How Local Businesses Get Cited by ChatGPT, Gemini and AI Overviews

Each assistant has a search layer behind it, and that layer is what decides whether you exist. Google’s AI Overviews and AI Mode draw from the regular Google index. Google’s documentation says a page must be indexed and eligible to be shown with a snippet, and that there are no additional technical requirements and no special optimizations necessary.

ChatGPT uses its own crawler, OAI-SearchBot, to surface websites in ChatGPT search, and a separate fetcher, ChatGPT-User, that visits a page when a user asks a question. Perplexity runs PerplexityBot for its index and Perplexity-User for live lookups. Both companies publish those names so you can check your server logs and your robots.txt for them.

For local queries there is a second source: licensed data. In July 2026 OpenAI began showing Yelp reviews, ratings, photos and business information inside ChatGPT answers, with Yelp branding and links, and by August Yelp reservations and waitlists could be booked without leaving the chat. Whitespark’s September 2025 study of 153 local queries across 17 categories and nine US cities found Facebook, Yelp and TripAdvisor were the review sources ChatGPT leaned on most, and that Los Angeles was the one city where Yelp beat Facebook.

Google’s AI features use what Google calls query fan-out: the system issues multiple related searches across subtopics and cites pages that do well across that wider set. Ahrefs measured the effect in March 2026 across 863,000 keywords and 4 million cited URLs. Only 37.9 percent of cited pages ranked in the top 10 for the original query, down from about 76 percent in July 2025. The rest were split almost evenly between positions 11 to 100 and pages outside the top 100.

Google: you do not need to create new machine readable files, AI text files, or markup to appear in AI Overviews or AI Mode.

A model quoting a page is looking for a sentence that answers the question on its own. Pages that bury the answer under a paragraph about how much you care do not get quoted. Pages that state the fact in the first line do.

The fan-out finding matters here. When someone asks an assistant who to call for a slab leak in Fullerton, the system may run separate searches for slab leak symptoms, repipe cost, Fullerton plumbers and PEX versus copper. A page that answers one of those sub-questions cleanly can be cited even if it would never rank for the original phrase.

1. Use a question as the heading, in the words a customer uses. "How much does a crown cost in Pasadena?" beats "Restorative Services".
2. Answer it in the first two sentences, with the number or the fact. Then explain.
3. Keep one question per section. A model lifts a section, not a page.
4. Name the place. "Irvine" and "Woodbridge" in the answer, not "our area".
5. Put your prices where you can. A page that says "crowns from" and a number is quotable. "Call for pricing" is not.
6. Date it. State when the price or the fact was last checked, so the model and the reader know it is current.

Write the way you talk to a customer on the phone. Short sentences read well on a phone screen and extract cleanly for a model. Every page on a Sitesmith build is written this way, because it is also how people read.

FAQ sections help. FAQ rich results mostly do not. Since August 2023 Google shows FAQ rich results only for well-known authoritative government and health sites, so the markup no longer earns the expanded listing for a plumber or a dentist. The markup is still valid and does no harm, but the value was never the markup. It is the question-and-answer text itself, which is exactly the shape an answer engine wants.

LocalBusiness schema is a different case. It states the business name, address, phone, hours, coordinates, price range and, for a restaurant, the cuisine and menu URL, in a form a crawler reads without guessing. Google says it uses that data for knowledge panels and carousels and asks for as many properties as you can provide. Use the specific type: Dentist, Plumber, Restaurant.

llms.txt is a proposal, published in September 2024 by Jeremy Howard, for a markdown file at the site root that summarizes the site for language models. It is easy to add and some documentation tools generate it automatically. It is also, so far, unread. Google’s John Mueller wrote on June 17, 2025 that no AI system currently uses llms.txt, and Google’s own AI features documentation says no AI text files are needed. Add it if you like. Do not expect anything from it.

Crawler access is the decision that does matter. OpenAI documents three agents. GPTBot crawls for model training; blocking it keeps your content out of training. OAI-SearchBot surfaces websites in ChatGPT search; blocking it removes you from ChatGPT answers. ChatGPT-User fetches a page when a user asks about it and does not follow robots.txt because a person initiated it. Google-Extended is the equivalent token for Gemini training, and Google says blocking it does not affect Search inclusion or ranking.

For a local business the choice is easy. Allow the search bots. If a customer asks ChatGPT for an electrician in Anaheim Hills and your site is blocked, the answer is your competitor.

• Check robots.txt does not disallow OAI-SearchBot, PerplexityBot or Googlebot.
• Decide separately about GPTBot and Google-Extended. Blocking those affects training only.
• Confirm the site is in Google Search Console and Bing Webmaster Tools and the sitemap is submitted. An unindexed page cannot be cited by anything.

An assistant assembling an answer about your business reads several sources at once: your site, your Google profile, Yelp, Facebook, a trade directory. Where they agree, it states the fact. Where they disagree, it hedges or picks the wrong one. A phone number that differs between Yelp and your site is not a small error to a model. It is two businesses.

The list of facts is short. Name, address or service area, phone, hours, services, prices you publish, license number, and the cities you serve. Write them once and copy them everywhere. When you change hours for the Rose Parade week or close for a Long Beach Grand Prix weekend, change them in every place that day.

Whitespark’s research and the Yelp integration point to the same conclusion for Los Angeles specifically: keep the Yelp listing complete, photographed and current, because in LA it is the review source ChatGPT reaches for first.

Reviews are the evidence a model uses to decide you are worth recommending, and now they are part of the answer itself. ChatGPT shows Yelp star ratings and review text inline. Google’s AI features draw on the same review data that feeds the map pack. BrightLocal’s 2026 survey of 1,002 US consumers found 97 percent read reviews for local businesses and 47 percent will not use one with fewer than 20 reviews.

Citations, the old local SEO word for directory listings, are back in a new role. They are the corroborating sources the model reads to confirm you exist and do what your site says. Google Business Profile, Yelp, Bing Places, Apple Maps, Facebook and your trade association listing are the minimum. Each one should carry the same facts and, where it allows, real photos.

• Ask every customer for a review by text within an hour of finishing. Google and Yelp both allow asking; neither allows paying.
• Reply to every review. Replies are text a model can read, and they show a business that answers.
• Claim and complete the six core listings. Match name, phone and hours to the site exactly.
• Pull live Google reviews onto the site so the page that gets cited also carries the proof.

Start with indexing, because nothing else works without it. A site that has never been verified in Search Console may have pages Google has never seen, and an unseen page cannot be cited by Google, ChatGPT or anyone else.

The Sitesmith Indexing service is a one-time $150 job with a checkable result: Google Search Console and Bing Webmaster Tools verified on your domain, the XML sitemap generated and submitted to both, robots.txt and canonical tags checked page by page, LocalBusiness and service schema validated, every page requested for indexing and confirmed, and a one-page report with the console screenshots. The accounts are created on your Google and Microsoft logins, so they stay yours. It takes about a week, most of which is waiting on Google to crawl.

After that, the Search plan at $99 a month covers the rest of this guide on a schedule: question research for your services and cities, pages written to answer them, the Google profile kept current, listings built and checked, and a monthly report of rankings, clicks and the AI answers you appear in. Indexing shows in days, local rankings in weeks, competitive terms in months.

Order of operations: indexed first, direct answers second, consistent facts third, reviews always.