How customers use ChatGPT to choose a local business
A customer who needs a dentist, a plumber or a cleaner used to type a few words into Google and open three tabs. Now some of them ask an assistant instead: “Who is a good emergency plumber near Somerville, and roughly what does it cost?” The assistant answers in a paragraph, often with a few links.
That changes what your website is for. It still has to convince a person. It also has to be readable by a machine that is trying to compare you with two or three other businesses. This guide is about the second job. It is part of our work on AI search optimization, and it is written for owners, not for SEO specialists.
The thesis is simple. Assistants compare what they can read. If your services, prices, service area, hours, reviews, reply time and booking link are written down in plain text on your site and on your profiles, there is something to compare. If they are hidden in a PDF, a photo, a pop-up or your head, there is nothing to compare, and a competitor who wrote them down gets the mention.
One honest limit comes first. Nobody outside OpenAI knows exactly how ChatGPT chooses businesses, and OpenAI does not publish a ranking formula. What follows is what the primary sources say, plus practical steps that make you easier to read and easier to measure.
What we know about how ChatGPT finds local businesses
Here is what OpenAI itself says, from its help pages and crawler documentation (read in October 2026):
- ChatGPT may search the web on its own. OpenAI says it “may search the web automatically” when a question would benefit from current information. Users can also switch search on by hand.
- It uses outside search providers. OpenAI says ChatGPT search “sometimes partners with other search providers” and typically rewrites the question into one or more targeted queries. The help page links to Microsoft’s and Shopify’s privacy statements for those providers. For Enterprise and Edu workspaces, OpenAI states that queries may go to the Bing search engine. For other plans, the help page does not say which provider handles which query. We go deeper in where ChatGPT gets its information.
- It uses location. OpenAI says ChatGPT may use an approximate location from the IP address, or a precise location if the user shares it, to find local results. It may share location with third-party providers for things like nearby business listings. Customers can also type a city or neighborhood into the question.
- It has its own crawler for search. OAI-SearchBot “is used to surface websites in search results in ChatGPT’s search features.” To be eligible, OpenAI says to allow OAI-SearchBot in robots.txt and to make sure your host or CDN allows OpenAI’s published IP addresses.
- It says placement is not guaranteed. OpenAI writes that ChatGPT ranks results using multiple factors and that placement is not guaranteed. It also warns that results and citations can be incomplete, outdated or incorrect.
What OpenAI does not say matters as much. It does not publish how businesses are ranked, how much reviews count, or whether schema markup helps. So when you read a confident “ChatGPT ranking factors” list, treat it as someone’s guess unless it cites OpenAI.
What an assistant can read about your business
Think of the questions a customer asks before choosing you. Each one needs an answer in text somewhere the assistant can reach.
| Customer question | Where the answer should be | What to write |
|---|---|---|
| What do you do? | Service pages | One page per service, with what is included and who it is for |
| Do you come to me? | Service pages, Google Business Profile | Towns or neighborhoods you serve, written as a list |
| What does it cost? | Service pages | A “from” price or a typical range, and what is extra |
| Are you open now? | Site footer, profiles | Hours, including holidays |
| Are you any good? | Reviews, your site | Review count and rating on your profiles; short quotes you have permission to use |
| How fast do you reply? | Contact page | The reply window you actually keep, stated once |
| Can I book online? | Every service page | A visible booking link or form, and what happens next |
| Are you licensed and insured? | About page, service pages | License numbers where your trade requires them, and insurance in plain words |
Two cautions. First, only publish what is true and what you can keep. A reply promise you miss is worse than none. Second, we do not claim that any assistant weighs reply speed, booking or price. We claim something smaller: a customer cares about those facts, and an assistant can only repeat what it can find. In Steady Demand’s 2026 study of 14,472 Gemini local citations (not ChatGPT; reported by MarTech in August 2026), about six in ten citations pointed to the businesses’ own websites. That is one study of one assistant, but it fits the idea that your own pages are the main source you control.
A 10-step plan to show up in ChatGPT
Work through these in order. Steps 1 to 4 take a day or two. The rest take a few weeks of steady work.
Step 1. Let the search crawlers in
If OAI-SearchBot cannot fetch your pages, you are not eligible for ChatGPT search answers. Open your robots.txt (it lives at yourdomain.com/robots.txt) and check three things:
- No rule blocks OAI-SearchBot, Googlebot, Bingbot or PerplexityBot.
- Your host, CDN or security plugin is not blocking those bots at the firewall. OpenAI asks site owners to allow its published IP ranges. Perplexity says the same for PerplexityBot.
- The pages you want read are not behind a login, a pop-up wall or a “type your phone number to continue” gate.
Here is an example robots.txt that allows these search crawlers. Replace the sitemap address with your own, and keep any rules your site already needs.
User-agent: OAI-SearchBot
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: Googlebot
Allow: /
User-agent: Bingbot
Allow: /
Sitemap: https://www.example.com/sitemap.xmlOpenAI says a robots.txt change can take about 24 hours to take effect for its search crawler. Perplexity gives the same figure.
