How Can HVAC Companies Track AI-Influenced Jobs?

What Does an AI-Influenced Booked HVAC Job Actually Mean?

An AI-influenced booked job is not the same thing as a click from a recognizable AI referrer. A homeowner may ask an assistant which HVAC company handles a failed compressor, remember a business name, search that name later, open a Google Business Profile, and call from the profile. The visible session may look like branded search or direct traffic even though an AI answer influenced the short list. The reverse can happen too. A referral may be visible, but the caller may already know the company from a maintenance agreement or neighbor.

Use three separate questions. Was the business visible in an AI or generative search surface? Did the customer say that surface influenced the choice? Did the lead become a booked service appointment and later a completed job? Each question requires a different record. Search Console can support the first. Intake and call context can support the second. Dispatch or CRM status supports the third. Combining all three into one claim without the joins creates a number that sounds precise but cannot be defended.

The operating definition should be narrow: confirmed AI influence means the customer explicitly named ChatGPT, Gemini, Perplexity, an AI Overview, AI Mode, or another identifiable AI experience as part of discovery or comparison. Assisted AI influence means the record contains supporting context, such as a referral plus a matching customer statement, but not a direct confirmation. Everything else stays unknown or assigned to another documented source. The goal is an auditable classification, not a larger attribution total.

Evidence: ai-visibility-boundarytraffic-source-boundaryfirst-party-funnel

What Should Your Dispatcher Ask on Every New Call?

1. Ask an open discovery question

   Use a neutral prompt such as, 'How did you first hear about us or decide to call today?' Do not list AI first, because leading the caller can inflate the category. Record the answer in the customer's own short words, not only in a forced dropdown.

2. Clarify the named source once

   If the caller says Google, online, search, or a recommendation, ask one follow-up: 'Was that a regular search result, an AI answer, a map listing, an ad, or something else?' If the caller is unsure, select unknown. Do not train dispatchers to convert uncertainty into an AI label.

3. Keep the service need separate

   Record no-cooling, furnace issue, maintenance, indoor-air concern, replacement estimate, or another approved service category in its own field. Discovery source should never replace the operational information needed to schedule the right technician and time window.

4. Confirm the booking outcome

   After the call, record whether the lead was eligible, declined, unreachable, scheduled, canceled, dispatched, completed, or sold according to the company's real workflow. The AI field has no business value if it is detached from this outcome chain.

Evidence: traffic-source-boundaryfirst-party-funnel

Which Fields Belong in the HVAC Lead Record?

Evidence: traffic-source-boundaryfirst-party-funnelprivacy-and-recording-boundary

How Should Website and Referral Data Be Captured?

Start with the standard traffic fields already available in the website analytics implementation: landing page, session source, session medium, referrer when present, and campaign values when they legitimately describe the incoming interaction. Google Analytics documents user-scoped, session-scoped, and event-scoped traffic dimensions as different views of origin and attribution. Choose the scope intentionally. A first-user source is not interchangeable with the source of the session that produced a form submission, and neither proves what a homeowner heard before visiting.

Store the technical observation beside the lead rather than overwriting the customer's answer. If the form session shows a recognizable referral and the customer names the same AI product, the two records reinforce a confirmed grade. If the referrer is present but the customer says a neighbor referred them, keep both facts. The referral may have helped the customer validate the company without being the first influence. If the customer names an AI assistant but the session is direct, do not discard the self-report merely because the browser supplied no referrer.

Avoid inventing UTM parameters for links the HVAC company does not control. Google's documentation explains that manual campaign parameters populate traffic-source dimensions for tagged destinations. That is useful on company-controlled campaigns and profiles, but it does not mean an operator can force external AI products to preserve a tag. The durable design accepts missing technical context and keeps unknown as a valid value.

Evidence: traffic-source-boundary

How Do You Connect a Phone Call to a Booked Appointment?

1. Create the lead at ring or answer

   Have the call platform or dispatcher create a lead record immediately with a call identifier, timestamp, dialed number, and permitted routing context. Do not wait until an appointment is booked, because unbooked inquiries are part of the denominator.

2. Attach the intake answer

   Place the self-reported discovery answer and normalized category on the same lead. If a recording or transcript is used, retain only the permitted evidence needed for review and keep access limited according to policy.

3. Write the appointment ID back

   When dispatch schedules service, store the appointment or work-order ID on the lead. The connection should be deterministic. Matching later by phone number alone can merge household members, repeat callers, or unrelated service requests.

4. Close the operational loop

   Update canceled, dispatched, completed, sold, and lost statuses from the field-service system. Keep booking date, completion date, and any approved revenue field distinct so the report can state exactly which outcome it counts.

Evidence: first-party-funnelprivacy-and-recording-boundary

Which Booking Stages Should the CRM Record?

A homeowner submits a form or reaches the call team

Create an inquiry record. This is the broad lead stage, not a booked job.

The request fits the service area, service type, and scheduling rules

Mark the lead qualified and record the qualification reason.

The homeowner accepts a date or arrival window

Mark the appointment booked and attach its work-order or appointment ID.

A technician is assigned or travels to the property

Mark dispatched according to the field-service system's real definition.

The service visit is closed with an approved completion status

Mark completed. Do not assume completed means replacement sold or invoice collected.

The company records a final customer conversion or sale

Use the company's sold or converted stage and keep the booked-job count available separately.

Evidence: first-party-funnel

How Should AI Influence Be Graded?

Evidence: traffic-source-boundaryfirst-party-funnel

What Does a Practical Attribution Workflow Look Like?

1. Observe visibility

   Review Search Console generative AI visibility where available, relevant landing pages, and dated prompt observations. This identifies exposure and coverage questions, not booked jobs.

