The 30-Day ChatGPT Ads Pilot We Would Run for an HVAC Company

Most “how to test a new channel” advice is too broad to be useful.

Run multiple creatives.

Track conversions.

Optimize.

All true. None of it tells an HVAC owner what the first month should actually look like.

Here is the 30-day pilot we would run for an established local HVAC company with available replacement capacity and a functioning intake process.

This is a blueprint, not a performance forecast.

Bottom line

One service. One market. One landing page. One commercial outcome. Roughly $2,000 in controlled media is enough to test the mechanics and begin evaluating lead quality, although local inventory and CPC determine how much data appears. The pilot succeeds only if it creates a defensible path to sold work.

The hypothetical company

Assume the company:

  • serves one metro
  • has 10–20 technicians
  • uses ServiceTitan or a comparable operating system
  • can take more replacement estimates
  • has a functioning call center
  • knows its average replacement economics
  • is already running Google Ads

We would not start with “HVAC.”

We would start with:

aging-furnace repair versus replacement

Why?

  • clear homeowner problem
  • meaningful job value
  • natural conversational intent
  • straightforward landing-page story
  • measurable appointment and sales path

Before day one: set the economics

The campaign should not launch until the company answers five questions.

1. What job do we want?

Furnace replacement.

2. Where do we want it?

A defined set of cities or ZIP codes with available capacity and attractive job economics.

3. What is the maximum acceptable acquisition cost?

Use actual gross profit and close-rate data.

Do not choose a target because an agency says “HVAC leads should cost X.”

For a framework on how to set your acquisition-cost threshold, see our cost guide.

4. Who owns the lead?

Name the intake manager.

5. How will the source survive?

Create:

  • ChatGPT Ads campaign source
  • dedicated tracking number
  • form-source fields
  • UTM convention
  • oppref preservation
  • CRM or ServiceTitan campaign record

If those steps are incomplete, the campaign is not ready.

Campaign structure

Objective

For the initial validation phase, use Clicks.

OpenAI currently recommends a starting maximum CPC bid of $3–$5. We would begin at $4 and adjust based on delivery and lead quality. See current pricing guidance.

The objective is not permanently locked to clicks.

It is a starting point while measurement is validated and enough conversion signal develops.

Budget

Use a $2,000 campaign-total budget over 30 days.

OpenAI recommends daily budgets for advertisers new to the platform, but a campaign-total budget is the stricter total-spend control. Daily spend can reach twice the selected daily amount on an individual day. See budget mechanics.

For a controlled client pilot, we prefer the hard ceiling.

Geography

Use only the cities, markets, or postal codes where supported and where the company actually wants jobs.

Context hints do not enforce geography.

Use campaign controls.

Ad groups

Use two.

Ad group 1: Repair versus replacement

Customer situation:

Aging furnace, repeated repairs, uncertainty about whether another repair is rational.

Ad group 2: Furnace replacement estimate

Customer situation:

Homeowner has already concluded replacement is likely and wants a local company to evaluate options.

Those are related but not identical.

They deserve distinct creative.

Context hints

Examples:

Repair versus replacement

Homeowners in the Denver metro area with a furnace over 15 years old who are comparing another repair with full-system replacement.

Replacement estimate

Homeowners in the Denver metro area planning furnace replacement and looking for a licensed local HVAC company to inspect the system and provide options.

These are relevance descriptions.

They are not guaranteed prompt targeting.

Read OpenAI’s context-hint guidance.

Creative plan

Launch at least five distinct ads across the two ad groups.

Angle 1: Decision support

Repair or replace your furnace? Get a local inspection and compare both options before deciding.

Angle 2: Aging equipment

Is your furnace nearing the end of its life? Talk with a licensed local HVAC company about replacement options.

Angle 3: Repeat breakdowns

Tired of repairing the same furnace? See whether replacement makes more financial sense.

Angle 4: Financing

Use only if the company actually offers clear financing.

