How to get your business recommended by ChatGPT

The short answer

ChatGPT recommends the businesses it can find, read, and quote. When a homeowner asks it for a roofer or an air conditioning company, it searches the web, reads a handful of pages, and names the businesses those pages describe well. To be one of them, a home-services business has to do five things:

1. Let ChatGPT's search crawler in. If OpenAI's crawler cannot reach your site, ChatGPT search will not show it.
2. Publish your own facts as data a machine can read. Prices, service areas, warranties, specs, and answers to your customers' questions, with your own numbers, in an llms.txt file, structured data, and downloadable datasets.
3. Be on the sites ChatGPT reads. Directories, review sites, and local articles carry real weight in its answers.
4. Say the same thing everywhere. One name, one phone, one service area, one description, on your site and every profile.
5. Measure, every week. One answer tells you little. The share of answers that name you, over weeks, tells you whether the work is landing.

None of this guarantees a recommendation, and anyone who promises one is guessing. It makes you findable and quotable, and then the weekly numbers show what moved. The rest of this guide takes each step in turn.

Why this matters now

OpenAI says ChatGPT has more than 900 million¹ weekly active users. BrightLocal's latest consumer survey found 45%² of U.S. consumers used ChatGPT or another AI tool for local business recommendations in the past year. And most owners do not yet know what to do about it: in Scorpion's survey, 80%³ of home-services business leaders said they are unsure how to appear in AI-driven search results.

The list at the top of an AI answer is short. SOCi found ChatGPT recommended 1.2%⁴ of the locations of the multi-location brands it studied, far below how often the same locations appeared in Google's local pack. That study covers large brands rather than independent contractors, but the lesson carries: ChatGPT names a few businesses, not a page of them.

Step one: let ChatGPT's search crawler in

OpenAI states that OAI-SearchBot⁵ is the crawler it uses to surface websites in ChatGPT's search features, and that sites opted out of it are not shown in ChatGPT search answers. Its help center puts the eligibility rule plainly: allow OAI-Searchbot to crawl the site⁶, and make sure your host or content delivery network allows traffic from OpenAI's published search-bot IP addresses.

Check three places:

• robots.txt. Look for a rule that disallows OAI-SearchBot by name, or a blanket rule that catches every bot.
• Your host or CDN. Bot protection on some hosts and CDNs turns away crawlers it does not recognize. A clean robots.txt does not help if the firewall says no.
• The training crawler is separate. OpenAI runs GPTBot⁷ for model training and controls it independently. You can decline training and still allow search.

OpenAI's help center for business workspaces also says ChatGPT may share search queries with Bing⁸ to return web results. So confirm Bing has indexed your main pages, not just Google.

Our guide to llms.txt and AI crawler access walks through each check. If you only do one thing this week, do this one: everything else depends on ChatGPT being able to read your site.

Step two: publish your own facts as data

This is the step most businesses, and most agencies, have never done, and it is where the gains are.

When ChatGPT chooses between two contractors, it leans on what it can quote: a price range for a job type, a warranty term, the cities a crew covers, how long an install takes, what a part costs to replace. If those facts live only in a salesperson's head or a PDF quote, ChatGPT cannot use them. If they are published on your site, in plain tables, with structured data that matches, it can.

Build three things:

• An llms.txt file. A plain-text map of your site for AI engines: your services, your prices, your data, and the pages that answer your customers' questions. It costs little, and it states what you want an engine to know in one place. Our llms.txt guide shows the format.
• Structured data for your own facts. Not just a name and address. Your price ranges as Offer data, your service areas, your warranties, and your product specs, all on one schema.org business entity with a stable identifier. Our guide to structured data for AI visibility lists which details matter.
• First-party datasets. Your own numbers, published as a table with Dataset markup and a CSV or JSON download: what jobs cost in your market, how long they take, what fails and when. A dataset answers a customer's question with your numbers, which is exactly the kind of source an engine can cite.

