Why I Don't Build the School Chatbot Everyone Asks For
It's the first feature almost every school asks me for now: a chatbot that talks to parents on its own. Drop it on the website, wire it to WhatsApp, let it answer questions around the clock so the office stops drowning in the same forty messages. On paper it's the obvious win of the AI era. I usually say no — or at least, not that.
Not because the technology can't hold a conversation. It can, fluently. I say no because of what a school answer actually is, and what happens the first time the fluent one is wrong.
A school answer is not a consumer answer
When a shopping chatbot guesses wrong about a refund policy, you're mildly annoyed and you move on. When a school chatbot tells a parent pickup is at 1 p.m. and it's actually noon, a child stands at a gate for an hour. When it improvises an answer about fees, an exam date, a medical form, or whether a specific student is on the approved-collection list, it isn't being unhelpful — it's manufacturing a fact about someone's child and broadcasting it with total confidence.
That's the asymmetry people underestimate. The parents most likely to message a school are anxious — about money, about safety, about a child who didn't come home on time. They are the worst possible audience for a confident guess. And a language model, by default, will always rather produce a smooth answer than admit it doesn't know. In most products that tendency is a quirk. In a school it's a liability with a name attached.
The thing schools are really asking for
Listen past the word "chatbot" and the actual request is simpler: stop making the office answer the same question forty times, and stop letting messages fall through the cracks. That problem is real and worth solving. It just doesn't require handing the conversation to a machine. It requires making the humans faster and the system tighter.
So instead of an autonomous bot, I build the unglamorous version:
- Draft, don't send. AI proposes the reply, pulling the exact line from the school's own handbook or calendar. A person who knows the family approves it. The parent gets a fast answer that a human stands behind.
- Retrieve, don't invent. When the bot does respond directly, it answers only from a known, school-approved source — "here is what the handbook says" — and refuses anything it can't ground. No source, no answer.
- Escalate anything specific. The moment a question is about a particular child — their attendance, their balance, their pickup — it goes straight to a human. General information can be automated. A specific child never is.
- Never drop a message. Most of the office's pain isn't answering hard questions; it's losing easy ones. A queue that guarantees every parent message is seen and closed out beats a clever bot every time.
Boring on purpose
The result doesn't demo like the future. There's no charming assistant with a personality; there's a faster office that stops repeating itself and stops forgetting people. But it has the one property a school cannot do without: a parent never receives a confident answer that no human would have given. The institution still speaks in a single, accountable voice — it just speaks faster.
I'm not against AI in a school. I put it to work every day. I'm against pointing the most overconfident tool we've ever built straight at the most anxious, highest-stakes audience a school has, and removing the one person who would have caught the mistake. Keep that person in the loop and AI is a quiet superpower. Take them out to save a few seconds and you've automated the fastest way to lose a parent's trust.
The chatbot everyone asks for is a personality. What a school needs is a memory that never drops a message and a mouth that never makes things up. Those are not the same product, and the second one is the one worth building.