How to Use ChatGPT to Find Customers: A Contractor's Prompt Playbook

In this illustration, you ask ChatGPT for “warehouses with old roofs in Dallas” and get 20 tidy rows of buildings and roof ages, each with a facilities contact. The table gives your rep a full morning of checking because none of the roof ages has a supporting record and none of the contacts has a source date.

A warehouse address does not establish when its roof was replaced. A facilities title does not establish who can approve your work. An email built around either unsupported claim starts the conversation with a mistake.

I would give ChatGPT a narrower job: prepare the questions and write from checked facts. Ground Control by PorchRocket supplies the building list. The prompts below show how to use that file for calls and follow-up without asking a writing tool to supply missing property evidence.

Where ChatGPT earns its place in prospecting

ChatGPT can help a contractor prepare sales conversations and draft messages when the prompt includes the property facts behind the offer.

Start with this morning’s sales work. A rep who needs to identify the person responsible for rooftop service should ask for a routing question, while a rep whose manager requested vendor information needs an email built from that conversation’s notes.

The official prompting guidance recommends explaining the desired result and supplying relevant context, with instructions about the format or boundaries when they matter to the task. Give the tool your approved service scope.

Sales task	Input to provide	Output to request	Owner’s review
Choose a customer profile	Service scope, travel limit, minimum account value	Qualification criteria and exclusions	Can the crew serve this account profitably?
Prepare a first call	Checked property facts and the offered service	Opening and routing question	Does the wording claim anything unverified?
Write after a conversation	Rep’s notes and the agreed next step	Short email draft	Does it reflect what the prospect requested?
Plan next week’s work	Contact outcomes and scheduled dates	Queue by property ID	Are opt-outs and promised dates respected?

I would build one approved prompt for each task and keep the accepted version beside the sales procedure. Reps can change the property facts while preserving the service offer and review standard.

The broader commercial prospecting method puts the building first. The guide to AI lead generation for contractors explains where writing tools fit beside property research. Pick the task before picking another tool.

Can ChatGPT find leads with verified building details?

ChatGPT can research public sources and inspect supplied images, but a claim about a particular building needs evidence before your rep uses it.

ChatGPT’s web search documentation describes searching for current information and returning citations; availability can depend on workspace settings. Use that capability for a manager’s published service area or a property’s contact page. Open the cited page and confirm the statement before adding it to a record.

Image analysis also exists. The official vision documentation warns that object counts can be approximate and descriptions can be wrong, which matters when your rep plans to reference a specific number of units in an email. Review the count. A roof image by itself does not establish an installation date.

For the Dallas warehouse request, audit the fields separately:

• Address: Does the source identify this building, rather than another location of the tenant?
• Roof age: Is there evidence about the current roof covering? A building construction date answers a different question.
• Contact: Does a current source connect the person or office to this property’s service purchasing?
• Need: Has anyone established a service requirement or a buying date?

Check those fields before outreach. If the rep tells a manager the roof is old, the manager can reasonably ask for the installation record. Without one, the rep has to withdraw the claim before explaining the service offer.

The accuracy guidance treats missing context and fact checking as separate parts of improving an answer. My sales rule follows that distinction: an empty field stays empty until a source or a conversation fills it.

Ask ChatGPT to identify where evidence is missing and return fewer records when needed. Request the source URL beside each researched fact. Open it and check that it describes the same property; a citation to a tenant’s homepage does not support a roof installation date.

The customer-profile prompt that sets the territory

A commercial HVAC campaign can start with warehouses of 25,000 square feet or more, at least five visible rooftop units and a defined service territory.

Use that as a planning example for a service department working within 23 miles of its north Dallas shop. The offer is scheduled maintenance and backup rooftop service. The service manager confirms equipment details during qualification and prices the scope around the crew’s capacity and travel time.

Copy this prompt and replace the operating facts with yours:

Build a customer profile for our commercial HVAC service campaign.

