Why an AI receptionist should never diagnose an emergency
The job on an emergency call is to deliver a fixed safety message and get your on-call person moving. Not to figure out what is wrong.
A caller says the basement smells like gas. A good receptionist does not ask where the smell is strongest, whether the pilot light is out, or what brand the furnace is. A good receptionist tells the caller to get out.
That is the rule we build into every call: the agent never diagnoses. Here is why.
It cannot see anything
A person on the phone describes a smell, a sound or a stain. A trained technician on site reads it differently. A voice agent is working from a few words, in a stressful moment, with background noise. The gap between what was said and what is true is where wrong advice comes from.
A confident wrong answer is the worst answer
AI language models are good at sounding sure. On an emergency call, "that's probably nothing, let's book you for Thursday" is exactly the failure to avoid. So the agent is not asked to judge. It is asked to recognize that a call may be dangerous and to respond the same way each time.
The words are fixed, not generated
For each emergency type, the safety message is written in advance and read as written. The model does not improvise it. For example:
- Gas smell: leave the home now, call 911 or your gas company from outside, and do not use light switches or anything that could spark.
- Carbon monoxide alarm: get everyone into fresh air and call 911. Do not go back inside until responders say it is safe.
- Burst pipe: if it is safe, shut off the main water valve and stay away from wet outlets and appliances. Then the team is alerted.
- Electrical hazard: stay away from the hazard, do not touch anything sparking or hot, turn off the main breaker only if it can be done safely, and call 911 for fire or smoke.
- Active roof leak: if it is safe, move belongings away, stay out of a room with a sagging ceiling, and leave and call 911 if there is structural damage.
Because the text is fixed, you can read every line ahead of time, and so can your lawyer and your insurer.
When in doubt, treat it as an emergency
A system that is unsure will occasionally treat a harmless call as urgent. That costs a short phone call. The opposite mistake can cost far more. Err on the side of the safety message.
What the agent does do
- Reads the safety message.
- Alerts you right away, or rings your on-call person if you turn on live routing.
- Collects the address and a callback number if the caller can give them.
- Records everything so you can review the call.
It does not tell the caller the cause, say a repair is safe, or promise a price or an arrival time.
What it is not
A phone agent does not replace 911, your gas company or your utility. It is a front desk that is honest about its limits. If you are evaluating any AI receptionist, ask to hear its emergency calls, word for word.