Every school communication platform eventually meets the same message. A parent, furious about a schedule change, typing things into a class thread they’d never say at pickup. A student testing what they can get away with in a homework group. The question isn’t whether it happens — ask any principal — it’s what the platform does in the three seconds between typing and delivery.
Bloomz now has an answer rolling out: profanity screening with an admin escalation queue, covering posts, comments, and messages. Here’s how it works, and — because moderation is one of those features where the design choices matter more than the checkbox — why it works the way it does.
Warn first. Hold on override. Never silently censor.
When a message contains flagged language, the sender sees a warning before anything is delivered, with the flagged words highlighted. Most people, given three seconds and a highlight, simply rephrase — the message sends instantly and nobody else ever knows.
Choosing “send anyway” is where our design departs from a naive filter: the message is not delivered and not discarded. It’s held in a review queue where the group’s own moderators — escalating to school administrators — read it and decide: approve it for delivery (it posts as the original author, not as the admin), reject it, or, for repeat offenders, suspend the author. Suspension is deliberately admin-only; a class teacher can publish or discard, but removing someone’s voice from the community is a school-level call.
This threads a needle that pure blocking and pure alerting both miss. Silent blocking breeds “why didn’t my message send?” support tickets and accusations of censorship. Pure alerting means the damage is done before anyone reviews it. Warn-then-hold gives the sender agency, keeps a human in every judgment, and means nothing offensive reaches a child or a class thread while it waits.
Severity decides the outcome — because not all flags are equal
A single on/off filter treats a swear word and a threat the same way. They aren’t the same, so Bloomz doesn’t treat them the same:
- Threats are blocked outright. No “send anyway,” no override. Some messages don’t deserve a second chance.
- Profanity and harassment warn first and hold for review on override — the flow above.
- Safety-sensitive language — a student who sounds like they’re in crisis — is delivered, while quietly alerting your team. This is the decision we’d defend hardest: blocking a cry for help so it reaches no one is the worst possible failure of a “safety” feature. The message goes through; the adults who need to know, know.
A wordlist you can see, not a model you can’t
The screening is a deterministic wordlist — no AI, no language model scoring your community’s speech. That’s a deliberate stance, consistent with how every Bloomz AI feature is built: AI never makes a judgment about a person. A wordlist is transparent (your district can see exactly what’s flagged), predictable (the same message is treated the same way every time), and explainable at a board meeting — none of which is true of a black-box toxicity model that quietly decides a parent’s dialect is “aggressive.” When a flag is wrong, a human reviewer overrides it; when a model is wrong, nobody can even say why.
Off by default, on your terms
Like Family Messaging, moderation ships off by default. Nothing changes for your community until a school administrator decides to turn it on. Some schools want screening on student-facing groups only; some want it everywhere; some have board policies to align first. The switch is yours, and until you flip it, Bloomz behaves exactly as before.
One honest scope note: screening reads text — post bodies, comments, messages. Words baked into an uploaded image or screenshot aren’t scanned today. We’d rather tell you that plainly than let a demo imply otherwise.
What this means for your evaluation
If you’re comparing platforms, add three questions to your rubric that a “profanity filter ✓” checkbox won’t answer:
- What happens on override? Delivered anyway, silently deleted, or held for human review?
- Are threats and crisis language handled differently from swear words? Ask specifically what happens to a message that sounds like self-harm.
- Can you see the rules? A visible wordlist beats an unexplainable model when a parent challenges a flag.
Community-safe messaging joins the same platform that already keeps conversations private, translated, and archived — one more reason the messaging layer your district trusts should be the one your district actually controls.