Measure AI recommendations for a home services company by asking the same set of real homeowner questions, by service and town, on a schedule, and recording which companies each assistant names and what it says. Then connect what you can to calls and booked jobs, knowing that phone-first leads leave gaps analytics will not fill. Keep mentions, recommendations, calls, and booked jobs as separate numbers.
Owners ask a fair question: if homeowners are asking ChatGPT or Google's AI who to call, is our company coming up? Answering it for a trade company has quirks. The questions are local, often urgent, and tied to a specific service, and the result is usually a phone call rather than a form.
This page applies the general approach in our article on measuring AI search results to the trades. It is a measurement guide for owners and marketing leads, not a promise of any result. Our home services overview covers the content work being measured.
Build a prompt set from real homeowner questions
Start with questions homeowners actually ask, drawn from your call logs and from the questions homeowners ask before booking. Write them the way a homeowner would type or say them, and include the conditions that change the answer: the service, the town, the urgency, and the kind of home or system.
| Prompt type | Example prompt (fictional) | What it tests |
|---|---|---|
| Service and town | Who can replace a water heater in [town]? | Unprompted recommendation for a core service |
| Urgent | My furnace stopped working tonight in [town]. Who can come out? | Visibility for after-hours work |
| Comparison | Is [company] or [competitor] better for a roof replacement? | How the company is described next to others |
| Fit | Does anyone near [town] work on boilers in older homes? | Whether specialty work is known |
| Branded | What do people say about [company]? | What the assistant says when asked by name |
Keep branded questions in their own group. An assistant describing your company when asked by name is not the same as recommending it to someone who asked who to call, and blending the two inflates the result.
Record each answer with its conditions
For each captured answer, keep the full prompt, the date, the assistant and mode, the location context, the companies named in order, what was said about each, and any linked sources. Location deserves extra care for trades, since local answers may differ depending on where the person appears to be. Save the original answer, not just a score.
Count carefully. Report how many answers named your company out of how many were checked, by service and town, and treat that figure as a description of your sample, not of every homeowner's experience. Before the first round, ask whoever runs the check, "If someone else repeated this next quarter, would they ask the same questions from the same places and record them the same way?" If not, fix the method before comparing anything.
Read Google's reports for what they cover
Search Console's Generative AI performance report covers "impressions for the following generative AI capabilities on Google Search": AI Overviews and AI Mode. Google defines those impressions as "how many times links to your site were shown to a user in a generative AI feature on Google Search," and says a missing report usually means the site "hasn't received enough impressions." An impression in that report is a link shown, not a call.
In Google Analytics, the AI Assistant channel covers visits "from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok," and it "excludes Google's AI Overviews and AI Mode," which count under Organic Search. OpenAI says "ChatGPT automatically includes the UTM parameter utm_source=chatgpt.com in referral URLs." Check what your own analytics receives before building a report on it.
The Business Profile adds the phone side. Google defines its calls metric as "the number of times a customer clicked on the call button on your Business Profile." That counts taps, not conversations, and it says nothing about where the homeowner first heard your name. Our page on what a home services Business Profile should say covers the profile itself.
Connect calls and booked jobs honestly
For many trade companies, the lead arrives as a phone call, and that is where measurement gets hard. A homeowner may ask an assistant, see a name, and then dial a number from a truck, a yard sign, or the Business Profile. Analytics will not connect those steps. Call tracking numbers on the website can show which pages produced calls, and a short question in the call script, such as "How did you hear about us?", adds the homeowner's own account. Record that answer in the homeowner's words in your booking or dispatch software and keep it separate from tracked sources.
Then follow the call to the job. Agree with the office on what counts as a qualified call: right service, right area, real request. Report AI observations, website visits, qualified calls, and booked jobs side by side, with the method used to connect them, and leave unknown sources unknown.
Watch for double counting. One homeowner might tap the call button on the profile, hang up, call back on the website number, and then book through a text thread. That is one opportunity, not three. Keep one record per homeowner request in the booking software, and set aside duplicates, wrong numbers, existing customers calling about a scheduled job, and requests outside your service area before reporting new business.
A fictional quarterly check
Here is how a fictional roofing company in Texas might run it. Each quarter, the marketing lead asks the same set of prompts across two assistants from three towns the company serves, records every answer in a shared sheet, and compares the companies named with last quarter. She adds Search Console's generative AI report, the AI Assistant channel in Analytics, Business Profile calls, and the office's tally of callers who said an assistant mentioned the company. The report ends with one decision, such as which service page needs a clearer answer, rather than a single score.
The prompts have to reflect the business. Whoever builds the prompt set should not have to invent which services and towns matter; the owner sets those, because they describe the work the company wants.
Measurement is not one of Hudson South's four steps, but it shows where the work behind them needs attention: the Authority Asset (approved business knowledge and evidence), Hudson Twenty-One (how the company communicates), Market Intelligence (the buyer questions and the market), and AEO Creation (the answer-first pages).
Start with a competitor comparison
On a free 30-minute call, Hudson South walks through how competitors in your market answer the homeowner questions you would want to be named for.
Get your free competitor analysisSources
- Generative AI performance report (Search). Search Console Help. Read 2026-10-15.
- Default channel group. Google Analytics Help. Read 2026-10-15.
- Publishers and Developers FAQ. OpenAI Help Center. Read 2026-10-15.
- Understand your Business Profile performance & insights. Google Business Profile Help. Read 2026-10-15.