Our starting territory is within 23 miles of our north Dallas shop. We want
warehouses and light-industrial buildings of at least 25,000 square feet
with 5+ visible rooftop units. Our offer is scheduled maintenance and
backup service. We have capacity to onboard three accounts this quarter.

Produce a table with these columns:
criterion | purpose | required evidence | exclusion | first question
Separate building fit from buying authority and current demand.

Use these facts for planning. Generate no addresses or contact details.
Ask for missing operating constraints before
recommending a broader territory.

Read the disqualifiers closely. Exclude buildings beyond the service boundary and accounts with access requirements the crew cannot support, even if the manager wants a quote. Put the reason in the campaign sheet so another rep does not reopen the same account next week.

For a roofing campaign, replace the rooftop-unit criterion with the roof age band and material your crew handles. A paving campaign can start with properties that have paved parking areas, followed by a field measurement when estimating begins. Keep each trade’s offer in its own prompt.

Use the contractor customer-profile worksheet to agree on the filters with the service manager. Before ordering a file, have the manager name two exclusions the rep would otherwise miss.

Prompts for a call script and a routing question

A first prospecting call should establish who handles the service decision and earn one next step, using only the building facts your team has checked.

For the warehouse campaign, start with responsibility for the maintenance agreement. A reception desk can route that question even when the contact column is empty. Give ChatGPT the verified facts and the question you need answered.

Write a first-call script for a commercial HVAC service rep.

Verified inputs: this is a warehouse in our service territory; our
property record shows 5+ visible rooftop units. Unknowns: who approves
work; equipment condition; the current service agreement.

Write an opening under 35 words and a routing question. Return separate
branches for reaching the manager or the wrong department. Add voicemail.
Ask one qualification question per branch. Offer our vendor information
as the next step. Use placeholders for our company and rep.

Avoid claims of a site visit, equipment failure, existing relationship
or permission to text. Return a script for human review.

A suggested opening for that exercise is:

Hello, I’m [rep] with [service company]. We provide commercial rooftop maintenance in [area]. Who handles the HVAC service agreement for your warehouse?

That gives the person answering a task they can handle. Once the rep reaches the person responsible for maintenance, I would ask whether the service agreement covers this building alone or a property portfolio, then record which office approves work at the site. Preserve the property ID.

Leave the service reason and callback number in a voicemail. Save the equipment discussion for a live conversation with the manager responsible, when the rep can establish that the building record refers to the right service location and ask about the scope of the agreement.

Practice the script aloud. Cut any phrase a technician would struggle to say naturally. The outbound sales playbook covers how the call fits a wider contact schedule.

Cold emails built from one checked property fact

A property-manager email should connect one checked account fact to the offered service and ask for a next step the recipient can answer easily.

Keep the subject tied to that request. An inquiry about vendor requirements needs no prediction about a failing compressor or a leaking roof. A match reason belongs in the research record even when it is too detailed for the first email.

Build sender fields into the approved template. The FTC’s summary of the CAN-SPAM Act identifies truthful headers and subjects, advertising disclosure and an opt-out as requirements.

Draft a commercial prospecting email using this account record.

Verified fact: the property is a warehouse inside our service area.
Our offer: commercial rooftop HVAC maintenance and backup service.
Our request: ask who manages HVAC vendor approval for this site.

Write a plain subject and a body under 85 words. Include one request.
Use [company], [rep] and [property address] placeholders. Mention only
facts supplied here. Do not imply a previous conversation, inspection,
urgent repair, endorsement or guaranteed savings.

Append our approved advertising footer. It must identify our company
and include our valid postal address. Explain a reply-based opt-out.
Leave missing footer fields as placeholders.
Return a draft. List any blanks we must fill before sending.

The working email can be this short:

Subject: HVAC vendor approval for [property address]

Hello [name or facilities team],

[Company] provides commercial rooftop HVAC maintenance in [area]. Who manages HVAC vendor approval for the warehouse at [property address]?

If that is your responsibility, may I send our service scope and vendor information?

[Rep], [company], [work phone]

Advertisement from [company]. [Valid physical postal address]. To stop our marketing emails, reply “unsubscribe.”