What about GPTBot? OpenAI documents GPTBot as a separate crawler for training its models. Disallowing it does not remove you from search, because OAI-SearchBot is its own setting. Whether to allow training is your call. A third agent, ChatGPT-User, visits pages when a user asks ChatGPT to open them, and OpenAI says robots.txt rules may not apply to it.
Step 2. Put your facts in text on the page
A page that says “Welcome to our family business” gives an assistant nothing to compare. A page that says (for illustration) “Drain cleaning in Somerville, Cambridge and Medford. Same-day visits Monday to Saturday. Most jobs $180-$450 before parts. Book online or call” does.
Google’s own guidance says the same thing in its terms: keep important content in text, make pages easy to find through internal links, and make sure pages are indexed and eligible for a snippet. Text inside images, sliders and chat widgets is easy to miss.
For each service, write one page with:
- What the service is, in one sentence a customer would use
- What is included and what is not
- Who it is for, and when it is the wrong choice
- Where you do it
- A “from” price or typical range, and what changes the price
- How long it takes
- How to book, with a visible button or link
If you are not sure what a good service page looks like, web design in Boston shows how we structure ours.
Step 3. Make your name, address and phone number identical everywhere
Your business name, address and phone number are the facts that tie your website, your Google Business Profile, your Bing Places listing, Apple Business Connect, Yelp and your social profiles to one business. If they disagree, any system that reads several sources has to guess.
Do a one-hour audit. Open each profile and copy the name, address, phone, hours and website into one sheet. Fix every difference, including “Street” vs “St.”, suite numbers and old phone numbers. Pick one spelling of the business name and use it everywhere.
OpenAI’s help page says ChatGPT may share location with third-party providers for things like nearby business listings. We do not know which listings feed which answers. That is the reason to fix all of them, not one.
Step 4. Complete your Google Business Profile
Google Business Profile is not ChatGPT. It matters because it is the largest public record of a local business, and because Google’s own AI features can draw on it (Google’s guide says Business Profiles can make products and services visible in AI responses and other results).
Google says local results depend on relevance, distance and prominence. Complete and accurate information makes a profile more likely to appear. More reviews and positive ratings can help local ranking, and replying to reviews can help you stand out. Google also states there is no way to request or pay for a better local ranking.
Check: verified profile, the most specific primary category, service areas, hours (including special hours), services with descriptions, a booking link, current photos, and your phone number. See local SEO and AI search visibility in Boston for how this fits together locally.
Step 5. Show prices, or a range
Many owners avoid prices because every job is different. A range with an honest “it depends on” list still beats silence. A customer asking an assistant “how much does a roof repair cost” gets an answer from somewhere. If your page has a range and a reason, your page can be that somewhere.
Write the range, the three things that move it, and what is not included. Date it (“Prices as of October 2026”). Update it when prices change.
Step 6. Earn and answer reviews
Reviews are the part of your reputation that you do not write yourself. Ask every satisfied customer for a review while the job is fresh, with a direct link, and reply to every review, good or bad, in a calm, specific way.
We do not know how much ChatGPT weighs reviews. Google does say review count and rating feed local prominence. Reviews also carry the words customers use for your service and your area, which is useful whatever reads them. Never buy or invent reviews. Google’s structured data rules say not to mark up fake reviews, and a fake review is a real risk to your reputation.
Step 7. Make booking and reply speed visible
If a customer asks an assistant for “a dentist who takes new patients and has Saturday hours,” the assistant needs those facts. State them: “New patients welcome. Saturday appointments 9am-1pm. Book online or call.” On your contact page, state your real reply window: “We answer calls Monday to Friday, 8am-5pm, and reply to forms within one business day.”
State only what you do. If the honest answer is that calls go to voicemail after 5pm, say “leave a message and we call back the next business morning,” then make that true. (If missed calls are your real problem, that is a separate fix from search, and we cover it elsewhere on this site.)
Step 8. Add schema markup
Schema markup is a block of structured data that states your name, address, services, hours and FAQs in a form machines read directly. Google says it is not required for its AI features and that there is no special schema for them. OpenAI does not list it as a requirement either. We still add it, because it states facts without ambiguity and it is one of the few things on a page that a person does not have to read to understand. We show a full example, with a test, in schema markup for local businesses.
Make sure the markup matches what visitors can see on the page. Google’s guidelines say not to mark up content that readers cannot see.
Step 9. Verify with Bing and submit changes
Bing Webmaster Tools is free. Microsoft’s own post on its AI Performance report says it shows when your site is cited in answers across Microsoft Copilot, AI summaries in Bing and select partner integrations. Microsoft’s advice for publishers includes clear headings, tables and FAQ sections, regular updates, and keeping address and hours current in Bing Places for Business. It also says IndexNow helps keep information fresh across search and AI experiences.
Be careful about what this proves. OpenAI says Enterprise and Edu searches may go to Bing, and it does not tell us how much Bing data reaches other ChatGPT plans. Verifying your site with Bing is cheap and useful for Copilot and Bing anyway. We just do not promise it moves ChatGPT.