Then write pages that answer one customer question each, using those numbers, under the name of a real person on your team. A page that answers what a roof repair costs in your city, with figures from your own jobs, is quotable. A generic post about "the importance of roof maintenance" is not.

Step three: be on the sites ChatGPT reads

ChatGPT reads more than contractors' own websites. In BrightLocal's test of local searches across ChatGPT Search, Gemini, Google AI Mode, and Perplexity, every platform used directories, and Yelp was used as a source in 33%⁹ of the searches. Yext's study of AI citations for location questions found 86%¹⁰ of them came from sources brands control or manage, such as their websites and listings. Yext sells listings management and studied other industries, not home services, but the direction is the same: what you publish, and what your listings say, is most of what the engines read.

In our own weekly measurements of home-services questions, which started in Phoenix, the website the engines cited most was energyshieldwindowsanddoors.com¹¹. The sites that matter differ by trade and market, so find out which ones the engines cite for your questions, then make sure you are on them with complete, current profiles. Our study of which websites AI assistants cite for Phoenix contractors is a worked example of how to find them for any market.

Reviews count too. SOCi reports that locations ChatGPT recommends average 4.3-star ratings¹². That is a vendor's correlation across large brands, not a rule, but it is a reason to keep asking happy customers for reviews on the sites the engines read.

Step four: say the same thing everywhere

ChatGPT has to work out that your website, your Google Business Profile, your Yelp page, and your BBB listing are one business. Make that easy:

• One schema.org business node on your site, with a stable identifier, referenced from every page.
• Links from that node to each of your profiles (the sameAs property), so an engine can connect them.
• The same name, address or service area, phone, and hours everywhere.

Before you "fix" a second listing or a second phone number, ask what else the business does. A second listing can belong to a separate line of business on purpose. If it does, the job is to keep the two clearly distinct, not to merge them.

This cleanup matters, but it is the floor, not the ceiling. Consistent listings stop the engines from getting confused. Your own published data, from step two, is what gives them a reason to choose you.

Step five: measure every week

Google rank does not carry over on its own. In our first market comparison, in Phoenix, the share of the businesses Google showed that Claude never named was 96%¹³. ChatGPT is a separate engine with its own searching, so the only way to know where you stand in it is to ask it.

Ask ChatGPT the questions your customers ask, with search on, more than once, and record whether your business is named, whether your site is cited, and whether you are recommended. Then do the same in Gemini, Perplexity, and Claude, because the engines read different sources and name different businesses. How to check whether ChatGPT recommends your business shows how to run that check by hand.

Answers do not always show their sources. Similarweb found roughly 6.8%¹⁴ of U.S. ChatGPT prompts had answers with web citations by May 2026. So track names and recommendations, not only links, and track them over weeks. Our methodology page explains how we measure this for clients, and our glossary defines the terms: named, cited, recommended, and share of voice.

What not to do

• Do not buy a ranking promise. No one controls what ChatGPT says. Buy measurable outputs: data published, listings fixed, citations tracked.
• Do not publish generic AI-written posts. They say what every other site says, so there is nothing in them to cite.
• Do not block OpenAI's search crawler to stop training. Block the training crawler if you want to; keep the search crawler allowed.
• Do not judge from one answer. Answers change from one ask to the next. Measure the pattern.

What this means for an owner

Start with access: confirm OAI-SearchBot can reach your site, today. Then build the data: an llms.txt file, structured data for your prices, service areas, and warranties, and at least one dataset that answers a customer question with your own numbers. Get onto the directories the engines cite for your trade, with the same facts everywhere. Then measure the answers every week and keep the facts current.

You can do this yourself. If you would rather have it done and measured for you, that is what CiteMy does, month to month. Either way, start with our free AI visibility audit: it checks whether OAI-SearchBot and the other AI crawlers can reach your site, reads your llms.txt and structured data the way the engines do, checks whether your name, address, and phone agree across the web, and emails you the findings that matter most.