Fill every bracket before sending and test that replies reach a monitored inbox. The footer is part of the draft, not a line to remove because it takes space. The sending requirements appear in the compliance section below.

When a manager requests a scope, give ChatGPT the call notes and approved document. Ask it to reference the requested service and agreed review date. Save that reply separately from the first-contact template.

Rehearse objections without inventing urgency

ChatGPT can help a rep practice responses to an incumbent vendor, a closed budget or a wrong contact before the rep speaks with a prospect.

I would practice the incumbent-vendor response first. A manager with a service agreement has answered an important qualification question. The next question is whether backup coverage or future vendor review is relevant.

Act as a commercial property manager for a sales practice session.
Our rep offers rooftop HVAC maintenance and backup service.

Start with the objection that your building already has a contractor.
Wait for my response. Then assess whether I acknowledged your answer,
asked a reasonable qualification question and avoided unsupported
claims. Suggest a shorter response in the rep's own plain language.

Continue with a budget objection, then a request to stop contact.
For the stop request, the correct action is to end the sales approach
and record the suppression. Accept that instruction without a rebuttal.
Do not invent testimonials, case histories or competitor failures.

An approved response can acknowledge the existing arrangement and ask whether the manager ever uses a backup provider. If the answer is no, ask whether a future vendor-review date is relevant. A clear refusal ends that branch.

Have the service manager approve response-time commitments and savings calculations before adding them to a reusable script. If dispatch offers scheduled service in that territory, the draft should offer a scheduled visit. Leave emergency response out of that script.

Keep a small record of objections heard during calls. Have ChatGPT group those notes by reason, with the property IDs attached. A group of wrong-contact replies calls for better routing research. Rewriting the same opening will not supply the missing responsibility information.

Follow-up prompts with a reason to return

A 21-day prospecting schedule gives each eligible account a small number of distinct contacts and lets the buyer’s response determine what happens next.

Use a schedule to organize the rep’s work. A manager’s instruction takes priority over the calendar, so a request for information next quarter should move that account out of the current sequence and put the requested date on the rep’s future queue. Record the reason for waiting.

Planned day	Eligible next action	Reason for the contact	Record before proceeding
1	Human call or reviewed first email	Identify the service purchasing role	Correct department and contact outcome
5	Send vendor information if requested; otherwise use an approved follow-up channel	Resolve the routing or document question	Request, delivery outcome or suppression
12	Follow up on a specific offer	Ask about an agreed scope review or vendor process	Response and next date
21	Close the active sequence	Ask whether future contact is wanted	Requested timing or campaign closure

Create the next-action queue from these prospecting notes.

Use only the property IDs and recorded events I provide. Separate
requested follow-ups from unresponsive accounts. Exclude every
suppressed contact. Respect dates the prospect requested.

For each eligible row, return these columns:
next date | approved channel from our record | reason | draft under 65 words
Add no text messages or automated calls. If channel eligibility is missing,
flag the row for review rather than writing a message.

Do not treat silence, an email open or a website visit as agreement
to buy. Return the queue for review before any messages are sent.

Replace empty check-ins with the next unfinished task. If a manager requested insurance information before reviewing vendors, send that document and ask where to submit the maintenance scope. Keep the request beside the draft so the rep can answer without searching another system.

How to feed a Ground Control file into ChatGPT

Upload a working copy of the Ground Control CSV or Excel file, preserve the delivered property facts and add outreach drafts in separate columns.

Start with ten records. The official spreadsheet-cleaning workflow recommends keeping the original file and reviewing a cleaned copy with change notes. Apply that discipline before a sales draft enters the CRM.

Map the delivered headers to a campaign sheet. Add your own stable IDs if the working copy needs them. Keep source values separate from anything the rep learns later.

