How many roofing inquiries come from AI? How to count your own

Is there a published figure for roofers?

Not one we know of that says where it came from, and our own work does not produce one. Our audit of 9,483 small business websites measured what the sites serve to a crawler, not where their calls come from. Its probe measured something narrower still: whether a business was named when one buyer question went to Claude with web retrieval on. 14.9% of 9,240 businesses were, across all 40 trades, on one question and one day. A name in an answer is not an inquiry, and no roofing-only figure was published.

So if you meet a share of roofing leads said to come "from ChatGPT", ask what was counted: calls, forms, visits to a site, or answers that named somebody. Those are four different numbers, and a count of one says nothing reliable about the others. The number worth having is the one you build from your own intake, and it takes one question and a sheet.

The question your office asks every new caller

Ask every new inquiry the same question before the inspection is booked: how did you find us? Ask it on the phone, put it on the web form, and ask it at the door if first contact happens there. Offer the same choices each time, with an AI assistant among them, and write the caller's own words beside the box they pick.

How did you find us?   (one choice; write their own words beside it)
  [ ] A neighbor, friend or family member
  [ ] Google search
  [ ] Google Maps
  [ ] An AI assistant (ChatGPT, Gemini, Claude, Perplexity, Copilot or another)
  [ ] A review site or directory: [which one]
  [ ] Our truck, a yard sign or a job we were doing nearby
  [ ] We have worked for you before
  [ ] Not sure
  [ ] Other: [their words]

If "An AI assistant": which one, and what did you ask it?   (optional)

Three rules keep the answers usable. Read the choices in the same order every time, or show them in the same order on the form, and suggest none of them. Keep "not sure" on the list, so a guess has somewhere to go instead of landing on Google by default. And record the answer given at first contact as the source of the job; if a later conversation suggests another, write it in a second field and leave the first alone.

The optional follow-up earns its place twice. "Which one, and what did you ask?" tells you which assistant your callers use, and the words they report are the raw material for the lines of your monthly panel, which AI visibility for roofers sets out. Keep only what the job record already needs; the question is about where the caller heard of you, not about the caller.

Carrying the answer from the first call to the signed contract

A roofing job can take weeks from the first call to a signed contract, and a replacement quoted in spring may be signed in summer. So the source is written once, on the job record, and carried through every stage: inquiry, inspection, estimate sent, contract signed. Count each stage under the month of first contact, and fill in the later columns as the jobs move.

Judge the channel by the signed column, not the inquiry column. A roof replacement is a large job, so three contracts signed from assistant inquiries in a quarter can matter more to the business than thirty inquiries that never booked an inspection. The ratios between the columns show how far assistant callers get compared with everyone else, a part of the customer journey that no answer records.

Mark storm months in the sheet. After hail or high wind, inquiries can rise from every source at once, and the share that names an assistant in a storm month tells you little about a quiet one. Compare storm months with storm months and quiet months with quiet months, the same separation the panel's storm log keeps.

Call tracking, and why one number serves a roofer better

Call tracking that gives each page, ad or source its own phone number answers the attribution question by giving up the one fact every page should share. A site that shows an office number, a tracking number and an old campaign's number gives an answer three to repeat, and a homeowner comparing your site with your Business Profile cannot tell which one is the business. In our audit, 36.9% of the sites published more than one phone number across their own homepage, contact page and about page; that is all 40 trades together, with no roofing cut. It is why one number heads the list of fixes in the overview for roofing contractors.

Track by page instead, and keep the number the same everywhere:

1. Put one number in the text of every page, in the footer and in the markup, the same number your Business Profile and listings carry.
2. Give every form a hidden field that records the address of the page it was sent from, so a form from the metal roofing page arrives labeled as one.
3. Tag the tap-to-call link on each page as an event in your analytics, named for that page. The number stays the same; the event says where the call started.
4. Ask the intake question on every call anyway. A caller who dialed from an answer, or from memory, started on no page of yours at all.
5. If a paid ad has to carry a tracking number, keep that number on the ad and out of the text and markup of your pages.

You give up knowing exactly which ad a call came from, and you keep one number that every page, listing and answer can agree on. After a storm, when a homeowner may be checking whether a contractor is local, that agreement is worth more than knowing which ad rang.

The two counts analytics and Search Console can add

A visit from an assistant's answer reaches Google Analytics as a referral when the browser passes on where the visitor came from; spotting AI traffic in Google Analytics covers the hosts and how to group them. For a roofer that count is a floor. A homeowner who reads your number in an answer and calls it never becomes a visit, which is zero-click search in the form this trade meets it.

For Google's own AI features there is a separate count. Search Console's generative AI performance report counts impressions, the times a link to your site was shown in AI Overviews or AI Mode, and groups them by page, country, date and device. It has no query dimension, and a site with too few impressions may not see it. The clicks from those features stay inside the ordinary Performance report under the Web search type, with the rest of your organic traffic, as Google's page on AI features says.

The page dimension is the useful one for a roofer. It shows which pages Google's AI features linked in a given month, the storm page or the slate repair page, and whether that moved after you rewrote one. It cannot tell you who called, what they asked or whether they signed.

Reading the tally beside the monthly panel

The tally says what callers told you. The panel says what the answers say, as a visibility score measured the way our methodology describes. Neither proves the other. A rising tally beside a flat panel can mean more homeowners are asking assistants, or that the office has started asking the question every time. A rising panel beside a flat tally can mean homeowners read your name and called without mentioning where they saw it.

So set the two side by side each month, with a third column for what you changed on the site and when. When both move in the weeks after a change you made, two independent readings point the same way, which is as close to proof as a roofer's own numbers get. When they disagree, read the callers' own words in the intake sheet before anything else; they are the only part of either reading that comes straight from the people who called.