
• Keyword-stuffed business names on Google. Against policy, gets reported by competitors, and creates NAP inconsistency between your legal name and your listing — actively harmful.
• Purchased reviews. Detectable, removable, and the resulting corpus lacks the specificity that makes reviews citable in the first place. You pay for something that fails at the exact job you wanted it to do.
• Templated city page farms. The single most common local SEO product, and it produces nothing retrievable. Thin pages with a town name swapped in contain no answer to any question.
• Chasing every new AI platform. The corroboration layer serves all of them. Build the entity properly and you are eligible everywhere, including engines that do not exist yet.

What the whole thing actually costs

Done in-house, the sprint above is roughly 25 to 35 hours of work spread across a month, plus chamber and association dues, which usually run a few hundred dollars a year. The schema block is a one-time build. The ongoing commitment is review generation, forum participation, and one or two answer pages a month. Done through an agency, expect the initial build to price like a small project and the ongoing work to price like a light content retainer — and be skeptical of anything quoted before someone has looked at your current diagnostic.

If you want to see how this maps to a broader visibility program, our AI visibility audit guide walks the full diagnostic process, and the local business AI search guide covers the strategic case for service businesses specifically.