Working field	Where it comes from	Treatment in the prompt
property_id	Campaign ID assigned by your team	Preserve exactly through every output
property_address	Delivered property record	Keep separate from any owner’s mailing address
matched_attributes	Delivered match reason and property fields	Retain original wording beside the draft
owner_or_contact	Optional research add-on or your checked research	Preserve missing values
contact_status	Rep’s dated research and conversations	Control eligibility and next action
first_call_question	New sales preparation	Mark as a draft
email_draft	New sales writing	Require review before sending

Contact research is an add-on to Ground Control. Request it when your team needs owner information or a named decision maker with role and available work contact details. Budget that separately from the building price.

Use only the fields needed for the task. A profile exercise can work from property IDs and attributes without a person’s email address, leaving the rep to add approved contact fields later when preparing the queue for a specific account and sending method.

Inspect the attached ten-row prospecting file before changing it.
Report the row count and headers. Flag any missing IDs. Ask me to
confirm the mapping to these fields:
property_id | property_address | matched_attributes | contact_status

After I confirm, create a working copy. Preserve source values and
all property IDs. Add these columns:
first_call_question | email_draft | review_note
Use only supplied property facts. Leave unknown contacts blank.
Leave contact research for a separately requested task.

Exclude suppressed records from drafting. Flag possible duplicates
for review instead of deleting them. Return the working file with
input and output row counts and a list of changes. Do not send messages.

The commercial prospect-list guide covers the underlying record structure. For the warehouse campaign, buildings with rooftop HVAC are the matching property category. Keep the match reason in the sheet even when the email uses only the address and service area.

Checking the output before it reaches your CRM

Reconcile ten drafted records against their sources before processing another batch so a changed address or unsupported equipment claim gets caught before import.

I would inspect all ten in the first batch. A plausible sentence can hide a changed number, and a correct address can sit beside the wrong owner. Open the working file beside the original.

Check these items:

• Every input ID appears in the output, with suppressed rows retained in the audit record and excluded from drafting.
• No tenant address has replaced the building address, and no owner mailing address has replaced the service location.
• A visible-feature match has kept its meaning. Five or more rooftop units must not become five aging units.
• Missing contacts remain blank, with any proposed research placed in a review column.
• Each scheduled action follows the rep’s recorded conversation and the applicable contact restrictions.
• Every message still contains its required sender details and opt-out information.

Read the draft from the recipient’s side. “We noticed your units need service” claims knowledge of condition. “We provide rooftop maintenance in your area” describes the offer. Choose the second opening when the record establishes building fit and nothing about condition.

Deduplicate carefully. Two buildings can share an owner while requiring separate service records. One campus can contain multiple buildings. Ask ChatGPT to flag possible overlaps and let the rep resolve the account structure before import.

Save the prompt version beside the approved working file, with research notes attached to property IDs. Import the reviewed fields into the CRM. Before another batch runs, fix any instruction that changed source values or assigned a contact to the wrong building.

CAN-SPAM and TCPA checks for 2026 outreach

CAN-SPAM applies to commercial B2B email, while TCPA requirements for calls depend on the number and technology involved; ChatGPT drafting supplies no contact permission.

The FTC’s CAN-SPAM guide makes no B2B exception. Commercial prospecting emails in 2026 need accurate sender information and a truthful subject. Identify the message as advertising and provide a valid physical postal address. Explain an easy way to stop future marketing emails.

Honor an email opt-out within 10 business days, and keep the opt-out mechanism working for at least 30 days after sending. The FTC also says hiring another sender does not remove your responsibility. Put those requirements into the sending procedure and test the reply-based opt-out used in the sample email.

For calls, the FCC’s artificial-voice ruling, paragraph 5 and footnote 13, explains that AI-generated voices fall under TCPA artificial-voice restrictions. Under 47 CFR 64.1200(a)(2) and (3), artificial or prerecorded voice marketing calls to wireless numbers and residential lines require prior express written consent. A business purpose does not exempt a mobile number.

I would use these prompts to prepare a person for a call. Add automation or marketing texts only after your sending process establishes eligibility, including applicable do-not-call rules and state requirements for the territory and method you choose.

