Picture this: it's the first 95-degree day of June, and a homeowner in your service area opens ChatGPT and types, “Who's the best AC installation company near me? My unit is 15 years old.” The answer comes back with three company names, a sentence about each, and a suggestion to get multiple quotes. If your company isn't one of those three names, you never knew that customer existed. No click, no call, nothing in your analytics.
That's the uncomfortable part of AI search. When a competitor outranks you on Google, you can see it. When ChatGPT recommends a competitor, it happens in a private conversation you'll never see unless you go looking. Here's how to look: measure whether AI assistants mention you, compare yourself against competitors, and fix the gaps. We'll use an HVAC company as the example, but the process works for any business that serves customers in a specific area.
Why AI answers are now a real discovery channel
For years, local discovery meant the Google map pack and the organic results below it. That's no longer the whole picture. Homeowners now ask ChatGPT, Perplexity, Gemini, and Microsoft Copilot the same questions they used to type into a search bar. Google itself shows AI Overviews above many results and offers AI Mode, a conversational search experience where the answer comes first and the links come second.
The difference matters. A results page shows ten options and lets the customer choose. An AI answer usually narrows the list to a handful of names, often with a short reason attached to each: “known for fast emergency service,” “strong reviews for heat pump installs.” Being one of those names is closer to a referral than a ranking. Being left out means you weren't considered at all. For background, our guide to generative engine optimization covers the basics.
Step one: build a list of real customer prompts
You can't measure AI visibility in the abstract. You measure it prompt by prompt. Start by writing down 15 to 30 questions your customers actually ask. Pull them from phone calls, website chat logs, the questions your techs hear in the field, and the search terms already showing up in Google Search Console.
For an HVAC company, a solid starting list mixes a few types of prompts:
- Direct recommendation prompts: “best AC installation company in [city],” “top-rated HVAC contractor near [neighborhood].”
- Service-specific prompts: “who repairs heat pumps near me,” “who installs ductless mini splits in [city].”
- Problem prompts: “my AC is blowing warm air, who should I call in [city],” “emergency furnace repair on a weekend in [city].”
- Cost and comparison prompts: “how much does a new central AC cost in [city],” “is it worth replacing a 15-year-old air conditioner.”
- Trust prompts: “most reliable HVAC company in [county] with financing,” “HVAC companies in [city] with good warranties.”
Write them the way a real person would, with casual phrasing and your city name. AI answers change a lot based on wording, so real language gives you a more honest picture than generic keywords.
Step two: run a manual prompt audit
Take your list and run every prompt through several assistants: ChatGPT, Google (AI Overviews in regular search, plus AI Mode), Perplexity, Gemini, and Copilot. Use logged-out or fresh sessions where you can, or a private browser window, so your past chats and location history don't tilt the results in your favor. If an assistant asks for a location, give it the city you're testing, not your office address.
For each prompt and each assistant, record the same handful of facts in a spreadsheet:
- Was your business mentioned? Yes or no.
- Which competitors were mentioned, and in what order?
- Was your website cited as a source, and which other sources were cited (directories, review sites, local news, competitor sites)?
- How was your business described? Note the exact wording and whether the tone was positive, neutral, or negative.
Run each prompt more than once. AI answers vary from one session to the next, so a single run is a snapshot, not a measurement. Two or three runs per prompt per assistant gives you a much steadier read. Yes, the first round takes a few hours. It's time well spent.
The metrics that actually matter
Once you have the raw data, turn it into a few numbers you can track month over month. That's how you'll know whether you're gaining or losing ground against competitors. You don't need a complicated dashboard. You need consistency.
Mention rate (your share of voice)
Out of all the answers you collected, what percentage named your business? Do the same calculation for each competitor. As a hypothetical example: if you ran 25 prompts across 4 assistants, twice each (200 answers), and your company showed up in 30 of them, your mention rate is 15%. If the competitor across town showed up in 90, theirs is 45%. That gap is the clearest single picture of who is winning AI recommendations in your market.
Citation rate
Mentions and citations aren't the same thing. An assistant can name your company without linking to your site, or cite your site without recommending you. Track how often your website appears as a source. Citations tend to drive actual clicks, and they show that the assistant found your content useful enough to reference.
Position and description
Being the first name in a list of three is different from being an afterthought at the end. Note where you appear and read how the assistant describes you. Does it mention the services you want to be known for, or something you stopped offering years ago? Wrong descriptions usually point to stale information somewhere online.