Step 10. Keep it current and date it
An assistant that finds an old price or an old phone number repeats it. Put a visible “Updated” date on service and price pages. Review them every quarter. When you change hours, prices or phone, change every profile the same day.
How to check where you stand
You can test this yourself in about half an hour, for free.
- Write 10 questions a real customer would ask. Mix kinds: “best [service] in [town]”, “how much does [service] cost in [town]”, “[service] open Saturday near [town]”, “is [your business] good”, and “[your business] vs [competitor]”.
- Ask each question in ChatGPT, Google (AI Mode or AI Overviews), Perplexity and Copilot. Use a fresh chat each time.
- In a spreadsheet, record whether you were named, who else was named, and which sources were cited.
- Open the cited sources. Look for wrong facts about you. Wrong hours or a wrong service area usually point to one profile or page you can fix.
- Repeat the same questions in two weeks, then monthly.
Expect noise. In the Steady Demand study of Gemini local queries, repeated searches shared the same sources only about 40% of the time and named the same top business only about 7% of the time. The same study found Gemini and ChatGPT agreed on the top business about 4% of the time. So one run proves nothing. Look at patterns over several runs: which facts get repeated, which sources keep appearing, which of your pages get cited.
Some data comes free. Google’s 2026 guide points to a Generative AI performance report in Search Console, and Google also counts AI Overview and AI Mode traffic in the Performance report under the Web search type. Bing Webmaster Tools has AI Performance, which counts citations in Copilot and Bing’s AI summaries. Neither tells you your “rank” in an answer: Microsoft says its numbers do not show ranking or a page’s role in an answer. Be wary of third-party tools that claim to see inside an AI system. Google says no third-party tool has access to its internal ranking or AI systems.
What nobody controls
This is the part most guides leave out.
- Placement. OpenAI says placement is not guaranteed, and Google says indexing and serving are not guaranteed either. Anyone who sells a promised spot in AI answers is selling something they cannot deliver.
- Which sources are used. OpenAI does not publish its provider list for every plan or its ranking factors.
- Consistency. The same question can give different answers on different days. OpenAI’s own help article on how its models are developed says there is an inherent element of randomness, so the same question may yield different answers.
- Accuracy. OpenAI warns that search results and citations can be incomplete, outdated or incorrect. An assistant can state something wrong about you even when your site is right.
- The customer’s context. Location, memory settings and wording change the answer. OpenAI says memory can influence the rewritten search query.
- Whether the customer clicks. Many answers satisfy the person without a visit. That is why your phone number, hours and booking link need to be right wherever they appear.
What you do control: whether your facts exist, whether they are consistent, whether crawlers can reach them, and whether you measure. Those are the steps above.
What does not work
- Keyword-stuffed “AI pages.” Google’s guide warns that mass-producing pages for query variations can violate its scaled content abuse policy. Write fewer, better pages.
- llms.txt as a fix. Google says its Search does not use llms.txt and that creating one will not help or harm visibility. We know of no measured effect for ChatGPT either. See llms.txt for what it is and why it is low priority.
- Fake mentions and fake reviews. Google says inauthentic mentions do not help, and its spam systems affect AI features.
- Rewriting every page for “AI voice.” Google says rewriting content for AI is not necessary, because these systems understand synonyms and intent.
- Paying for placement. We found no OpenAI documentation of a paid route into search answers, and Google says you cannot pay for local ranking.
The pattern is the same as classic SEO: useful, accurate, crawlable pages win slowly. Google puts it this way: optimizing for generative AI search is optimizing for the search experience. For a plain comparison of the terms, read GEO vs SEO vs AEO. For Google’s version of this story, see Google AI Overviews.
A four-week plan for a small business
Week 1. Check robots.txt and firewall for OAI-SearchBot, Googlebot, Bingbot and PerplexityBot. Make a sheet of name, address, phone, hours and website from every profile. Fix every mismatch.
Week 2. Complete your Google Business Profile and Bing Places listing. Verify your site in Bing Webmaster Tools and Google Search Console. Write down the 10 customer questions and run the first test.
Week 3. Rewrite your top three service pages with the table in this guide: what, where, price, hours, booking, reply window. Add a visible “Updated” date.
Week 4. Add schema markup that matches the new pages. Ask your last 10 customers for reviews. Run the 10 questions again and compare.
After that, re-test monthly and fix the facts that assistants get wrong.
How Magnet Mint can help
We are a Boston studio that builds websites for local service businesses that customers, Google and AI assistants can read. For AI search, we start with a written audit: a map of your pages, a service page template, missing information, schema, a check of your listings, and a record of what AI answers say about you today, with a re-check after the fixes. The GEO/AEO audit is from $750 and takes 1-2 weeks.
If you want to see where you stand first, ask for a free demo (30-minute call; demo shared within 3 business days). Or call 617-991-9912, Monday to Friday, 9am to 5pm ET.
Published . Updated .