Know your teen’s okay. Even the part they hide.
Serious risk almost never uses the obvious word. LarkWard is being built to read the whole conversation.
How it works
Nothing here is a banned word.
Which is exactly why a keyword filter finds nothing in this thread. Read it the way a person would, and it changes.
- Friend u get the footballs?
- Your teen yeah grabbed 2
- Friend after school by the field?
- Your teen bet, same as last week
Design preview. The product is not yet released.
Beyond the messages
It isn’t only what they say.
Risk shows up as a pattern before it shows up in words. An app that appears at midnight. A number nobody recognises that talks more than every friend combined.
Every app, and when it turned up
What’s on the phone, when we first saw it, and how active it is. An app that appeared at 11:40 at night and never existed before is worth knowing about.
Who they talk to, and how much
Messages, photos exchanged, and the hours it happens in. We show you the counts and the timestamps. We don’t tell you who the person is, because we don’t know and you probably do.
Design preview. The product is not yet released.
A number on its own doesn’t mean anything.
Hours in an app. A contact you don’t recognise. A name that changed on Tuesday. Separately they are trivia. Together they are the thing you needed to know.
What we’re building toward11 hours on Snapchat, 60% of it after 11pm, almost all of it with one number that turned up three weeks ago. Here is the thread that concerns us.
A new name doesn’t make it a new person
Rename a number, and anything that watches names loses track of it. LarkWard follows the number, so the rename changes nothing except that now you know it happened.
Design preview. The product is not yet released.
Keyword filters see words. We read the conversation.
We aim to catch more, never everything, and we’ll stay honest about the difference.
Context, not keywords
A word-matching scanner reads a thread like the one above and finds nothing to flag. Contextual analysis reads the whole exchange instead: who, when, how often, and whether the story holds together. That is where coded risk actually shows itself.
Precision first, on purpose
A tool that flags everything gets muted within a week. We would rather surface few, high-confidence concerns than flood a parent with noise. It is a deliberate design choice, and the harder one to build.
Honest about the limits
No detector catches everything, and we will not claim ours does. We will tell you what it is good at, where it is weak, and how often it gets things wrong, including when that is inconvenient for us.
The risks a teenager actually faces
Not a cartoon list of fears. The real, adult-grade dangers of growing up online, surfaced calmly and only when they matter.
- Vaping & substances
- Grooming & risky contact
- Alcohol
- Self-harm signals
- Sextortion
- Bullying
Open, not optional
Your teen is told, not asked.
Protecting your child is a parent’s call, and LarkWard doesn’t hand that decision to a teenager. What it will never do is run in secret.
On their phone, in plain language
The app is visible on the home screen and says what it does: what reaches you, and what never does. If it stops running, you are told.
Open about the what, private about the how
They see the scope. They don’t see the detection method, because publishing that is a manual for evading it.
Design preview. The product is not yet released.
We won’t flag your kid for how they talk.
Detection systems can quietly over-flag slang, accent, and the way different families talk. That is unfair, and it is also just inaccurate. So we are building the measurement in from the start.
- Measure Test detection across dialects and how different kids actually talk.
- Close the gap Where we over-flag one group, we treat it as a bug to fix.
- Publish Report how fair the detection is, in the open, before we claim anything.
Capable doesn’t mean a permanent record.
Recent history only
We keep what matters for a bounded window, then it auto-deletes. No archive of a child’s whole life.
Never sold
A family’s data is not a product. It is never sold, and never shared except with people the parent approves.
Honest, never covert
Your teen knows it’s there. We’re open about what we do, and private about how detection works, because publishing the method just teaches evasion.
The honest answers
Is LarkWard available yet?
Not yet. LarkWard Technologies, Inc. is actively building the product, iOS first. Nothing is for sale, there is no app to download today, and we don’t have customers yet. If you’d like to hear when it’s ready, get in touch.
Will my teen know LarkWard is there?
Yes. That is a design principle, not a setting. The teen installs their own app and can see the controls in effect. We are not building a spy tool. What they will not see is the analysis itself, because showing a teenager exactly what gets flagged just teaches them to evade it.
Will parents be able to read the actual messages?
Yes, when something is flagged. The design is to show you what we’re worried about and the thread behind it, so you can make a human call. It is not a browsable feed of your child’s entire life. That history is bounded and auto-deletes.
Will it catch everything?
No, and we won’t pretend otherwise. The goal is to catch far more than keyword filters, especially coded and contextual risk. Anyone promising to catch everything is really selling you a flood of false alarms.
You say “fair by design.” Can you prove it?
Not yet, and we won’t claim it before we can. Fairness measurement is built into our development plan and is a gate we have set for ourselves before launch. When we have results from real evaluation data, we’ll publish them, including the parts that don’t flatter us. Until then this is a commitment, not a result.
Know they’re okay. On your family’s side.
We’re building it now. If you’re a parent, a clinician, or a family-law professional who wants to follow along, we’d like to hear from you.