Blog about GEO and SEO

Where does ChatGPT get information about local businesses?

ChatGPT has no private database of local businesses. For local questions it mostly runs a live web search, reads a handful of pages it can fetch – your website, directories, review sites, news and maps listings – and summarises them, sometimes mixing in older facts absorbed during training. That means your visibility depends on being present, consistent and machine-readable across the open web, not on anything you submit to OpenAI.

Which sources does ChatGPT actually use for local queries?

When you ask something like "best dentist near Kreuzberg" or "plumber open on Sunday in Austin", ChatGPT typically triggers its search mode. It issues one or more queries, retrieves a short list of results, opens some of them, and writes an answer with links.

The pool of material it draws on breaks down into four layers:

LayerWhat it containsHow current it is
Live web searchPages returned by ChatGPT's search index and third-party search providersMinutes to days
Pages fetched on demandThe specific URLs ChatGPT opens to read details (hours, prices, services)Real time
Training dataText seen when the model was trained, including old copies of sites and directoriesMonths to years old
Session contextYour location, your previous messages, files or links you pasteImmediate

OpenAI has not published a full list of its search partners or index composition, and the mix has changed over time, so treat any precise claim about "ChatGPT uses X search engine" with caution. What is stable and well documented is the crawler side: OpenAI operates distinct user agents for different jobs.

User agentPurpose
GPTBotCrawls pages that may be used for model training
OAI-SearchBotBuilds the index used by ChatGPT's search feature
ChatGPT-UserFetches a page when a user or the assistant follows a link in a conversation

Blocking GPTBot does not block search. Blocking OAI-SearchBot does remove you from ChatGPT's search surface, which is usually the opposite of what a local business wants.

How does ChatGPT know where I am?

Location in ChatGPT usually comes from one of three places: an approximate location derived from the user's connection, a location the user typed into the prompt, or a location remembered from earlier in the conversation or from memory settings. Behaviour differs between the web app, the mobile apps and the API, and OpenAI can change the defaults.

The practical consequence: many local answers are generated from a city-level or neighbourhood-level hint, not a precise GPS point. ChatGPT is not computing driving distance the way a maps app does – it's reading text that says "located at 12 Oak Street, Shoreditch" and matching it to the area in the prompt.

So the address, neighbourhood name, nearby landmarks and service area need to appear as plain text on your pages. If your location only exists inside an embedded map widget or an image of a business card, it is effectively invisible.

Why does ChatGPT sometimes name businesses that closed?

Because the model's training data is older than the web. If search doesn't fire – short prompts, offline-ish questions, or a model variant without browsing – ChatGPT answers from memory, and that memory can include a restaurant that shut two years ago or a phone number you changed.

Outdated mentions elsewhere on the web make this worse. An old directory page with your previous address keeps feeding the wrong fact to both search and training crawls. Cleaning up stale listings is unglamorous local SEO work that pays off directly in AI answers.

Another failure mode: when sources disagree, the model may average them or pick one arbitrarily. Two different phone numbers across your site, Google Business Profile and a directory create genuine ambiguity. Consistency is not a nicety here – it's what lets a model state a fact confidently.

Which directories and review sites matter most?

There is no public ranking of which sources ChatGPT trusts for local data, and it varies by country and industry. What's observable is that AI answers about local businesses tend to cite a repeatable set of page types:

  • Your own website, especially a dedicated location or contact page.
  • Maps and profile platforms (Google Business Profile, Apple Business Connect, Bing Places).
  • Review platforms relevant to your vertical (general review sites, restaurant and hospitality platforms, trade-specific directories).
  • Local press, city guides, "best X in Y" listicles and association member directories.
  • Industry registries, chambers of commerce and licensing bodies.

The "best X in city" listicles deserve special attention. AI answers for comparison-style prompts frequently lean on them, because they already contain a curated list in the exact shape the answer needs. Getting included in credible local roundups is one of the highest-leverage activities for AI visibility [check: share of local AI answers citing listicle-type pages].

How do I make my own site a usable source?

Put the facts in text, in one obvious place

Create a page per physical location with: full postal address, phone, email, opening hours including holidays, service area, payment and accessibility details, parking, nearest transit stop. Write them as sentences and lists, not graphics. Add a short "About this location" paragraph that a model can quote verbatim.

Add LocalBusiness structured data

Schema.org markup doesn't guarantee a citation, but it removes ambiguity about which string is the phone number and which is the address. It also feeds classic search features that AI engines read.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Dentist",
  "name": "Northside Dental Studio",
  "url": "https://example.com/locations/northside",
  "image": "https://example.com/img/northside.jpg",
  "telephone": "+1-512-555-0142",
  "priceRange": "$",
  "address": {
    "@type": "PostalAddress",
    "streetAddress": "412 Oak Street, Suite 3",
    "addressLocality": "Austin",
    "addressRegion": "TX",
    "postalCode": "78702",
    "addressCountry": "US"
  },
  "geo": {
    "@type": "GeoCoordinates",
    "latitude": 30.2672,
    "longitude": -97.7431
  },
  "areaServed": ["Austin", "East Austin", "Mueller"],
  "openingHoursSpecification": [
    {
      "@type": "OpeningHoursSpecification",
      "dayOfWeek": ["Monday","Tuesday","Wednesday","Thursday"],
      "opens": "08:00",
      "closes": "18:00"
    },
    {
      "@type": "OpeningHoursSpecification",
      "dayOfWeek": "Friday",
      "opens": "08:00",
      "closes": "14:00"
    }
  ],
  "sameAs": [
    "https://www.google.com/maps/place/...",
    "https://www.yelp.com/biz/...",
    "https://www.facebook.com/..."
  ]
}
</script>

The sameAs array matters more than people expect: it explicitly ties your site to your profiles, which helps any system trying to decide whether three listings describe one business or three.

