SecurityChannels, platforms and more
AI Chatbot for Security Companies: Two Rules Come First
A security firm can get real value from a chat widget. Quotes, surveys and service visits all answer well. But two rules have to be written first, because getting them wrong is worse than having no chat at all.
Read this part first
Rule one. The bot never takes a report of a break-in, an alarm sounding now, or an intruder on site. That message gets one response: contact emergency services, and here is the monitoring line. No clarifying questions, no address capture, no being helpful first. Chat queues, and a person in that moment belongs on a phone.
Rule two. The bot never confirms what security a named address has. Not the system type, not the brand, not whether it is monitored, not where the cameras are, not when an engineer or a guard attends. That is exactly what a burglar would try to collect, and a chat widget is a cheap, anonymous place to try. The refusal applies even when the person says it is their own house.
Treat the widget as a place people probe
Most chat widgets face customers. Yours faces the public, and part of the public is doing homework. Nobody needs a name or a reason to open your chat window and ask about a property.
The dangerous requests rarely look dangerous. They look like a customer who has lost a code, a landlord checking what was installed, or a new occupier asking who monitors the flat. Each is reasonable, each is unverifiable, and a helpful assistant is the wrong instinct.
So write the refusal as an absolute, not a judgement call. Language models are agreeable by design, and a rule with an exception in it will find the exception. One sentence, no conditions: the assistant does not discuss security arrangements for any address, and here is the line that does.
The refusal list
Put these into the bot instructions before any feature work, and test each one yourself.
Test the awkward versions. Ask as an angry customer, as someone claiming to be a police officer, as a person threatening a bad review. A rule only counts if it survives pressure, which is the point of our guardrails guide.
Make the emergency route impossible to miss
Do not rely on the model spotting an emergency. Someone in a panic types badly, and a filter waiting for the word intruder will miss half of them. Say in the opening message that this is an automated assistant, that emergencies go to the numbers shown, and when staff are around.
Put the emergency number and your monitoring line in the widget from the first message, on every page, and in the out-of-hours reply. Nobody should have to phrase a message correctly to see them. Then escalate generously: a needless handover costs two minutes. Our handover protocols guide covers the mechanics, though this sector should sit further towards the human.
What it can safely do
Once the refusals hold, a chatbot earns its place. Everything below is information about your business, not about a customer's site.
Surveys: collect contact details, not vulnerabilities
A site survey request is the most useful thing the widget produces, and it is where teams get careless. It is tempting to ask which doors are weak and where the blind spots are, so the surveyor arrives prepared.
Do not. Keep the request to name, contact details, property type, rough size, whether anything is installed already, and some times that suit. The detail belongs with a surveyor, not in a transcript in an inbox.
The same restraint applies to service requests. The bot takes the fact that a visit is needed and a way to reach the customer. It does not need to know what the panel is doing.
Account changes need a verified channel
Keyholder lists, contact updates, passcodes, duress words and cancellations all carry weight, and none can be verified in a chat window. The bot explains which channel handles them and passes a call-back request to a named person. It should refuse a code even when a customer volunteers one, and never repeat one back. Your account team has an identity process; the bot's job is to get people to it quickly.
Volume, channels and cost
Installer enquiry volume is steady rather than spiky, with evenings busier than most expect. The free plan is $0 forever with no card: 2 seats, 500 messages a month, one widget, one domain, mobile app, 30-day history. Paid plans are Starter at $25 a month and Professional at $95 a month, with 20% off on annual billing. Annual billing takes 20% off; the ladder is on the pricing page. If you also sell network security, see our cybersecurity company guide.
Frequently asked questions
Can the chatbot take a report of a break-in?
No. A break-in, an intrusion in progress or an alarm going off right now must never be handled in a chat window. The reply routes the person to emergency services and to your monitoring line, on the first message, with no questions asked first. Chat queues, chat is not answered to a response standard, and a person in that moment should be on a phone. Build this rule first, then test it yourself.
Why must the bot refuse to say whether an address is monitored?
Because that is the exact question a burglar would ask. Whether a site has an alarm, what type it is, who monitors it, when engineers attend, whether cameras record, whether a guard patrols overnight: none of it may be confirmed in a chat window, for any address, ever. It applies even when the person claims to be the owner. Identity cannot be checked in chat, so the answer is always the same polite refusal and a route to your verified line.
What can it safely do, then?
Plenty, once the two refusals are in place. It answers what services you offer and where you cover, explains monitored and unmonitored systems in general terms, gives your published price ranges, books a site survey, takes a service request, and answers questions about accreditation and insurance. All of that is information about your business rather than about a customer’s site.
Can it make changes to a customer’s account or keyholder list?
It can route the request, never action it. Keyholder changes, passcodes, duress words, cancellations and contact updates all need identity verification, and a chat widget cannot verify anyone. The bot should also refuse to receive a passcode or duress word at all, and never repeat one back. It explains which verified channel handles account changes and passes the request to a named person.
What does it cost to run one?
The free plan is $0 forever with no card: 2 seats, 500 messages a month, one widget on one domain, mobile app, 30-day history. Most installers outgrow that. Paid plans are Starter at $25 a month and Professional at $95 a month, with 20% off on annual billing. Annual billing is 20% off.
Answer the enquiries. Refuse the rest, every time.
EzyConn trains on your service pages, holds the rules you write, and pushes handovers to your phone on every plan. Free forever, no card.
Start freeLast updated . Nothing here is security, legal or regulatory advice. If a crime is in progress, contact emergency services. Follow your own procedures and the licensing requirements that apply where you work. View more guides.