Maintain contact eligibility in your own records. A purchased phone number or email is contact information; it does not document consent to every channel. Record the number type alongside permission where required. Log requests to stop.

Give ChatGPT a review task with clear limits:

Review these outreach drafts against our supplied sending policy.
Flag unsupported property claims. Check sender identity and opt-out
wording. Confirm that suppressed rows have no outgoing draft.
List any channel or permission fields that require human review.
Leave compliance certification to human review. Invent no consent records.

Resolve flagged rows before loading the sending queue. Record a request to stop marketing contact immediately so every rep who opens the account can see the instruction, including someone preparing a fresh draft months after the original campaign closed. Remove pending outreach.

What a better building list is worth over five years

Two signed HVAC accounts can produce $46,800 in five-year gross profit against a $4,176 acquisition budget for a 96-building campaign.

Use this planning model: each signed account produces $11,700 in annual maintenance revenue at a 40% gross margin after field delivery costs, including travel. Assume five paid years per account, unchanged pricing and margin, and no added repair sales. The campaign has 48 sales hours at a fully allocated $47 an hour, including research, outreach, qualification visits and quotes. That rate includes sales travel and allocated tool costs; there is no separate advertising expense. No contact add-on is purchased; contact research fits inside those hours. Use the high end of Ground Control’s published building price.

Annual gross profit per account is $11,700 x 40% = $4,680. Five-year customer lifetime value, or CLV, is $4,680 x 5 = $23,400 in gross profit. This simple version assumes five paid years and applies no discount rate.

Campaign assumption	Mixed building list	Ground Control fit list
Buildings worked	96	96
Sales hours	48	48
Conversations	12	24
Qualified visits	6	12
Proposals	4	8
Accounts signed	1	2
Property-list cost	$0	96 x $20 = $1,920
Allocated sales cost	48 x $47 = $2,256	48 x $47 = $2,256
Total acquisition spending	$2,256	$4,176
Customer acquisition cost	$2,256 / 1 = $2,256	$4,176 / 2 = $2,088

The fit-list funnel assumes conversations with 24 of 96 buildings, or 25%, versus 12 of 96, or 12.5%, for the mixed list. Both campaigns convert half their conversations to qualified visits, two-thirds of visits to proposals and one-quarter of proposals to signed accounts. The fit list finishes at 2 of 96 buildings, about 2.1%. Targeting changes the first stage; the later conversion rates stay equal.

Allocate the fit campaign’s 48 hours as 12 for research and preparation, 16 for outreach, 12 for qualification visits including sales travel, and 8 for proposals. For the mixed list, allocate 24 hours to research and unsuitable records, 14 to outreach, 6 to visits and 4 to proposals. Both budgets total 48 hours. Keep the same acquisition budget in the low, base and high cases below; only signed accounts change.

Customer acquisition cost, or CAC, divides all campaign acquisition spending by accounts won. Lead-list ROI divides gross profit less list and sales costs by those costs: ($46,800 minus $4,176) / $4,176 = about 1,021% over five paid years. First-year gross profit is $9,360, leaving $5,184 after acquisition spending.

Fit-list outcome	Accounts won	CAC	Five-year gross profit	Five-year ROI (rounded)
Low	1	$4,176	$23,400	460%
Base	2	$2,088	$46,800	1,021%
High	3	$1,392	$70,200	1,581%

At two accounts, CAC payback is $2,088 / ($4,680 / 12) = about 5.4 months of account gross profit, assuming that profit accrues evenly after service starts. It excludes the time spent prospecting before signing. One signed account leaves the fit campaign with a higher CAC than the mixed-list case. I would examine the lost accounts before ordering a larger territory.

Run your own retention and cost assumptions through the lead-list ROI calculator. The decision rule is to expand a property segment when signed-account gross profit and measured sales hours support the acquisition cost.

How to use ChatGPT for prospecting, step by step

Start with one property segment and human-reviewed outreach over 30 days, using the account outcomes to decide whether the territory deserves more sales hours.

