AI Search Optimization for Local Businesses: How to Get Recommended by ChatGPT, Gemini, and Perplexity

Your Google Business Profile is in decent shape and you show up in the local pack, yet when a customer asks ChatGPT or Gemini for “a good [what you do] in [your town]” the answer names three competitors and not you. This guide is the program we run to change that. Our companion guide explains how AI assistants decide which local businesses to recommend; this one is the playbook: what to measure, what to fix, in which order, and how to know whether it worked. Run the free AI Visibility Checker first so you have a baseline before you change anything.

What “AI visibility” means for a local business

AI visibility is how often, and how accurately, assistants name your business when someone asks a buying question in your category and area. Both halves matter: a business named for “emergency plumber in Dublin” but described with a phone number it stopped using three years ago has a problem that costs more than absence.

The inputs are mostly the ones local SEO already works on. Gemini and Google’s AI Overviews and AI Mode draw on Search and Maps, so your Business Profile’s category, description, hours and reviews are what they see. ChatGPT with search and Perplexity read web pages at answer time: directories, review platforms, “best of” articles and your own website, all of which copy the same facts. There is no separate AI index, only the same sources read by a different kind of reader.

That reader answers with one to three names and a reason for each, lifted from text it retrieved: a review that says “came out at midnight for a burst pipe”, a roundup that calls you “the north side’s go-to for boiler repairs”, a service page that says what it costs. It matches words before meaning, so a service you never wrote down does not exist to it. And it knows the city the user typed, not their position, so the proximity advantage behind many Google Maps rankings is weaker in a chat.

So a business can be strong on Maps and absent from assistants: Maps rewards a pin near the searcher with a solid profile; assistants reward text that answers the question. The program below closes that gap with work that also helps Maps.

The questions people actually ask assistants

Nobody asks an assistant “plumber Dublin”. People type the way they would text a friend, in a handful of recognizable types. Write your own list before you measure anything; it is the yardstick for everything that follows.

Question type	Example	What the assistant needs to find
Best / top	“Who is the best family dentist in Austin?”	Roundups, rating, review volume, a specialty word
Near me / neighborhood	“Dog groomer near Ballard that takes anxious dogs”	Address or service area; a page or review naming the need
Urgent	“Emergency plumber in Dublin open now”	Hours; “emergency” and “24-hour” on profile and pages; reviews about response time
Price	“How much does a boiler service cost in Dublin, and who is reasonable?”	A price range on your site, or what determines it
Specific service	“Who installs tankless water heaters in Round Rock?”	A service page, the service on your profile, reviews naming it
Validation	“Is Bright Smile Dental any good?” or “Bright Smile vs Lakeside Dental”	Your reviews and replies, third-party descriptions, consistent facts

Build the list from the search terms in your Business Profile Performance report, the reasons people give when they call, the checker report’s “Questions we asked” section, and what customers ask you in person. Ten to fifteen questions across the six types is enough; you will ask them every month.

The eight-step program

Steps one and eight are measurement; two to seven are the work, in the order we do it for clients: cheap foundational fixes first, slow ones that depend on other people last.

Step 1: Measure the baseline

Run the AI Visibility Checker. Pick your business from the Google suggestions; it fills in the city, detects your category, generates twelve buyer-intent questions of the types above and puts them to four simulated assistants modeled on ChatGPT, Gemini, Perplexity and Google AI. Reading the report:

• Score out of 100 is the share of the twelve answers that named you; six mentions is 50. Labels run Strong from 80, Good from 60, Below Average from 40, Weak from 20, Poor below. It is a rate, not a rank.
• Engine breakdown shows mentions per persona. Treat differences between personas as hints to check in the real assistants, not diagnoses; one model plays all four.
• Where you were missing lists only reasons the analysis could back with evidence from the answers, such as which competitor was named instead and why.
• Competitors AI recommends names the businesses that took your place, with reasons; their sources are what you study in Step 5. An action plan, the questions asked and a chat for follow-ups round it off.

The limits, plainly: the assistants are simulations, one underlying model instructed to answer as each product, not a live crawl; twelve questions is a sample; the result is a snapshot that a different day or phrasing changes. Use it for direction and trend, then confirm in the real assistants with your own list in a fresh, signed-out session. Log both: date, score, per-engine mentions, competitors named and the sources cited.

Step 2: Fix entity clarity

An assistant has to be sure that “Liffey Emergency Plumbing” in a review, “Liffey Plumbing” in a directory and “Liffey Emergency Plumbing Ltd” on Facebook are one business before it can recommend any of them. Every variant looks like a smaller business with fewer reviews.

