ChatGPT for HVAC: where it helps in the office, and where it stops
ChatGPT is a general-purpose writing assistant. In an HVAC office it is genuinely useful for the words — the customer reply, the follow-up note, the plain-English version of a tech's write-up — and it stops exactly where the work depends on your records: which rooftop unit is under an agreement, what is on the truck, and which call is about to slip.
The detail.
What it is actually good for in an HVAC office
The assistant is a fast writer with no memory of your business. That makes it a good fit for the parts of the job that are only words — and a bad fit for anything that depends on your records.
It is good at drafting. Paste in a tech's terse note — "RTU-3 no cooling, found failed contactor, replaced, holding 55 supply air" — and ask for a short update a property manager can read. It is good at rewriting: take the renewal note you send every year and ask for a plainer version, or a shorter one, or the same note with a firmer ask. It is good at structuring: ask it to list the steps a tech should work through on a no-heat call, or to turn a loose punch list into a checklist with owners and due dates.
It is also good at summarising a long email thread into the two or three decisions buried in it. That is real time back in an office that fields a hundred messages a day.
What it cannot do is verify any of it. Everything it writes is a plausible draft, not a fact, so the useful mental model is a fast junior writer who has never set foot in your shop — great for the first pass, never the last word.
Where a general-purpose assistant stops
The assistant has no access to the things that make a commercial call what it is. It does not know that RTU-3 has two repairs this year and is nine years into a twelve-year life, that the site is on a full-coverage agreement, or that the same customer has a second building with a compressor on order. It does not know your labor rates, your agreement terms, what is on the truck, or which tech is certified for the work. Ask it anyway and it will answer confidently — inventing a model number, a code section, or a refrigerant charge that sounds exactly as sure as the correct answer next to it.
It also leaves no trail. Whatever it drafts, someone still has to attach it to the right customer, the right site, and the right piece of equipment, and file it where the next person can find it. A draft that lives in a chat window is not a record.
The last limit is confidentiality. Treat any general-purpose assistant like a public channel. Do not paste customer names, service addresses, account numbers, pricing, payment details, or anything covered by an agreement. If you would not email it to a stranger, it does not belong in the prompt.
A simple rule for what goes in and what stays out
Keep the assistant on the words and keep the records where they belong. Draft the reply there; send it from your own system. Rewrite the note there; file the finished service record against the unit. Summarise the thread there; keep the account history where the whole office can see it.
One habit covers most of the risk: before you send anything the assistant wrote, check every specific — names, numbers, dates, coverage, and anything technical — against the record. The writing is the cheap part; the checking is the job.
Where the records live
Thermal is the other half of the pairing. It keeps the customer, the site, the equipment, the agreement, the work order, and the invoice on one spine, so every job starts from the unit rather than a calendar slot. The service history, the warranty dates, and the agreement coverage are already attached when the call comes in, and a renewal gets proposed before it lapses instead of after.
That is the split worth holding onto: use a general-purpose assistant for words, and keep the record in the system that knows the equipment. Thermal describes its own automation the same way — a suggestion shows its reason before you accept it, and you can turn it off. QuickBooks Online sync is in early access; Sage Intacct and Business Central are on the roadmap.
Questions this raises.
Is ChatGPT good for HVAC work?
For the words, yes — drafting customer replies, rewriting notes, turning a tech's write-up into something a property manager can read, and summarising a long thread. It is not good for anything that depends on your records, because it cannot see your equipment, agreements, schedule, or parts.
Can ChatGPT look up my equipment or service history?
No. A general-purpose assistant has no connection to your business records, so it cannot know which unit is under an agreement, what has been repaired, or what is on the truck. If you ask, it will often produce a confident answer that is simply made up — verify every specific against the record.
What should I never paste into ChatGPT?
Customer names and service addresses, account and payment details, pricing, credentials, and anything covered by a service agreement. Treat it like a public channel: if you would not email it to a stranger, keep it out of the prompt.
Does a general-purpose assistant replace a field service system?
No. It produces text; it does not keep the record. A field service system is where the equipment, agreement, work order, and invoice live and stay attached to the customer. The assistant can draft around the edges; the system is the source of truth.
The records, on your own book.
If you want to see the agreements, the equipment history, and the schedule read from your own work orders rather than a spreadsheet, that's a demo.