Cited sources
Keep a running list of the sites that keep showing up as sources: review platforms, directories, the local business journal. That's where the assistants are pulling from. If your competitors are on them and you're not, you've found your to-do list.
Tools that help you track AI visibility
A spreadsheet and a few hours are enough to start. Over time, other tools can fill in the picture. Here are the main categories and what each one is good for:
- Manual tracking spreadsheets: the cheapest and most transparent option. You see the exact answers, which keeps you honest about what's really happening.
- AI visibility trackers: software that runs your prompt list on a schedule across several assistants and reports mention rates against competitors over time. Check how each one samples answers and locations before you trust the trend lines.
- Google Search Console: AI Overview and AI Mode activity is counted inside your normal search performance data. You can't isolate it cleanly, but impressions and clicks on question-style queries give you a signal.
- Referral traffic in your analytics: look for sessions from chatgpt.com, perplexity.ai, gemini.google.com, and copilot.microsoft.com. The volume is often small, but it's real traffic from AI answers.
- Brand search trends: when more people see your name in AI answers, more of them tend to search for it directly. A steady rise in branded searches is an indirect but useful sign.
For a quick first look before you build anything, run your site through our free GEO analyzer. It scores how ready your site is for AI answers and compares you against local competitors, so you know where to focus your manual audit.
Why your competitors keep getting recommended
When a competitor shows up again and again, it's rarely luck. AI assistants tend to favor businesses that are easy to verify and described the same way across the web. These are the signals we see most often:
- Review volume and recency. A company with hundreds of recent reviews that mention specific services (“replaced our heat pump in one day”) gives an assistant a lot to work with. A company with thirty reviews from 2021 doesn't.
- Consistent business information. Same name, address, phone, hours, and service area on Google, Yelp, the BBB, industry directories, and your own site. Conflicting details make you harder to trust and easier to skip.
- Clear service pages. A dedicated page for AC installation, another for heat pump repair, another for ductless systems, each naming the areas you serve. One generic “Services” page rarely gets cited.
- FAQ content that answers real questions. Pages that directly answer “how long does an AC install take” or “what size unit do I need” are exactly the kind of text assistants quote.
- Structured data. Schema markup for local businesses tells machines plainly what you do, where, and for whom.
- Mentions on trusted sites. Local news, chamber of commerce listings, and industry associations add independent confirmation that you do good work.
What to do to close the gap
Go back to your audit and look at the prompts where competitors appear and you don't. Then work through the fixes in roughly this order:
- Fix your business information everywhere first. It's tedious, but everything else stands on it.
- Build a steady review habit. Ask every satisfied customer, every week, and encourage them to mention the specific service. Our guide on getting more Google reviews lays out a simple system.
- Create or rewrite your service and location pages so each one clearly answers who you serve, what you do, and why someone should call you. Add FAQ sections built from your audit prompts.
- Get listed on the sources your audit showed being cited for your competitors.
To go deeper on the content side, read how to get recommended by ChatGPT. And if you run an HVAC business, our HVAC marketing page shows how we approach the trade specifically.
How often to re-measure
AI answers shift as models update and new content gets indexed, so a one-time audit goes stale quickly. Re-running your core prompt list monthly is a sensible rhythm for most local businesses. Keep the prompts identical between runs so the numbers stay comparable. Expect improvements to show up gradually over a few months, not in a week. Seasonal businesses like HVAC should add an extra check before peak season.
FAQ
Can I see exactly what ChatGPT tells customers about my business?
Not their individual conversations, no. You can ask the same kinds of questions yourself, in fresh sessions, and record the answers. Repeated over time, that shows how often and how you're being recommended.
Does ranking well on Google mean I'll show up in AI answers?
It helps, since many of the same signals matter, but it's not a guarantee. AI assistants pull from reviews, directories, and third-party sites as well as your website, so a company with a strong site and weak reviews can rank on Google and still get skipped in AI answers.
How many prompts do I need to track?
Fifteen to thirty is a practical range for a local service business. With fewer, a single odd answer can swing your numbers. With many more, the audit becomes hard to repeat every month, which defeats the purpose.
Start with what you can measure
You don't have to guess whether AI assistants are sending customers to your competitors. A prompt list, a spreadsheet, and an afternoon will tell you. At DataDrivenHQ, a woman-owned agency based in San Diego County, we run these audits for local businesses across the country and turn the gaps into a clear plan. If you'd like a second set of eyes on your results, book a free call and we'll walk through them with you.