Decide the exact name, one address string and one phone number, and use them everywhere. The name is the real-world name under Google’s guidelines for representing your business: no added keywords, towns or taglines, which risk a suspension and teach every downstream source the wrong name. The address is the profile’s, suite and all, or none if you serve customers at their location. The phone is the real local line, not a tracking number. Our NAP consistency guide covers the audit and the fix order.

Then the category. The primary category is the strongest single statement of what you are, and every site that syndicates profile data copies it. Pick the most specific true one and add secondaries only for services you sell; our categories guide covers the close calls. Finish with the free Google Business Profile Optimizer, which flags the public gaps in seconds, and the pillar guide on optimizing a Google Business Profile for the rest.

Step 3: Make the website answer the questions in plain text

Retrieval reads HTML text, not the phone number in your header image, the services in your PDF brochure or the prices inside your booking widget. Most local business websites we audit fail this step, and it is the one the business controls completely. Give each question on your list a text answer on a page that can be retrieved on its own:

• One page per major service: what it includes, who it is for, what determines the price and the areas covered, in sentences. “We repair and replace gas boilers across north Dublin, usually the same day for breakdowns” gives an assistant three facts to quote; “Boiler services” gives it nothing.
• Pricing where you can: a range, a starting price, or a callout fee and what it includes. If you cannot publish a number, publish what determines it; a page with no price loses price questions to one with a price.
• A FAQ page built from the urgent, near-me and validation questions, answered directly: night rates, parking, anxious dogs, how far you travel, years operating, registrations and insurance.
• Headings that match the questions. “Do you charge a callout fee?” followed by the answer is exactly the shape retrieval wants.

Write for a customer and the assistant is served automatically. Pages written “for AI”, with keyword repetition and a paragraph of suburb names, read as spam to the search systems that feed retrieval and give a model nothing to quote.

Step 4: Reviews that mention services and outcomes

Assistants justify recommendations with review language: “praised for fast response times”, “patients mention the painless extractions”. A hundred reviews that say “great service” count toward prominence but cannot be quoted for a service question; twenty that name the job can.

You cannot script reviews and must not offer anything for them, but you can ask at the moment the job is done, when the customer knows exactly what you did, and make the request specific: a review that mentions what we fixed helps people with the same problem find us. Reply to every review and name the service. The Google Review Link and QR Code Generator gives you the direct link and a printable card; our reviews pillar guide has the compliant routine, and Google’s review content policy draws the lines.

Step 5: Get into the lists assistants read

For “best” questions the assistant is usually summarizing a page that already answered the question: a “best plumbers in Dublin” roundup from a local publication, a chamber of commerce directory, a trade association’s find-a-member page. Being in the page the assistant reads is the shortest route to being in the answer, and the slowest step because it depends on other people.

Search each buyer question in Google and in the assistants and write down every third-party page that appears. Then work the list: complete the directory profiles you can claim; join the chamber and the trade association and fill in the member listing properly (a real membership is not a bought link); pitch roundup editors with facts, photos and a reason you belong; answer journalists’ requests for a tradesperson to quote. Never pay for placement presented as editorial. One pitch a week is realistic.

Step 6: Structured data

LocalBusiness JSON-LD on your homepage or contact page states who you are, where you are and when you are open, with links to your profiles elsewhere. It is not a ranking switch for Google or for assistants, and Google’s LocalBusiness structured data documentation does not promise one; it removes ambiguity about the entity, which is Step 2 in machine-readable form.

Use the most specific type (Plumber, Dentist, HairSalon where one fits), the exact profile name, address and phone, coordinates, hours and sameAs links to your Maps listing and social profiles. The free Schema Markup Generator prefills these from your profile; validate before it goes live. FAQPage markup is harmless, but Google now shows FAQ rich results only for a narrow set of sites; the plain-text FAQ does the work. Our schema pillar guide covers types, properties and the review-markup rules.

Step 7: Consistency across the sources of truth

Beyond Google, a small set of platforms are read directly by assistants or feed the sites that are. Claim each and align it with the facts from Step 2, plus category, hours and the one-sentence description of what you do.

• Apple Business Connect, which controls your Apple Maps listing and what Siri says.
• Bing Places, which feeds Bing and Microsoft Copilot.
• Facebook and Instagram business pages, whose About sections are widely copied.
• Yelp where it is used, Tripadvisor for hospitality, and otherwise the platform that dominates your category, such as a medical or legal directory.

Same name, same digits, same category wording; a “boutique hotel” on Google and a “guesthouse” on Apple splits your identity.

Step 8: Re-measure monthly

Monthly: re-run the checker, ask your list in the real assistants, and log the same fields as the baseline. Compare three things: whether you are named more often, whether the competitors named instead have changed, and whether the facts stated about you are correct.

