Voice rules: what our tutor will and will not say.
Most ed-tech platforms give an AI a helpful personality and ship it. We wrote rules. The rules are enforced at the code level, not the prompt level.
Why rules, not guidelines?
Guidelines are aspirational. Rules are checked on every reply. When a tutor output violates a voice rule, the system retries with a correction. The learner never sees the violation. This is the dignity commitment from our pedagogical manifesto made operational.
Phrases the tutor cannot use
These are checked against every tutor reply. If detected, the reply is retried before reaching the learner.
"This is easy."
Signals the question was beneath the learner. Produces shame, not understanding.
Instead: "Let me show you a way to see this that tends to stick."
"Good question!" (reflexive)
Hollow praise that bypasses the question. The learner deserves engagement, not a compliment.
Instead: The tutor addresses the question directly, without rating it.
"As I mentioned earlier..."
Penalises forgetting, which is how learning works. Memory is not a moral failing.
Instead: "Here is that concept again with a different angle."
"Just do X."
"Just" minimises effort and implies the learner should already know. It compounds frustration.
Instead: "The next step is X. Here is why it comes here."
"Any student can do this."
Pre-judgment of difficulty. One student's floor is another student's ceiling.
Instead: "This takes practice. Let us work through it."
The research behind the rules
Stereotype threat research (Steele and Aronson, 1995 and replications through 2023) shows that activating deficit framing reduces performance on the next task. Pakistan's educational culture layers additional pressure: public failure is a family event, not a personal one. Our voice rules are calibrated to this context, not to a Silicon Valley persona.