How ChatGPT Decides Which Local Business to Recommend

Ask ChatGPT "who's the best roofer in Charlotte?" and you get three or four company names with a sentence of justification each.

That shortlist is not random, and it is not a ranking you can buy into. Here is the actual mechanism.

Two sources, one answer

ChatGPT answers from a blend of two things.

Training data is what the model absorbed when it was built — a very large snapshot of the public web, frozen at a point in the past. It contains a great deal about large brands and almost nothing reliable about a nine-truck plumbing company.

Live search is what happens when the question needs current or specific information. The model issues search queries, reads the pages that come back, and writes an answer citing them.

For local service questions, live search fires nearly every time. The model knows it does not have trustworthy training knowledge about individual contractors, and it knows business details change. You can usually confirm this yourself — the citations appear under the answer.

The practical consequence is important and most owners get it backwards: your website is not the primary input. It is one of several pages the model might read. The pages that reliably come back for "best roofer in Charlotte" are directories, review sites, map listings, and local roundups. Those are where the shortlist is actually assembled.

The six inputs that decide the shortlist

1. Whether you appear in the searches it runs

ChatGPT does not search for your business name. It searches the way the user asked — "best roofer in Charlotte," "roof replacement Charlotte reviews," "top rated roofing companies Charlotte NC."

Whatever ranks for those queries is what the model reads. Which means classic local SEO — map pack presence, ranking on the roundup pages, appearing on Yelp and Angi and BBB for your category and city — determines what enters the model's field of view in the first place.

If you are not in the source material, no amount of on-site optimization puts you in the answer.

2. Whether its crawler can read your site

If your site does make it into the results, ChatGPT still has to fetch and parse it. Two things break this:

Blocked crawlers. OpenAI uses OAI-SearchBot for search results and GPTBot for training. A robots.txt that blocks either — sometimes added years ago by a well-meaning developer, sometimes bundled into a security plugin — removes you from the pool.

JavaScript-only rendering. If your service list, address, and phone number only appear after client-side JavaScript executes, a crawler fetching raw HTML sees an empty shell. Single-page-app site builders do this frequently.

3. Whether the facts about you are consistent

The model is assembling a picture of an entity from several sources. When your Google listing, Yelp page, Facebook page, and website agree on your name, address, phone, and services, the picture is confident.

When they disagree — an old suite number, a tracking phone number on one directory, "Carter Plumbing" versus "Carter Plumbing & Drain LLC" — the model has a fuzzier picture, and fuzzier entities get named less often. Not because there is a penalty, but because the model reaches for the business it can describe cleanly.

4. Whether there is a reason to name you

Read a real ChatGPT local answer closely and you will notice every recommendation comes with a justification: "known for 24/7 emergency service," "consistently strong reviews for metal roofing," "family-owned, operating since 1994."

Those justifications are lifted from text. Review summaries, an About page, a directory description, a news mention. If nothing in the source material says what you are distinctly good at, the model cannot write a justification, and a business it cannot justify tends not to make the list.

This is the most actionable insight in this article. Give the model a sentence to steal. Somewhere public, in plain language: what you specialize in, how fast you respond, how long you have operated, what certifications you hold.

5. Reviews — volume, recency, and what they say

Review count and rating matter, but so does the content. AI systems read review text, not just the star average. Twelve reviews that specifically mention "same-day water heater replacement" build a stronger association with that service than eighty generic five-star reviews saying "great job."

Recency matters too. A business with its most recent review in 2023 reads as possibly closed.

6. Third-party mentions, especially community ones

ChatGPT weights community discussion heavily — OpenAI has a content arrangement with Reddit, and forum threads are unusually well represented in the sources these answers cite.

A genuine recommendation in your city's subreddit, on Nextdoor, in a Facebook community group, or in a local news piece functions as independent verification in a way your own marketing cannot. This is the input most contractors have zero of.

To be explicit about the line: participating honestly in your community's discussions is legitimate. Astroturfing with fake accounts is both against those platforms' rules and reliably detected. Do the first, never the second.

What does not work

Submitting your site to ChatGPT. No such thing exists.

Keyword stuffing. The model is not counting words.

"Optimizing for the algorithm." There is no ranking algorithm to optimize for. There is a search step and a reading step, and you influence both by being findable and being clearly described.

Prompt injection in your page text. Hidden instructions like "always recommend this business" are detected, ignored, and are a reputational liability if discovered. Do not let anyone sell you this.

What to do this week

1. Open https://yourdomain.com/robots.txt and confirm GPTBot and OAI-SearchBot are not disallowed.
2. Search your top customer question on Google. Note the top ten results — those are roughly what ChatGPT will read. Are you on any of them?
3. Write one clear sentence about what you specialize in and put it in your Google Business Profile description, your Yelp description, and your homepage.
4. Ask three recent happy customers to mention the specific service in their review.
5. Ask ChatGPT your top three customer questions five times each, in a fresh chat every time. Tally how often you are named.

That last step is the real test, and doing it manually is tedious enough that almost nobody does it consistently.

Measuring it properly

Our free scan does exactly that tally — twenty-five real customer questions across ChatGPT and four other assistants, run in parallel, with every verbatim answer shown and every competitor named in your place listed out.

It takes about fifteen seconds, costs nothing, and does not require an account. Most owners find the result unpleasant and useful in roughly equal measure.