If you searched for how to get HVAC company recommended by ChatGPT, you have probably seen the same checklist everywhere: match your name, address and phone across listings, polish your Google Business Profile, claim Bing Places, add schema, collect reviews. None of that is wrong. The mistake is doing all of it before you know what the AI tools actually say about HVAC in your cities.
Most owners test once. They type “best HVAC company near me” into ChatGPT, don’t see their name, and start fixing everything at the same time. That single prompt tells you very little, because answers change from run to run, from city to city, and from platform to platform.
A better approach is to test the way a good tech diagnoses a system: run real homeowner prompts, log what comes back and which sources get cited, then fix the sources that matter in your market. Here is how to do that, and when to do it so you are ready before the first cold snap.
Key takeaways
- Test before you fix. Run 10 to 20 service and symptom prompts per city, several times each, before you change listings or pages.
- Log sources, not just names. The sites each AI tool cites show you which listings, reviews and pages it trusts in your market.
- Treat each platform separately. ChatGPT, Perplexity, Gemini and Google AI Overviews draw on different data, so a win on one does not carry to the others.
- Check near me prompts against the map pack. Many local searches still show a map pack instead of an AI answer, so Maps visibility still counts.
- Time it to the season. Test in late summer, fix sources in the shoulder season, and re-test before heating calls spike.
What is documented and what is guessed about AI local recommendations
Very little about how AI tools choose a local contractor is officially documented. What Google and OpenAI publish covers crawling, indexing and eligibility, not a ranking formula for “best HVAC company in town.”
That matters because a lot of advice online states guesses as facts. Before you spend money, know which is which.
What Google and OpenAI actually say
Google’s Search Central guidance on AI features says the same SEO fundamentals apply to AI Overviews and AI Mode. Your pages need to be crawlable, indexed and eligible to appear in Search, and there is no special markup that guarantees inclusion.
For Maps and local results, Google’s Business Profile Help explains that local ranking is based on relevance, distance and prominence. Gemini and AI Overviews sit inside the Google ecosystem, so that local data is part of the picture.
OpenAI documents a crawler called OAI-SearchBot, which is used to surface websites in ChatGPT search. If your site blocks it, your pages may not show up in ChatGPT search answers. OpenAI does not publish a full list of where ChatGPT gets local business data.
What is still guesswork
One widely shared claim says a single location data provider supplies most of ChatGPT’s local results and that Google rankings have no effect at all. Other testers have pushed back, arguing the original figure was misread and that review directories such as Yelp show up in most of their test runs.
You will also see reports of licensing deals between review platforms and AI companies. Deals like that can change what gets cited, but treat them as reported news until you see them confirmed by the companies involved.
The practical point: do not build your plan on one statistic. The only data that should drive your fixes is what the AI tools say in your cities, for your services.
| Claim | Status | What to do |
|---|---|---|
| Blocking OAI-SearchBot can keep your site out of ChatGPT search | Documented by OpenAI | Check robots.txt and any firewall or CDN bot rules |
| AI Overviews use standard Search eligibility | Documented by Google | Confirm your key service pages are indexed |
| Google local ranking depends on relevance, distance and prominence | Documented by Google | Keep your Business Profile complete and accurate |
| One directory supplies most ChatGPT local data | Disputed | Log cited sources in your own market |
| Google rankings have no effect on ChatGPT | Unproven | Track both and keep investing in Google |
Pro tip: Some security plugins and CDN bot settings block AI crawlers by default. Ask your web developer to confirm OAI-SearchBot is allowed at the server level, not just in robots.txt.
Build a prompt set by service, symptom and city
A useful test set mirrors how homeowners actually ask: by service, by symptom and by location. Aim for 10 to 20 prompts per city that cover your money services, not just “best HVAC company.”
Homeowners rarely ask an AI tool for a company in the abstract. They describe a problem, often at a bad moment, and ask who to call. Your prompt set should reflect that.
| Prompt type | Example prompt | What it tests |
|---|---|---|
| Best-of | “Who is the best HVAC company in [city]?” | Overall brand prominence |
| Emergency | “My furnace stopped working tonight. Who can come out now in [city]?” | 24/7 and emergency signals |
| Symptom | “My heat pump is blowing cold air. Who should I call in [city]?” | Whether you are tied to repair problems |
| Symptom | “Furnace keeps short cycling. HVAC repair near [neighborhood]?” | Repair relevance in a specific area |
| Replacement | “Who installs heat pumps in [city] and offers financing?” | High-ticket install visibility |
| Maintenance | “Furnace tune-up companies in [city] with good reviews” | Seasonal maintenance and review strength |
Vary the location
Test your core city plus two or three suburbs in your service area, including one near the edge where you want more calls. Results for a suburb can look very different from results for the city it borders.
