How Can Roofers Track AI-Influenced Inspections?

What Counts as an AI-Influenced Roofing Inspection Request?

An AI-influenced roofing inspection request is a first-party inquiry in which the homeowner says an AI assistant or generative search experience helped surface, compare, or validate the company. It is not simply a visit from organic search, a Business Profile call-button click, a spike after a hailstorm, or a prompt-monitoring result that names the roofer. Those observations can support the analysis, but none of them contains the homeowner's decision by itself.

Roofing creates a difficult path because discovery often moves across surfaces. A homeowner may ask an assistant what to do after wind damage, see a company name, search the name later, read reviews, open the Business Profile, and call from the profile. The visible source may be Google or direct even though an AI answer shaped the shortlist. Another homeowner may land through a recognizable referral but choose the company because a neighbor recommended it. The technical path and the human reason are different fields.

Use a narrow classification. Confirm AI influence when the homeowner identifies ChatGPT, Gemini, Perplexity, Google AI Overviews, AI Mode, or another identifiable assistant as part of discovery or evaluation. Use assisted when the customer statement and independent context point in the same direction without direct confirmation. Use contradicted when the homeowner names another source. Keep every incomplete record unknown. Honest unknowns make the confirmed group more useful.

Evidence: ai-visibility-boundarytraffic-source-boundaryinspection-funnel

Which Roofing Inspection Stages Should Stay Separate?

Evidence: inspection-funnelprivacy-contractor-boundary

What Should the Roofing Intake Team Ask?

1. Start with an open source question

   Ask, 'How did you first hear about us or decide to contact us today?' Let the homeowner answer before presenting categories. This reduces the chance that an intake script plants the AI answer it is trying to measure.

2. Clarify ambiguous online answers once

   If the answer is Google, online, search, or a recommendation, ask whether it was a regular result, map listing, ad, AI answer, another person, or something else. Accept 'not sure' without changing it to AI.

3. Capture the remembered decision clue

   Ask what the homeowner remembers seeing or asking. A short paraphrase such as 'roofer for wind-lifted shingles near me' is enough. Do not ask for private chat histories or screenshots as a default requirement.

4. Record inspection intent separately

   Use a distinct field for leak check, visible shingle damage, maintenance review, replacement estimate, storm follow-up, or another company-approved category. Discovery source should never substitute for the information needed to route an inspection safely.

5. Close with the actual disposition

   Record out of territory, not qualified, follow-up needed, requested, scheduled, canceled, no access, completed, or another approved outcome. The influence label matters only when it stays attached to this operational record.

Evidence: traffic-source-boundaryinspection-funnelprivacy-contractor-boundary

Which Fields Belong in the Roofing Lead Record?

Evidence: traffic-source-boundaryinspection-funnelprivacy-contractor-boundary

How Should Call and Website Tracking Support the Record?

Technical tracking should corroborate the lead record, not rule it. A dedicated call-tracking number can identify the page or campaign that displayed the number. A form can preserve its landing page and tagged campaign values. GA4 can describe first-user, session, and event-level acquisition context. Each system answers a bounded question about the observable route. None knows every conversation, prompt, review, referral, or offline comparison that happened before contact.

Keep original and normalized values. The original record might say session source 'google' and the normalized field might say organic search. A customer answer might say 'Gemini showed your company after I asked about a roof leak.' Store both without forcing them to agree. When they conflict, the conflict is useful. It may reveal a multi-touch journey, a redirect that removed context, a shared device, or a mistaken recollection. It is not permission to choose the category that makes the report look stronger.

Direct traffic deserves restraint. Google's own documentation defines direct traffic as lacking a clear referral source and lists several reasons that source information can be missing. A direct visit after a storm is not automatically an AI visit. Keep it direct in the technical field. If the homeowner separately confirms AI influence, the record can hold both facts at once.

Evidence: traffic-source-boundaryprivacy-contractor-boundary

How Should Storm Activity Appear in the Attribution Report?

Storm activity belongs in a context layer. NOAA's Storm Events Database documents significant weather events using National Weather Service records. A roofing company can use an event date, event type, and geography to explain why overall inspection demand changed. That can help operators compare normal periods with storm-response periods and plan staffing. It cannot tell the company which lead came from an AI answer.

Create a separate storm-context field or reporting band. Examples include no matched event in currently available NOAA data, recorded event in the broader market, or recorded event in the service area during the defined review window. NOAA's current dataset metadata says monthly updates can have up to a 120-day delay and cautions that contributed information may remain unverified by NWS. A recent no-match cannot prove that no storm occurred. Record the database access date, reconcile later releases, and describe the join exactly.

Weather proximity also does not establish roof damage, insurance coverage, claim eligibility, or repair need. The inspection workflow determines what the company observed under its approved procedures. Insurers, adjusters, contractors, and legal professionals have different roles that vary by jurisdiction and contract. The attribution report should stop before making those determinations.

Evidence: storm-context-boundaryprivacy-contractor-boundary

Which System Should Answer Each Measurement Question?

Evidence: ai-visibility-boundarytraffic-source-boundaryprofile-action-boundarystorm-context-boundaryinspection-funnel

How Should Each Inspection Request Be Graded?

The homeowner identifies an AI assistant or generative search feature and the lead record reaches inspection requested.

Grade confirmed AI influence. Preserve the customer's wording and report the later inspection statuses separately.

The homeowner gives an ambiguous online answer while a recognizable referral or matching journey context exists.

Grade assisted only. State which independent clue exists and do not count the record as confirmed.

The technical route suggests AI, but the homeowner names a referral, prior relationship, ad, or another source.

