Here is the short answer: ChatGPT recommends the business it can read, verify, and corroborate across the open web. It does not rank you the way Google does. When someone asks it for the best salon, plumber, or gym in your city, it assembles an answer from your website, the places that mention you, and your reviews, then names the business it is most confident about. So the job is not to chase a ranking. It is to become the most legible and most corroborated option in your category and your city.
That is a different discipline from the local SEO you already know, and most of the checklists floating around get parts of it wrong. Below is what actually moves the needle in 2026, in the order I work through it with a client, ending with the one thing that quietly breaks all of it.
How ChatGPT finds a local business, and why it is not Google
Start with a fact that reorganizes everything: ChatGPT does not run on Google. Its search is powered mostly by Bing's index, plus its own crawler and Google's review data. If you have spent years tending your Google Business Profile and never once thought about Bing, you may be invisible to the single most-used AI assistant in the country. Claiming and completing Bing Places for Business is the highest-leverage hour most local owners have never spent.
The other assistants pull differently. Google's Gemini and its AI Overviews read the Google stack you already know: Maps, your Business Profile, the entity graph. Perplexity names whatever it can cite and verify, so it leans on sources with clear authorship. One business, three different front doors, and you have to be legible to all of them.
| Assistant | What it mostly reads | Your first move |
|---|---|---|
| ChatGPT | Bing's index, your own site, editorial mentions, Google reviews | Claim Bing Places for Business |
| Gemini and Google AI Overviews | Google Business Profile, Maps, the entity graph | Perfect your Business Profile |
| Perplexity | Sources with clear authorship it can cite | Earn editorial mentions |
Now the part that surprises people. When BrightLocal reverse-engineered 800 ChatGPT local searches across twenty industries and twenty cities, the sources broke down roughly like this: about 58 percent were the businesses' own websites, about 27 percent were editorial mentions such as articles and blog posts, and only about 15 percent were directories. Wikipedia alone accounted for close to 39 percent of those editorial mentions. The directories that did show up were Three Best Rated, Expertise, and Tripadvisor, not Yelp, not Facebook, not Google Maps. The exact mix shifts as the engines change, but the hierarchy has held: your own site and being named in someone else's "best of" list matter more than the directories most owners pour their time into.
Google ranks pages. ChatGPT assembles an answer. You are not optimizing for a position. You are optimizing to be the option it trusts enough to name.
Claim the profiles the models actually read
The models build their picture of you from structured profiles first, because structured data is easy to trust. Three of them matter, in this order:
- Google Business Profile, done to the AI standard. Set your primary category to your highest-revenue service and add every relevant secondary category. Write full service descriptions in plain language. Fill the service area, hours, photos, and the Q&A. The models read the categories and the descriptions, not your intentions.
- Bing Places for Business, because of everything above. Most of your competitors have not claimed it. That is the opening.
- Apple Business Connect, which feeds Siri and Apple Maps. Quick to set up, rarely done.
Then the rule that ties them together: your name, address, phone, and business description have to match across all three, plus your website and any directory that lists you. A machine reading four slightly different versions of your address does not average them. It loses confidence and reaches for a competitor it is more sure about.
- Primary category set to your money service, secondaries filled
- Service descriptions written in full sentences, not keywords
- Bing Places and Apple Business Connect claimed, not just Google
- Name, address, and phone identical everywhere, down to the "St" versus "Street"
- Photos and hours current, Q&A seeded with real questions
Reviews are training data, not a scoreboard
You have been taught to chase the star average. The models care more about what the reviews say. A detailed review that names the service, the outcome, and the neighborhood teaches the model something it can repeat. "They re-piped a rowhome in Manayunk in two days and the quote held" is worth more than ten reviews that say "great service." One is language the model can extract and attribute to you. The other is noise.
So the ask you make of happy customers changes. Instead of "leave us five stars," it becomes "mention what we did and where." Keep them recent, because recency signals a business that is still operating, and respond to them, because a reply is more text that confirms you are real and attentive.
