Kids First
Designing an AI tone check for co-parents in conflict
Context
Web App Design
Conversational AI
Family Coordination
Methods
Problem Framing
Competitive Analysis
Interaction Design
Prototyping
Role
Product Design
UX Research
Team
5 Designers
Engineering Team
PM
Project Manager
Timeline
2023-2025
Status
Paused pre-launch (legal and funding constraints)
Overview
A message that stops a fight before it starts.
Kids First is a co-parenting web app that brings schedules, tasks, and messaging for separated parents into one child-focused space.
This case study focuses on the tone check, an AI feature that reviews every message before it is sent and offers calmer wording, giving parents a moment to pause before a question turns into a fight.
Built by a team of designers, engineers, and PMs over two years. The product paused before launch due to legal and funding constraints; this case study covers the design work and the decisions behind it.
Problem Space
Communication is the app’s core feature. It’s also where parents stop using it.
Co-parenting apps exist so separated parents can coordinate in one place, with a clear record, instead of through scattered texts.
The team's user research with separated parents found that messaging is where it breaks. A pickup question can read like an accusation. One bad exchange can turn a thread into a fight.
Parents who get burned avoid the app or go back to texting. The app only works when both parents use it, so one parent leaving takes the shared calendar, tasks, and record with them.
The problem to solve. How might we make messaging safe enough that both parents keep using the app?
The design challenge: How do you interrupt one harmful message in the moment without turning the app into a referee? Step in too little and the harm gets through. Step in too much and parents feel policed, and they leave anyway.
Who is this for
Built for parents in conflict, but any parent can have one bad message.
Messaging serves two kinds of users: parents in active conflict and parents who mostly cooperate but sometimes don’t. That is why the tone check has two levels.
Nudge the mild,
Stop the harmful message.
The tone check reviews each message before it sends. It is always on for both parents. It responds in proportion to the harm, and offers calmer wording the parent can use or edit.
Checks before it sends.
The other parent never sees the message. It protects the relationship at its most fragile moment.
Nudge the mild.
A softer version is offered, and the parent can still send.
Stop the hurtful.
The send is blocked, with a clear reason.
Offers a way forward.
Calmer wording on the spot, so the parent isn’t stuck and has less reason to leave the app.
The tone check judges the words.
The parent owns the message.
The tone check decides how a draft reads: fine, might read as blaming, or hurtful. It then suggests calmer wording. It never decides what the parent means, only how it is said, and the parent always chooses the final words.
Designing for a model that can be wrong
A tone check will sometimes be wrong. These are the failures we planned for and what each one led us to design.
Research
Tone guidance exists.
Stopping the message doesn’t.
The audit (2024)
I reviewed 10 co-parenting platforms and compared 7 of them across 6 features. Scheduling, messaging, and expenses were well covered. Real-time guidance appeared in none. The closest was OurFamilyWizard's ToneMeter, an optional add-on that flags language but lets the parent send anyway.
Market update (October 2026)
Since the audit, OurFamilyWizard added an AI rewrite version (2025), and TalkingParents (March 2026) and AppClose (May 2026) launched tone tools of their own. All remain optional, and none blocks the send, according to their public documentation.
What this meant for us: Detecting tone wasn’t the gap. What happens next was. Competitors treat tone as advice. We treated it as a checkpoint.
Design Strategy
Step in as much as the message needs,
and no more.
The tone check is always on for both parents, and responds in proportion. A message that might read as blaming gets a nudge the parent can ignore. A hurtful one is stopped, and the parent can still use a rewrite, edit it, or dispute the flag. The one thing they can't do is send a hurtful message as written. The real design decision was where to draw that line.
Two principles
Friction in proportion to harm.
A mild message gets a suggestion. A hurtful one gets a pause that names the words and offers a way forward.
The model detects tone. The parent decides meaning.
The system never writes for the parent. It only stops what shouldn’t go out as written.
The Tradeoff
A hard stop protects more than a warning, but it will sometimes stop a message that was fine, and no line between mild and hurtful will be right every time. We accepted that cost for the hurtful level, because a wrongly stopped parent loses a few seconds, while a hurtful message that gets through can end the thread. Two things keep the cost low: the nudge never blocks, and the dispute "Not hurtful? Tell us" gives the parent a way out.
User Interface
A quiet interface, so the tone check reads as a moment and not an alarm.
Kids First needed a visual language calm enough for a high-stress context: warm, not clinical, and quiet enough that the tone check stands out when it matters. Color is reserved for it. The nudge uses a softer tone, and the stop a stronger one. Everywhere else the palette stays neutral.
When a message is stopped, the banner names the words, the send button is disabled, and the rewrites sit directly under it, so the parent never lands on a dead end.
An AI feature is a decision about how much to step in. Detecting tone wasn’t the hard part. The hard part was deciding where a nudge ends and a stop begins and accepting that the line will sometimes be wrong.
Design the failures, not only the happy path. A model will be wrong, so the dispute, the explanation, and “Edit it myself” mattered as much as the rewrites.
Small decisions carry trust. Button copy and layout, like “Use this” and “Send anyway," decide whether a parent feels helped or policed.
Details cost time. I over-detailed some screens early on, which slowed the team. Designing the whole story first, with a shared design system, kept us moving.
What I’d do next. Test where the line between mild and hurtful sits with real parents, and measure what happens after a nudge and after a stop.