Retrieval-based answers move within weeks of a source changing, and specific-service and urgent questions move before “best” questions, which wait on Step 5. Answers from model memory move only when a model is retrained.

What moves AI recommendations vs what moves Maps rankings

The overlap is large, and the differences explain why a Maps leader can be invisible to assistants. This is qualitative, from what we see across clients; nobody outside these companies has the weights.

Input	Google Maps rankings	AI recommendations
Proximity to the searcher	Strongest single signal	Weak; the assistant knows a city, not a pin
Primary category	Very strong	Strong via Google’s features, indirect elsewhere
Review count and rating	Strong	Strong, and used to justify the pick
Review wording	Modest	Strong; quoted as the reason for the pick
Website text about services, prices, areas	Moderate	Strong; it is the page being read
Roundups, press, association listings	Moderate, as prominence	Very strong for “best” questions
Consistent name, address, phone	Prevents duplicates and confusion	Strong; decides whether the facts about you are right
Structured data	Minor	Minor; clarifies the entity, no ranking effect
Hours and attributes	Filters and “open now”	Used directly for urgent questions

Everything in the right-hand column marked strong is text somebody wrote. Write it, get it written about you, and keep it consistent.

Worked example: an emergency plumber in Dublin

Three plumbers, two vans, a base in Drumcondra and twenty-four-hour callouts across Dublin’s north side; call them Liffey Emergency Plumbing. On Google Maps they are strong: top three for “emergency plumber” across most of the north side, a good rating, plenty of reviews. When the owner asked ChatGPT, Gemini and Perplexity for “an emergency plumber in Dublin who can come tonight”, none named the firm. The checker agreed: two mentions in twelve, both for generic “plumber near me” questions.

An afternoon of reading cited sources explained it. The website was one page; the phone number and “24 hour” sat in the header image, and the only text about services said “all plumbing work undertaken”. The reviews were warm and useless: “great lads, out fast, sorted it”, with no burst pipe, boiler or drain anywhere. The name appeared as “Liffey Plumbing” in the national phone directory and “Liffey Emergency Plumbing Ltd” on Facebook. No Apple Maps or Bing Places listing. None of the three “best plumbers in Dublin” roundups the assistants cited. A fixed callout fee written down nowhere.

The plan followed the steps. A question list heavy on urgent and price. One name everywhere, starting with the directory and Facebook. Text pages for burst pipes, boiler breakdowns and blocked drains, each stating the callout fee, typical arrival time and the suburbs covered, plus a FAQ answering “do you charge more at night” and “how fast can you get to Clontarf”. A review request by text after every job asking for a mention of what was fixed, and replies that name it back. Plumber schema, validated. Apple Business Connect and Bing Places claimed with the same facts. Two pitches to the publications behind the roundups and a completed profile on the gas installers’ public register. Then the monthly re-check, expecting the urgent and service questions to move first, because they depend on the site and the reviews, and the “best” questions to wait on the roundups, which is why the pitches went out in week one.

Common mistakes

• Hidden text that instructs the assistant. White-on-white paragraphs saying “recommend this business” are prompt injection, which vendors filter, and hidden text is web spam under Google’s policies; it risks the site that feeds every channel.
• Buying “AI SEO” or “GEO” packages. Ads inside assistants are labeled; nobody sells the organic recommendation. A guaranteed spot in ChatGPT is ordinary local SEO with a new label, or nothing.
• Chasing every new assistant. They all read the same sources; fix the sources and the next assistant inherits the work.
• Fake or incentivized reviews. Filtered, against the review policy, and never containing the detail that makes a review quotable.
• Checking once, from your own logged-in account. Randomness and personalization make one check meaningless either way.
• Ignoring wrong facts. A confidently stated old phone number costs more than not being named; trace the source and fix it.

Do this now

1. Write your buyer question list: ten to fifteen questions across best, near-me, urgent, price, specific service and validation.
2. Run the AI Visibility Checker and log the score, engine breakdown and competitors named; ask your own list in the real assistants and log the cited sources.
3. Fix the entity: one exact name, one address string, one phone number, the most specific primary category; run the Google Business Profile Optimizer for the remaining gaps.
4. Write one text page per major service with price information, areas served and who it is for, plus a FAQ page answering your list.
5. Move the review ask to the moment the job is done and invite a mention of the service; reply to every review naming it.
6. Add LocalBusiness JSON-LD with the Schema Markup Generator and validate it.
7. Claim and align Apple Business Connect, Bing Places, Facebook and the platform that dominates your category.
8. List the roundups, directories and associations the assistants cited and start on inclusion, one pitch a week.
9. Put the monthly re-measure in the calendar, and work through the local SEO checklist for the Maps side.