How should HVAC emergency calls be handled when the conversation happens by text?

When a missed call turns into a text about a gas smell, a carbon monoxide alarm, sparks, a burning smell, or water near electrical equipment, the first reply should be a locked, pre-approved safety message: get out, and call 911 or the gas utility from outside. At the same moment, a person on your team should get an alert to call the customer. An AI should never improvise safety advice, give troubleshooting steps, or try to schedule a tune-up in the middle of an emergency.

By the MyWorkFlo team · Published September 27, 2026 · Safety guidance checked against federal pipeline-safety and emergency-physician sources the same day.

Why are emergencies different when the conversation is by text?

On a phone call, a dispatcher hears panic in someone's voice and reacts. By text, the only signal is the words, typed fast and often misspelled. The customer may be standing in the house the whole time. That puts two jobs on any text-back system: recognize the danger from a short, messy message, and respond with wording that is safe no matter what the underlying fault turns out to be.

It also changes what "fast" means. A text-back tool that asks three qualifying questions before noticing the word "gas" has wasted the minutes that matter. Detection has to run on every message, before anything else happens.

What should the customer be told for each type of emergency?

Every instruction below is about the person (leave, move away, call 911), never about the equipment. That line is deliberate, and the next section explains why.

What the customer texts	What the customer should be told	What your team does
Gas smell, "rotten eggs," propane smell	Leave the house now. Don't use light switches or a phone while inside. Once safely outside, call 911 or the gas company's emergency line.	Gets an instant alert and calls the customer once they're safe
CO alarm going off, or furnace running and someone feels dizzy or nauseous	Get everyone outside into fresh air right now and call 911.	Gets an instant alert and follows up after responders clear the home
Sparks or smoke from equipment	Move away from it and leave the area if you safely can. If you see flames, call 911.	Gets an instant alert and calls the customer
Burning smell near HVAC or electrical equipment	Move away from the area. If you see smoke or flames, leave the house and call 911.	Gets an instant alert and calls the customer
Water or flooding near the breaker panel or electrical equipment	Stay away from the area and don't touch anything electrical nearby. If it seems dangerous, call 911.	Gets an instant alert and calls the customer
No heat in freezing weather	Acknowledge it as urgent and move it to the front of the line. It isn't usually a safety-script situation.	Decides on a night dispatch or the first morning slot, based on your rules

The gas guidance matches federal pipeline-safety advice: leave the area, don't operate switches or use a phone near the leak, and call 911 from a safe distance (PHMSA). The carbon monoxide guidance follows the American College of Emergency Physicians: if the alarm sounds, move outside to fresh air, call 911, and don't go back in until responders say it's safe (ACEP).

Why must the AI never improvise in an emergency?

A language model is good at sounding helpful, and in an emergency that's exactly the risk. Asked "I smell gas, what do I do?", a model could reasonably say "try shutting off the gas valve at the meter." That sounds responsible, but it sends a person toward the hazard and assumes the smell is coming from where the model thinks it is. The same goes for "check if the breaker tripped" when there's water on the floor, or "open the windows" when a CO alarm is sounding and the first priority is getting out.

MyWorkFlo follows a bright-line rule:

Person-directed instructions are allowed and pre-scripted: leave, move away, don't touch anything, call 911.
Equipment-directed instructions are never allowed, however trivial they sound: check, reset, flip, open, shut off, inspect.
The wording is locked. The safety scripts are written by the platform in advance, not generated by the model. The business can't edit them, so nobody can soften them on a busy day.
After escalation, the AI stops. It doesn't keep qualifying or offer to book. Any further texts get a short fixed reply confirming the team has been notified and to call 911 if it becomes life-threatening.

This is part of a broader rule. Outside emergencies, MyWorkFlo also won't guess at repairs, invent arrival times, or quote prices it isn't sure about.

How does the instant human handoff work?

Detection runs in two layers. First, a fixed keyword check runs on every incoming message before the AI does anything, and it's deliberately biased toward over-triggering: a false alarm costs your team a glance at a text thread, and a missed one could cost far more. Second, the AI itself is instructed to flag anything that could plausibly be gas, carbon monoxide, fire, or flooding near electrical, even phrasing the keyword list didn't anticipate.

When either layer fires:

The locked safety script goes to the customer immediately.
Your staff get an alert by text message and push notification with the customer's number and a prompt to call now.
The conversation is marked urgent in your dashboard, with the full text history for whoever calls back.

You can add your own trigger words during setup. Some shops add phrases like "pipes frozen" or "elderly parent, no heat" so those conversations alert a person right away. Custom triggers get a generic urgent response rather than a hazard-specific script, because the system can't know what the hazard is.

How should emergencies be handled for Spanish-speaking customers?

A household might text in Spanish, English, or a mix of both, sometimes in the same message. A detector that only checks English, or that only checks Spanish after it has decided the customer "is a Spanish speaker," will miss real emergencies. MyWorkFlo checks every message against both English and Spanish phrase lists, such as "huele a gas," "huele a quemado," "sale humo," and "alarma de monóxido," and sends the locked safety script in the customer's language. The Spanish scripts are direct translations with the same person-directed meaning, for example: "Por favor sal de la casa ahora mismo y, ya estando afuera y a salvo, llama al 911 o a la línea de emergencias de tu compañía de gas."

English and Spanish texting is included on every MyWorkFlo plan (see pricing). For the full picture of how missed calls become text conversations, see missed call text back for HVAC, and for how to set up on-call coverage, see how to handle after-hours HVAC calls.

Common questions

What should a customer be told if they text that they smell gas?

Leave the house now, don't use light switches or a phone while inside, and call 911 or the gas company's emergency line once safely outside. The reply should not include troubleshooting steps, and a person on the HVAC team should be alerted immediately.

Why shouldn't an AI write its own emergency advice?

Because a well-meant but wrong instruction, like telling someone to go find the gas valve or reset a breaker, can put a person in danger. Safety wording should be written and approved in advance, locked so it can't be changed on the fly, and limited to instructions about the person, never about the equipment.

Does emergency detection work in Spanish?

With MyWorkFlo, yes. Emergency phrases are checked in both English and Spanish on every message, such as "huele a gas" or "huele a quemado", and the locked safety script goes out in the customer's language.

Is no heat in freezing weather treated as an emergency?

It's urgent but usually not a safety emergency, so MyWorkFlo treats no heat as a priority call rather than triggering the safety script. If your shop wants it escalated like an emergency, you can add your own trigger words, such as "pipes frozen", which alert your team immediately.

What happens after the safety script is sent?

The conversation is marked urgent and your staff get an immediate alert to call the customer. The AI stops trying to qualify or book. Any further texts get a short fixed reply confirming the team has been notified and to call 911 if it becomes life-threatening.

Put locked safety scripts between your customers and a guess.

Start 14-day free trial

Sources

primis.phmsa.dot.gov — PHMSA, what to do if you suspect a gas or pipeline leak (accessed Sep 27, 2026)
emergencyphysicians.org — American College of Emergency Physicians, carbon monoxide safety (accessed Sep 27, 2026)
myworkflo.com — MyWorkFlo emergency detection and control modes