Make your website legible to a machine
Here is where a good-looking site quietly fails. Many modern templates render their content with client-side JavaScript, which means the words a human sees are not in the raw HTML a crawler first receives. If your core answer, your services, and your location load only after a script runs, the model may never read them. Server-rendered HTML is not a nice-to-have for AI visibility. It is the floor.
On top of that floor, two things do the work:
- Structured data. Add LocalBusiness or ProfessionalService schema, Service schema for what you offer, Organization schema with sameAs links to your profiles, and FAQPage schema on any page with a real question-and-answer section. This is how you hand the model a clean, labeled version of the facts instead of making it guess.
- Extractable writing. Open each page with a direct answer. Use question-shaped headings. Use numbered steps for processes and a comparison table for tradeoffs. Put your FAQ in visible HTML, not hidden behind a click. Write the way you want to be quoted, because being quoted is the entire game.
The move most guides get wrong: llms.txt
Almost every AI-visibility checklist published this year tells you to add an llms.txt file to your site, a small text file that supposedly tells the models how to read you. Skip it. Ahrefs studied 137,210 domains in May 2026: 28 percent had published the file, and 97 percent of those files received zero AI traffic. The bots do not even look for it, they only fetch it when pointed at it, which for a local business is never. It is an hour spent on a rumor. Put that hour into your schema and your business-information consistency instead, where the models are actually reading.
Get named where ChatGPT is already reading
Since roughly a quarter of what ChatGPT cites for local is editorial, the highest-value outside work is getting named in other people's lists. The "best [your service] in [your city]" articles, the local press roundups, the industry association and chamber pages, the curated directories like Three Best Rated and Expertise that the study actually surfaced. Every independent site that describes you the same way raises the model's confidence that you are the real answer. Corroboration is the currency. One mention is a claim. Five consistent mentions across sites the model trusts is a fact.
How to tell if any of this is working
Three ways, from crude to precise:
- Ask the assistants directly, from a logged-out or incognito session, using the exact phrasing your customers would. Vary it a few times, because answers change with the user, the location, and the phrasing. Note who gets named, including the competitors, because that tells you who to study.
- Watch your analytics. In GA4, under Reports, Acquisition, Traffic acquisition, you will start to see chatgpt.com and perplexity.ai show up as referral sources once the assistants send you clicks. Small numbers, high intent.
- For an ongoing read, use an AI-visibility monitor such as Profound, Peec, or LLMrefs, which track how often you are named across the assistants over time. This is the layer that turns a guess into a measurement.
Where this usually breaks, and what I do about it
Notice that almost none of the above is a single clever tactic. A local business is rarely invisible to AI because it missed one trick. It is invisible because of fragmentation: an address that disagrees across a dozen tools, a website the model cannot read, reviews scattered across platforms, no schema, a Bing profile nobody ever claimed. The models meet a business that cannot describe itself consistently, and they move on.
That same fragmentation is why your software bill is too high. Every disconnected tool holds its own slightly different version of the truth, and you are paying for all of them to disagree. The two problems are the same problem.
That is the work I do. I run one audit that finds where you are invisible to AI and where you are overpaying for tools that overlap, then I install a single system that fixes both, so the savings help fund the visibility work. Fire a standard agency and you lose the tactics the day you stop paying. Work with me and the infrastructure stays installed. That is by design.
Frequently asked questions
Does ChatGPT use Google to find local businesses?
Mostly no. ChatGPT's search leans on Bing's index plus Google's review data, not Google's local rankings. Claiming Bing Places for Business is one of the most overlooked moves a local owner can make.
How long before ChatGPT starts recommending my business?
Plan in weeks, not days. The assistants have to crawl and index your updated profiles and site, then see you corroborated across enough independent sources to be confident. Recent reviews and fresh content speed it up.
Do I need an llms.txt file?
No. The current data shows the major assistants do not read it. Put the time into schema, server-rendered content, and consistent business information instead.
Is local SEO dead?
No. Your Google Business Profile, your reviews, and consistent business information are still the foundation. AI search raises the bar on machine-readability and on being described the same way across the web, but it did not replace the fundamentals.
Can I do this myself?
Yes. The checklist here is real and most of it is a matter of hours and discipline. The audit is for when you would rather have it done, done consistently, and consolidated into one system you own.
