A user opens your mini-program, taps allow on the notification prompt because it's the only way to finish checkout, and within a day you've sent three promos. By the fourth, the toggle is off. We've watched this loop enough times to stop blaming the copy and start blaming the model.
Most teams treat push like a free broadcast channel. You install the SDK, request the token, and suddenly you can reach a person anywhere, anytime, on the device they sleep next to. So you do. And then the operating system, the host app, and the user all quietly build walls — notification summaries, scheduled delivery, one-tap mute — because the channel only works when it's scarce.
The mute button is faster than your funnel
In a mini-program the permission prompt isn't a separate OS step the way it is on iOS. It's wrapped inside the host app, and the host has its own rules about when a push even gets delivered. Some hosts batch notifications and only show the one you'd have sent at 3am after the user wakes up. Others throttle how many a mini-program can send per day. That's not friction to be defeated; it's the host protecting its own relationship with the user, and you're borrowing it.
So the first mistake isn't a bad headline. It's assuming the channel is yours. It isn't. You're a guest in someone else's notification tray, and guests who talk too much get shown the door.
Permission isn't consent
We'd argue the biggest myth in this space is that an accepted prompt equals a relationship. It doesn't. Opt-in is a loan, not a bank account. The user lent you attention on the assumption you'd be useful, and the interest rate is brutal: one irrelevant ping and the trust drops, two and they're reaching for settings.
The teams that keep their delivery rates up don't send more. They send fewer things the user would have opened anyway — a ride that's arrived, a package that's out for delivery, a cart someone abandoned while genuinely intending to come back. Transactional and clearly-wanted beats promotional every single time, and it's not close.
Timing beats volume
There's a temptation to send at the moment you want the conversion — end of quarter, flash sale, Tuesday because the dashboard says Tuesday. Users don't live in your dashboard. A reminder about a dinner reservation sent at 9am is noise; sent at 4pm it's a small favor. We've seen open rates triple on the same message just by moving it from a batch window to a moment that fit the user's day.
The catch is that the user's day differs across markets, across hosts, and across time zones you didn't know you had. A universal runtime helps here precisely because it can read the local clock and quiet hours from one place instead of five.
The payload is the product
What's inside the push matters as much as whether it arrives. A notification that says "Check your app for an update" is a chore wrapped in a notification. One that says "Your refund of $12.40 was approved" is a small piece of resolution delivered to the lock screen. The best pushes don't ask the user to do anything — they complete something the user already cared about.
Keep the payload honest about intent. If it's a promo, say it's a promo; don't dress a discount as a system alert. People learn to ignore the wolf-cries fast, and when you finally have something real, the tray is already muted.
| Failure | Who pays | What actually fixes it |
|---|---|---|
| Prompt shown at first launch | The user who never opted in | Ask after value, not before it |
| Three promos in a day | Your delivery rate, next week | A weekly cap users can see and adjust |
| Sent at 3am local time | The person who mutes you | Honor quiet hours from the runtime |
| "Tap to see what's new" | Everyone, instantly ignored | State the concrete thing in the payload |
| Promo dressed as alert | Trust, permanently | Label intent honestly in the message |
| No way to dial it down | The frustrated, who go nuclear | Per-topic controls, not one master switch |
The honest tradeoff
None of this is free. A tighter send policy means fewer touches, and fewer touches means the growth team loses a lever they were comfortable pulling. Per-topic controls are real engineering, and quiet-hour logic is fiddly across regions. There's a real argument that a young mini-program should be a little loud while it learns what its users want. We won't pretend the restraint is obvious on day one.
But the cost of being muted is worse than the cost of being quiet. A muted channel is a channel you already paid to build and can no longer use. The teams that win this aren't the ones with the cleverest copy. They're the ones who treated the notification as a promise and kept it.
One runtime, many hosts
Here's the part that's easy to miss. Each host app implements notifications differently — different prompts, different caps, different quiet-hour behavior. If you hard-code your push logic to one host, you relearn all of it on the next. A runtime that targets several hosts can normalize permission, scheduling and quiet hours into one API, so the same restraint applies whether you land in a messaging app or a payments app. You write the discipline once; it carries across.
We're not saying build less. We're saying build the right thing once, and let it travel.
Sources
- web.dev — Notifications — practical guidance on permission and delivery best practices.
- MDN — Notifications API — how the underlying notification mechanism actually works.
- Think with Google — Consumer behavior — how people react to mobile messaging in practice.
- StatCounter — Platform share — why host and OS differences matter across markets.
- Google Developers — Web Fundamentals — background on push and notification fundamentals.