Keep NAP identical everywhere

Name, Address, Phone should be byte-for-byte the same on your site, your Google Business Profile and every directory. "Suite 3" vs "#3", "St" vs "Street", a local number on one page and a toll-free number on another – each variation is a chance for a model to hedge or pick wrong.

Do I need to let OpenAI's crawlers in?

Yes, if you want to appear in ChatGPT answers with a citation. Check your robots.txt for blanket blocks added by a developer or a CDN's "block AI bots" toggle.

A permissive configuration for a local business looks like this:

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: GPTBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Google-Extended
Allow: /

User-agent: *
Disallow: /cart/
Disallow: /checkout/
Sitemap: https://example.com/sitemap.xml

If you have a policy reason to refuse training use, block GPTBot only and keep OAI-SearchBot and ChatGPT-User allowed. Also verify at the server level: firewalls, bot-protection rules and rate limits block AI crawlers far more often than robots.txt does. A 403 to OAI-SearchBot is invisible in your analytics but fatal to AI visibility.

Does llms.txt help for a local business?

llms.txt is a proposed convention – a Markdown file at your root that points AI systems at your most important pages. No major engine has confirmed it as a ranking or retrieval input, so treat it as cheap insurance rather than a fix.

# Northside Dental Studio

> Family dental practice in East Austin, Texas. Open since 2011.
> Address: 412 Oak Street, Suite 3, Austin, TX 78702. Phone: +1-512-555-0142.

## Key pages
- [Location and hours](https://example.com/locations/northside): address, hours, parking, transit.
- [Services and prices](https://example.com/services): treatments with indicative prices.
- [Insurance accepted](https://example.com/insurance): list of accepted plans.
- [FAQ](https://example.com/faq): appointments, emergencies, cancellations.

## Contact
Email: hello@example.com. Emergency line: +1-512-555-0199.

The real benefit is discipline: writing it forces you to identify the five pages that answer the questions people actually ask.

How do I check whether ChatGPT mentions my business?

  1. Build a list of 15–30 prompts a customer would type: "best [service] in [neighbourhood]", "[service] open late near [landmark]", "how much does [service] cost in [city]", "is [your brand] any good".
  2. Run them in ChatGPT with search enabled, in a fresh chat, with the relevant location set.
  3. Record whether you appear, which competitors appear, and which URLs are cited.
  4. Open the cited URLs. Those are the pages that currently control the answer – your work is either to get onto them or to outrank them.
  5. Repeat monthly. Answers shift as the index refreshes, so a single test tells you little.

Also check server logs for OAI-SearchBot, ChatGPT-User, PerplexityBot and ClaudeBot hits. If they never appear, something is blocking them.

What should I fix first?

PriorityActionEffort
1Unblock AI crawlers in robots.txt and at the firewallLow
2Fix NAP inconsistencies across site, maps profile and top directoriesMedium
3Create or upgrade a location page with hours, services, prices in textMedium
4Add LocalBusiness JSON-LD with sameAs linksLow
5Remove or correct stale listings showing old addresses or numbersMedium
6Earn mentions in local roundups, press and association directoriesHigh
7Build a steady flow of recent reviews on the platforms in your verticalOngoing

FAQ

Can I submit my business directly to ChatGPT?

No. There is no submission form or business dashboard. The only way in is to be visible on the open web: a crawlable site, accurate profiles and mentions on pages that search engines already index.

Does blocking GPTBot remove me from ChatGPT answers?

Not by itself. GPTBot relates to training data; OAI-SearchBot relates to the search index used in live answers. Blocking only GPTBot keeps you eligible for cited search answers, though some older facts about you may fade from model memory over time.

How long before changes show up in ChatGPT?

Site changes have to be recrawled and reindexed first, and third-party listings update on their own schedules. Expect weeks rather than days, and longer for anything that depends on new mentions across other sites [check: typical refresh window observed for OAI-SearchBot].

Do reviews affect whether ChatGPT recommends me?

Indirectly. Models read review text and ratings from pages they retrieve, so a strong, recent review presence gives the model positive language to quote. Review volume is not a ranking signal in any documented sense, but it changes what the available text says about you.

Why does ChatGPT give a different answer to the same question twice?

Because retrieval and generation both involve variation: different searches fire, different pages are opened, and the model samples wording. Treat AI visibility as a distribution across many runs, not a fixed position.

If you want a quick read on whether AI assistants can actually find and cite your location pages, run a free check in Zgadka – it scores your site on GEO and SEO and shows which crawler, markup or content gaps are standing between you and the answer.


Fazier badge