Run each prompt with the city typed out, and also run a near me version. If you can, have a tech or office staffer who lives in a target suburb run the near me prompts on their own phone, since location settings shape what local results appear.
Keep prompts plain
Write like a homeowner, not a marketer. Skip industry terms your customers never use, and include a few prompts with no brand or service category, just the problem.
Add one prompt that names you directly, such as “What do people say about [your company] in [city]?” That shows how AI tools describe you when someone already knows your name. For a wider view of how this fits into strategy, see what good LLM SEO strategy looks like beyond traditional rankings.
How to run the test so the results mean something
Run each prompt several times, in a clean session, on each platform, and log every business and source. AI answers vary, so one run is an anecdote and three runs start to show a pattern.
This is a few hours of work spread across a week. It splits well between office staff during slow hours.
- Set up clean sessions. Use a logged-out browser, a temporary chat or a fresh account with memory turned off, so past conversations do not bias the answer.
- Run each prompt three times per platform. Space the runs apart, ideally on different days, and keep the wording identical.
- Log every business named, in order. Record your company and every competitor, including where each appears in the answer.
- Record the cited sources. Click every link or source chip and note the site: your website, Google Business Profile, Yelp, Angi, BBB, a local news story, a manufacturer dealer locator or an article.
- Note what the answer says about you. Check hours, services, service area, phone and any claims like 24/7 service. Wrong details point straight to a listing that needs fixing.
- Repeat for each city. Use the same prompt set so you can compare markets side by side.
- Save dated screenshots. You will need them for the re-test, and they settle arguments about what changed.
If the workload looks heavy, start with 10 prompts and two cities. A smaller set you actually repeat beats a big set you run once.
Score your results and find the sources to fix
Turn raw runs into a simple score per prompt, then look at which sources keep showing up next to the companies that win. Those sources are your fix list.
| Result in a run | Points | What it means |
|---|---|---|
| Named first or recommended outright | 3 | Strong position for that prompt |
| Named in the list, not first | 2 | Visible but not preferred |
| Named with wrong or outdated details | 1 | Visible, but a source is feeding bad data |
| Not named | 0 | Gap to investigate |
Average the points across runs for each prompt. Then calculate a simple visibility rate: the share of runs where you were named at all.
Do the same for your top three competitors. A low score matters less if nobody else scores well either, and a lot more if one competitor shows up in nearly every run.
Read the cited sources
Sort your log by source. When the same site keeps appearing beside the winning companies, that is where to work first.
- Your website. If it gets cited, make sure service and city pages say plainly what you do, where you do it, and whether you offer emergency service. Vague pages give AI tools nothing to quote.
- Google Business Profile. Check primary and secondary categories, services, hours and service area. Our guide to what actually drives Google Maps rankings for local service brands covers the details.
- Bing Places. It is easy to forget and often out of date. Claim it and correct it so any AI tool that draws on Bing data finds the same details as everywhere else.
- Yelp and other directories. If Yelp, Angi or BBB keep getting cited, claim the profile, match your categories to your real services and respond to reviews.
- Review recency and detail. Reviews that mention the actual job, like a furnace repair or heat pump install, give AI tools specific language to match against symptom prompts.
- Third-party mentions. Local news, trade associations and manufacturer dealer pages can show up as sources. Note which ones cite competitors but not you.
- Consistency. Any listing with an old phone number or address is a candidate for the wrong details you logged.
Listings, reviews and SEO work best when they tell the same story. See local SEO vs listings vs reviews for how they fit together.
Pro tip: Fix only what the test points to. If no AI tool in your market ever cites a particular directory, it should not be first on your list, no matter what a checklist says.
Score each AI platform on its own
Each platform pulls from different places and presents local results differently, so score them separately. A strong ChatGPT score with a weak AI Overviews score points to different fixes than the reverse.