Angle 5: Warranty and installation quality

Use only with specific, supportable terms.

The ads should not all say the same thing with different verbs.

OpenAI recommends distinct variations and benefit-focused copy. See the creative guide.

The landing page

The page headline:

Repair or replace your furnace?

The page should include:

  • signs replacement may be worth evaluating
  • reasons another repair may still be reasonable
  • service area
  • license and relevant credentials
  • actual warranty terms
  • financing, if available
  • what happens during the estimate
  • phone and short form
  • evidence the business can substantiate

It should not include:

  • every HVAC service
  • generic “comfort you can trust” copy
  • fake urgency
  • unsupported savings claims
  • a 12-field form

The page exists to continue the conversation.

Measurement setup

OpenAI layer:

  • Pixel installed
  • lead event tested
  • oppref preserved
  • UTMs added
  • optional Conversions API prepared
  • duplicate events prevented with event IDs

Operating layer:

  • dedicated tracking number
  • ChatGPT campaign source
  • lead quality disposition
  • appointment status
  • estimate completed
  • job sold
  • revenue

ServiceTitan can assign trackable phone numbers to campaigns and connect booked jobs to revenue. See ServiceTitan’s marketing documentation.

Week one: validate the plumbing

The first week is not about clever optimization.

Confirm:

  • ads passed review
  • campaign is serving
  • geography is correct
  • links work
  • landing page loads quickly
  • calls route properly
  • forms arrive
  • source enters the CRM
  • OpenAI events fire once
  • office staff know the campaign exists

Review every inquiry manually.

A bad event implementation can train the campaign toward the wrong action.

Week two: judge relevance

Now ask:

  • Are the conversations and clicks aligned with furnace decisions?
  • Are leads inside the service area?
  • Are they new customers?
  • Are they asking about the promoted service?
  • Is the page producing calls or forms?
  • Is one ad angle clearly stronger?
  • Is response time acceptable?

Pause ads that are clearly irrelevant.

Do not change the bid, geography, landing page, and creative at the same time.

Preserve the ability to learn.

Diagnose the bottleneck.

Low delivery

Possible causes:

  • bid
  • narrow inventory
  • account or ad review
  • overly constrained geography
  • weak creative coverage

Clicks but no leads

Possible causes:

  • low relevance
  • weak landing page
  • misleading ad promise
  • tracking failure
  • slow page
  • poor CTA

Leads but few bookings

Possible causes:

  • out-of-area traffic
  • wrong service
  • intake failure
  • slow response
  • customer expectation mismatch

Bookings but no sales

Possible causes:

  • lead quality
  • sales execution
  • pricing
  • appointment quality
  • capacity or scheduling

The ad platform is not always the problem.

Day 30: make a decision

Evaluate:

  • total spend
  • impressions and clicks
  • unique inquiries
  • qualified leads
  • booked appointments
  • completed appointments
  • sold jobs
  • booked or completed revenue
  • all-in acquisition cost
  • unresolved opportunities still in the pipeline

Then choose one of four outcomes.

Scale

Lead quality and economics justify more budget or geography.

Continue the same test

The signal is promising but the buying cycle is incomplete.

Change the proposition

The campaign delivered activity, but the promoted service or landing page is wrong.

Stop

The inventory, relevance, or economics do not justify further spend.

“New channel” is not a reason to continue a weak campaign indefinitely.

What this pilot should produce even before scale

At the end of 30 days, the company should know more than whether it “got leads.”

It should have:

  • a working ChatGPT Ads account
  • proven measurement
  • service-specific creative
  • a reusable landing-page framework
  • real local CPC and delivery data
  • lead-quality evidence
  • a clearer view of whether the channel deserves more investment

That operating knowledge is the early-mover advantage.

For the full picture of how ChatGPT Ads work for home services, read the operator’s guide.

Sources

Ready to test ChatGPT Ads in your market?

ChatDemand handles everything — from ad to booked job.

Start Testing ChatGPT Ads →