1. Days 1 and 2: define the account. Give ChatGPT your service scope and operating constraints. Approve the customer profile with the service manager, including exclusions and the smallest account worth onboarding.
2. Day 2: choose the first offer. Request call and email drafts around one next step, such as sending a maintenance scope. Remove commitments your dispatch or estimating team has not approved.
3. Day 3: order the matching buildings. In Ground Control, select your commercial territory, warehouse or industrial type, 25,000+ square feet and 5+ visible rooftop units for the HVAC campaign. Commercial records cost $10 to $20 each, with the file within 48 hours after payment. Choose owner or decision-maker research add-ons if needed.
4. Days 4 to 6: prepare ten rows. Preserve the delivered facts, assign campaign IDs and confirm the headers before ChatGPT drafts. Research missing purchasing responsibility. Apply suppression records and channel eligibility before writing account-level messages.
5. Days 7 to 10: review and start contact. Approve the first ten records against their sources. Make human calls through your approved process and send eligible, reviewed emails. Record replies in the CRM each day.
6. Days 11 to 20: follow the buyer’s answer. Use ChatGPT to prepare requested information and organize next dates. Ask who approves service spending. Arrange authorized qualification visits where the prospect agrees to discuss a scope.
7. Days 21 to 27: price the qualified work. Confirm equipment details and delivery costs during the site review. Prepare the scope and have the service manager approve commitments. Give ChatGPT approved notes when asking it to edit proposal language.
8. Days 28 to 30: count signed accounts and costs. Reconcile sales hours, list spending and account gross profit. Review wrong-fit reasons before changing the filters. Carry requested future dates into the next month’s queue.

For roofing outreach, use the commercial-roof property list and swap in your supported roof age band and material. Tie the email offer to an authorized survey or maintenance review.

A weekly scorecard for better prompts

Track signed-account gross profit alongside the conversations and qualified visits that produced it, with weekly counts tied to the property IDs your reps worked.

Ask ChatGPT to summarize the records behind those numbers. Require property IDs for examples and leave explanations tentative when the notes do not establish a cause. A count of written emails tells you how much text was produced. The scorecard needs the buyer’s response.

Recorded pattern	Question for the sales manager	Change to test
Buildings repeatedly fail qualification	Which filter or exclusion was missing?	Tighten the next property segment
Calls reach the wrong office	Who buys service for this property?	Improve contact research and routing language
Managers respond but decline a visit	Does the offer fit their stated need?	Revise the first requested step
Quotes stall after site review	Was scope, authority or timing unresolved?	Add the missing qualification question

Keep the successful prompt and change one part for the next small batch. Record which version each rep used. Review the underlying accounts before crediting a wording change with more sales, particularly when the territory or property mix changed too.

For public buildings, Bid Control by PorchRocket finds city, county and school opportunities with bid papers, due dates and addenda. Use that public-work process when the next step is responding to a solicitation. ChatGPT can help prepare language from checked documents; the buyer’s packet governs the response.

The Ground Control file behind your next campaign

The fastest way to start commercial prospecting with ChatGPT is to order matching buildings from Ground Control by PorchRocket and draft outreach from those records. Select business type, building size and features such as 5+ visible rooftop units, roof material or paved parking areas. Commercial records cost $10 to $20 per building and arrive within 48 hours after payment. Report a wrong match within 30 days for a credit back. These are prospects your team contacts; their owners have not asked for quotes.

Build the order around what the rep needs to see:

• A service territory and building category the crew can cover, such as warehouses and industrial properties in the approved Dallas area.
• Building-size bands and visible features tied to the offer, with a reason each property matched.
• Owner information and decision-maker contact research as optional add-ons, with no charge for a contact we do not find.
• A CSV or Excel file your team can map to its campaign sheet, then provide to ChatGPT for reviewed drafting.
• An exact price before payment, with commercial orders starting at ten buildings.
• Properties sent only when Ground Control is 90%+ sure they match, backed by the 30-day credit for a wrong property.

Choose the buildings your crew can serve, then give the rep words worth saying to the person who runs them.