ChatGPT
ChatGPT search shows sources alongside many local answers, and it may show a list or map of businesses. Pay close attention to which directories and review sites appear, and whether your own site is cited at all.
If your site never appears as a source, check OAI-SearchBot access first. It is the one documented, fixable cause.
Perplexity
Perplexity puts citations front and center, which makes it the easiest place to see which pages it trusts. It often cites articles and directory pages, so a weak score here can point to thin third-party coverage rather than a listings problem.
Gemini
Gemini is a Google product and tends to lean on Google’s own local data. Weak Gemini results usually send you back to your Business Profile, reviews and Google-indexed pages.
AI Overviews and the map pack
For many near me searches, Google shows the map pack without an AI Overview at all. So for every near me prompt, record two things: whether an AI Overview appeared, and where you landed in the map pack.
If you skip the map pack, you can conclude you are invisible when you are actually getting the calls. Our post on AI Overview SEO mistakes that keep brands out of the answer covers the content side of earning a mention.
A seasonal timeline for testing, fixing and re-testing
Run your baseline in late summer, spend the shoulder season fixing the sources your test flagged, and re-test in October before heating calls pick up. That way your fixes have time to show up before demand does.
Seasonality matters more in HVAC than in most trades. A company with a wave of AC reviews in July and nothing after can look current for cooling prompts and stale for heating prompts.
- August: baseline. Run the full prompt set with heavy weight on heating, furnace and heat pump symptom prompts. If it is already September, run the baseline this week.
- September: fix sources. Update Business Profile services and hours for heating season, refresh furnace and heat pump pages, correct any wrong listing details, and ask maintenance customers for honest reviews in a way that follows each platform’s review policies.
- October: re-test. Use the same prompts, cities and platforms, then compare scores against August. Expect gradual movement, not overnight changes.
- Spring: repeat for cooling. Run the same cycle for AC prompts before the first heat wave.
AI answers shift without warning, so a twice-a-year test cycle is a minimum. Busy markets may want a quick monthly spot check on the five most valuable prompts.
Frequently asked questions
How do I check whether ChatGPT recommends my HVAC business?
Run 10 to 20 realistic homeowner prompts for each city you serve, three times each, in a clean logged-out session. Log every business named and every source cited, then score the results. One quick search is not enough, because answers vary between runs.
Why doesn’t ChatGPT show my HVAC company?
Common causes include a site that blocks OAI-SearchBot, thin service and city pages, missing or outdated directory listings, and few recent reviews. Your test log will tell you which sources competitors are getting credit from that you are not. Start there instead of guessing.
Do Google reviews and rankings affect ChatGPT recommendations?
OpenAI does not publish how ChatGPT weighs Google data, and industry claims conflict. Google reviews and rankings clearly matter for Gemini, AI Overviews and the map pack, which is reason enough to keep investing in them. Track ChatGPT separately to see what moves it in your market.
Which directory listings do I need for AI search?
Claim and correct your profiles on the major directories, such as Yelp, Bing Places and Foursquare, since it costs little and prevents bad data. Then put real effort into whichever ones your test shows AI tools citing in your cities. If a directory never shows up as a source, it belongs lower on the list.
How long does it take to get recommended by AI tools?
There is no documented timeline, and no one can guarantee placement. Listing corrections can show up relatively quickly, while reviews and third-party mentions build over months. That is why the plan above starts testing in late summer, well ahead of heating season.
How do emergency and 24/7 HVAC searches work in AI?
AI tools look for clear signals that you actually answer after hours, such as accurate hours on your listings and pages that state emergency service plainly. Make sure every listing says the same thing, and that someone or something actually answers the phone at night. Our guide to after-hours lead response covers the call-handling side.
How Ad Leverage approaches AI search for HVAC companies
We start every AI search engagement with the testing method above: prompt sets by service, symptom and city, repeat runs on each platform, and a scoring sheet that shows where you stand against local competitors. The fix list comes from that data, not from a generic checklist.
From there, the work usually spans your website, Google Business Profile, Bing Places, directories like Yelp, and review flow. Because we work in ServiceTitan accounts every day, we tie visibility changes back to booked jobs where tracking allows, so you can see whether better AI placement is turning into revenue.
If you want a baseline before heating season, our AI search team can run the test and walk you through what it shows.