2. Capture inquiry context

   Create the lead ID, retain available session or call context, and ask the neutral discovery question at first contact.

3. Grade influence

   Apply the confirmed, assisted, contradicted, or unknown standard with a short reason. Never let a model or vendor silently upgrade weak evidence.

4. Join the booking

   Connect the lead to the appointment or work order and record the actual booking status from the dispatch system.

5. Join the completion

   Add dispatched, completed, sold, lost, and approved value outcomes without treating one stage as another.

6. Audit the report

   Sample records, inspect contradictions, track unknown rates, and publish only language supported by the evidence class being counted.

Evidence: ai-visibility-boundarytraffic-source-boundaryfirst-party-funnel

How Should Direct and Unknown Traffic Be Handled?

Direct traffic is a technical classification for a visit without a clear referral source. Google lists several reasons source information can be missing, including untagged links, redirects, offline documents, direct entry, and tracking interference. It should not be renamed 'AI traffic' simply because AI assistants sometimes produce journeys without a visible referrer. That shortcut would turn every unexplained session into the answer the marketer wanted.

Keep two unknowns if necessary. Technical source unknown means the browser or analytics record does not identify an origin. Discovery source unknown means the customer did not provide a usable answer. A lead can be technical-unknown and customer-confirmed AI. It can also have a visible referral while the customer gives no discovery answer. These combinations are useful because they show which part of the measurement chain needs work.

Report the unknown share beside confirmed and assisted counts. If unknown declines after dispatcher training or form changes, that is a process-quality improvement, not proof that AI influence increased. If direct traffic rises, investigate tagging, redirects, privacy settings, and intake completion before assigning a strategic story. The defensible report preserves uncertainty instead of hiding it.

Evidence: traffic-source-boundary

Can Google Business Profile Calls Count as AI-Influenced Jobs?

Google Business Profile performance can show call-button clicks and website-link clicks associated with the profile. Google defines the calls metric as the number of times a customer clicked the call button. A click is not the same as an answered call. An answered call is not the same as a qualified request. A qualified request is not the same as a booked appointment. The profile metric belongs at the interaction layer, not the booked-job layer.

A profile call can still become part of a confirmed AI-influence record. The call platform or dispatcher needs to create a lead, the customer needs to identify AI influence clearly, and the dispatch system needs to record the booking. The report can then say the customer called through the Business Profile and reported AI influence. It should not say the Business Profile or AI caused the job unless a valid causal method supports that stronger claim.

Use profile interactions as a reconciliation check. A surge in call-button clicks with no rise in created call leads could expose a routing, tracking, or answer-rate problem. A rise in booked jobs with flat profile calls may point to forms, direct dialing, branded search, repeat customers, or other paths. The value comes from comparing layers without pretending they are identical.

Evidence: profile-action-boundaryfirst-party-funnel

What Should a Weekly HVAC Attribution Report Show?

Visibility layer

Show available generative AI impressions, relevant page trends, and dated prompt observations with an explicit note that these are not lead or booking counts.

Intake layer

Show total inquiries, discovery-question completion, confirmed AI, assisted AI, contradicted, and unknown classifications.

Booking layer

Show booked appointments by influence grade, plus booking rate within each grade when the denominator is complete.

Operations layer

Show dispatched, canceled, completed, sold, and lost outcomes according to documented field-service definitions.

Quality layer

Show unlinked calls, duplicate leads, missing work-order links, classification changes, and sampled records that failed review.

Privacy layer

Show retention exceptions, unauthorized access events, recording-consent gaps, and vendor issues to the designated compliance owner, not the marketing leaderboard.

Evidence: ai-visibility-boundaryfirst-party-funnelprivacy-and-recording-boundary

How Should Prompt Checks Be Used Without Overstating Causation?

Prompt checks answer whether a selected model produced a selected response for a selected query, location context, account state, and moment. They can reveal whether the HVAC company appears, which pages or third parties are cited, and whether important service facts are missing or wrong. They do not identify the total population of homeowner prompts, and they do not prove that a response produced a booking.

Store prompt observations as a separate table with the exact question, tool, date, location assumptions, signed-in state when known, output summary, cited URLs, and reviewer. Use a fixed panel for trend comparison, but rotate a smaller exploration set to catch new service questions. Do not merge the number of appearances into the CRM as if each appearance were a lead. The prompt table supports diagnosis and content planning.

Compare prompt observations with first-party outcomes carefully. If confirmed AI-influenced bookings repeatedly mention heat-pump financing and prompt checks show weak or inaccurate coverage of that topic, the combined evidence can justify a content or entity correction. It still does not prove incremental lift. Make the operational decision, label the inference, and measure the next period with the same definitions.

Evidence: ai-visibility-boundary

What Should an HVAC Company Implement First?

• Write one definition each for inquiry, qualified lead, booked appointment, dispatched call, completed job, sold job, confirmed AI influence, assisted AI influence, and unknown source.
• Add a neutral discovery question and one permitted follow-up to every phone and form intake path.
• Create a stable lead ID at first contact and carry it into the appointment or work-order record.
• Store customer-reported discovery separately from session source, referrer, campaign values, and call-routing context.
• Use confirmed, assisted, contradicted, and unknown grades with a short reviewer note.
• Keep Google generative AI visibility, prompt observations, profile interactions, inquiries, bookings, completions, and sales in distinct report layers.
• Sample records every reporting cycle and correct classification rules when the evidence does not support them.
• Review call notice, consent, transcript access, retention, vendor contracts, customer disclosures, contractor-advertising language, and applicable privacy requirements before production use.
• Report unknowns and data gaps openly. A smaller defensible number is more useful than a larger invented one.