The first thing to understand about chat as a support channel is that the person on the other end has already decided your line is worth a risk. She's typed to an unknown number from a phone that someone else might check. Your job as a designer is to make that risk as small as possible.
Voice engineers optimise for connect time. Chat engineers for a survivor-support line optimise for something else entirely: forensic silence. Nothing about the interaction should be visible when the phone is closed.
The design constraints
1. No notifications.
The default WhatsApp notification tone is fine for a food delivery. It's a disaster when a partner is in the kitchen. Every reply from the service must arrive silent. The Business API supports this — you just have to build for it from the start.
2. No read receipts, no last-seen, no typing indicator.
All three of these can be observed by someone who has access to the same phone. All three must be off by default. Not opt-in — off from message one.
3. An abandon word.
The user needs a way to instantly clear the conversation and replace it with something benign. We use "recipe" as the default and let the client pick their own. When she types it, the flow returns a recipe (or a weather forecast, or a bus schedule — pick a genre that survives a glance). She can show the phone to her partner and the conversation looks like a food chat.
4. Handoff to humans, fast.
Automation triages severity. It does not counsel. In a crisis line, the survivor knows within two exchanges whether she's talking to a person, and if she's not, she stops. Design for a four-minute human handoff. Any longer is a bounce.
5. Data doesn't touch your servers.
Everything you build sits on the nonprofit's own infrastructure. Not yours. Not a third party's. Consent forms explicitly. And nothing about the conversation is ever readable from WhatsApp's metadata — no chat summaries in previews, no first-line notifications when the phone is locked.
What chat is bad for
Chat is not universally better than voice. In an active emergency — someone locked in a room, someone with a weapon in the house — voice with an on-call police liaison is faster and more legally protectable. A good design offers both, and gives the user a two-tap way to escalate a chat to a call.
Why this is a marketing agency's problem
It probably shouldn't be, honestly. This is design work that belongs with the programme team, and if the programme team can afford a specialist product designer, they should hire one. Most can't.
The people we most want to reach through crisis lines live in the last mile, and the last mile is where our work already lives. So when a survivor-support nonprofit asks us to help with WhatsApp, we help. But we always ask them the same question first: is the programme team ready to counsel by chat, in real time, at scale? If the answer is no, the interface doesn't matter. Build the counselling capacity first.