Grade contradicted and preserve both records. Investigate the path without overriding the customer statement.

The source answer is missing, the caller is unsure, or the records cannot be joined safely.

Grade unknown. Improve the next intake interaction instead of guessing at historical attribution.

A storm record overlaps the inquiry date but the homeowner gives no AI discovery evidence.

Keep storm context separate and leave AI influence unknown or assigned to the actual documented source.

Evidence: traffic-source-boundaryinspection-funnelstorm-context-boundary

How Do You Join the Inquiry to the Completed Inspection?

1. Create the lead once

   Assign a durable lead ID at the first call, form, message, or booking request. Store the original entry time and source context before later systems normalize it.

2. Qualify the request

   Record whether the property and service need fit the company's territory and inspection policy. Keep qualification separate from the homeowner's discovery answer.

3. Attach the schedule event

   Write the accepted appointment date or service window back to the same lead ID. A requested inspection without an accepted time stays requested, not scheduled.

4. Record access and completion

   Use the company's approved operational statuses for access confirmed, no access, canceled, rescheduled, or completed. Do not infer completion from a calendar slot.

5. Freeze the attribution grade

   After the defined reconciliation window, preserve the grade and reason used for reporting. Later sales outcomes may update separately without rewriting the original source evidence.

Evidence: inspection-funneltraffic-source-boundaryprivacy-contractor-boundary

Where Do AI Prompt Checks Fit in the Measurement Plan?

Prompt checks answer whether a named system produced a named business under a documented prompt, location, account state, and date. They can reveal visibility gaps, inconsistent recommendations, and the kinds of source pages an answer cites. They do not identify the people who later contacted the company. A prompt result and an inspection request can occur in the same week without one causing the other.

Store prompt observations in a separate table with the exact prompt, market, platform, date, response state, business-name presence, cited source when visible, and reviewer. Do not paste the result into the customer record as if it were the customer's journey. At the aggregate level, compare prompt visibility, Search Console AI impressions, total inquiries, confirmed AI-influenced requests, and completed inspections. Label the comparison descriptive unless a stronger study design supports more.

A useful prompt panel stays stable long enough to show change. Rewriting prompts after every inspection produces an attractive story and a weak measurement system. Keep a fixed core panel, document any added storm-response prompts as a separate cohort, and avoid claiming that a recommendation caused a request merely because both moved together.

Evidence: ai-visibility-boundarytraffic-source-boundarystorm-context-boundary

What Should the Weekly Roofing Attribution Report Show?

Evidence: ai-visibility-boundaryinspection-funnelstorm-context-boundaryprivacy-contractor-boundary

Which Rates Can the Roofing Company Calculate Honestly?

Start with intake coverage: leads with a recorded discovery answer divided by all inspection inquiries in the same period. Then calculate confirmed AI influence among all inquiries, scheduling among confirmed AI-influenced requests, and completion among confirmed AI-influenced scheduled inspections. Publish the numerator, denominator, date range, and influence rule beside every rate. A percentage without those four items cannot be audited.

Do not remove contradicted or unknown records from the all-inquiry denominator to make AI share look larger. Do not combine requested and scheduled inspections. Do not call an inspection completed because a technician was dispatched. Do not call a completed inspection revenue, a sold job, or an insurance outcome. Each step has a different operational owner and a different failure mode. The report is stronger when the drop-offs stay visible.

Use minimum sample rules defined by the company before reviewing results. With a small count, report the records and classifications without implying a stable trend. Compare periods only when intake training, influence rules, service territory, and storm context are sufficiently consistent. If any of those changed, annotate the break rather than hiding it inside a percentage.

Evidence: inspection-funneltraffic-source-boundarystorm-context-boundary

What Should the Weekly Quality Review Check?

• Every inspection inquiry has one durable lead ID and a creation timestamp.
• The discovery answer is present, or the source is explicitly unknown rather than backfilled.
• Confirmed AI influence includes a named assistant or generative search experience in the homeowner's own account.
• Technical source, customer-reported influence, service intent, and storm context remain separate fields.
• Inspection requested, scheduled, access confirmed, completed, canceled, and no-access outcomes are not collapsed.
• Business Profile call-button clicks are not reported as connected calls or inspection requests.
• Direct traffic is not reassigned to AI without separate customer confirmation.
• NOAA storm context records the database access date, reporting lag, geographic scope, and later reconciliation without becoming customer-level attribution proof.
• Unknown and contradicted records remain visible in the denominator.
• Personal information, vendor access, retention, recording, licensing, and insurance boundaries follow the company's approved policies.

How Can a Roofing Company Implement This in Thirty Days?

1. Week one: define the states

   Write one operational definition for inquiry, inspection requested, qualified, scheduled, no access, completed, and later commercial outcomes. Assign an owner to each status transition.

2. Week two: add neutral intake

   Add the open discovery question, one clarification, the original answer, normalized source, and attribution grade. Train intake staff with examples that include unknown and contradicted records.

3. Week three: join approved systems

   Connect lead IDs across the phone, form, scheduler, and CRM without collecting extra personal or claim data. Test direct, organic, profile, tagged, and missing-source cases.

4. Week four: run the first reconciliation

   Review every classification, preserve disagreements, calculate rates with visible denominators, and add storm context only at the defined aggregate level. Fix process gaps before interpreting performance.

5. Day thirty: freeze version one

   Document the rules, reporting window, vendor access, retention policy, and review owner. Future changes receive a version date so period comparisons remain interpretable.

Evidence: traffic-source-boundaryinspection-funnelstorm-context-boundaryprivacy-contractor-boundary