Push Notifications: Why Yours Are Being Ignored, and What Fixes It
Push is the cheapest channel an app has. You don’t pay per message, there’s no ad spend, and it lands right on the lock screen.
That’s also the problem. Because it costs nothing to send, it gets overused, and a lot of apps wear the channel out within a few months of launch.
We’ve built more than 50 Android and iOS apps at Color Leaves, and the setup itself has never been where things go wrong. Firebase Cloud Messaging on Android, APNs on iOS, save the device token, send a payload. A decent developer gets that running in a day. The trouble starts with the decisions around it: when you ask for permission, what you send, how often you send it, and what you count as success. Get those wrong and push doesn’t just underperform. It drives people out of your app faster.

You only get one real shot at permission
iOS shows its permission dialog once. If the user taps “Don’t Allow”, the only way back is through Settings, and very few people go there on their own.
Android used to be the easier platform here. Since Android 13, it isn’t. Apps now have to ask for notification permission at runtime, and if the user says no twice, Android stops showing the prompt.
Still, plenty of apps throw the dialog up on first launch, before the user has done anything. It’s the worst time to ask. They don’t know the app yet, they haven’t got anything out of it, and they have no reason to agree.
Ask when the value is obvious instead. Right after someone places an order, offer to tell them when it ships. After they save a search, offer to let them know when something matches. After they follow someone, offer updates on that person’s activity.
You can also ask in two steps. First show your own screen explaining what you’ll send and why. If they say no there, the system dialog is still unused and you can try again another day. If they say yes, the system prompt goes to someone who has already agreed, and far more of them accept it.
iOS has another option a lot of teams don’t use: provisional authorization. From iOS 12 onwards, an app can send notifications quietly to Notification Centre without asking first. The user sees real notifications and decides from there whether to keep them. If your app’s value takes a few days to show, this often works better than asking upfront.
It’s worth the effort. Airship’s benchmark studies have shown for years that users who opt in to push stay at roughly twice the rate of those who don’t. Some of that is selection bias, because people who opt in were probably more engaged to begin with. Even so, the gap is too big to treat the opt-in screen as an afterthought.
Transactional and promotional are not the same thing
Nearly every notification is one of two types, and they need very different handling.
Transactional notifications are about something the user did, or something happening to their stuff. The order shipped. The payment failed. Someone replied. The cab is two minutes away.
People want these. They’re the reason most users leave push turned on in the first place. Send them quickly and make them precise.
Promotional notifications are about something you want the user to do. A sale, a new feature, a reminder to come back.
A few of these are fine. Too many and people get annoyed. Each one uses up a bit of the goodwill that keeps notifications switched on.
The mistake most apps make is running both through one system with a single on/off setting. So the user who wants delivery updates but not sale alerts has one choice: switch everything off. Keep the two separate, in your code and in your settings screen. Give people a way to mute marketing and most of them will leave transactional notifications on. Don’t, and they’ll turn off the lot.
What gets a notification opened
A few things hold up across almost every app.
Be specific. “Your order from Sharma Traders is out for delivery” does far better than “You have an update”. The first one tells the user whether it matters before they even unlock the phone.
Tie it to something they did. Notifications triggered by a user’s own action do much better than broadcasts. A price drop on a product they looked at beats a store-wide sale.
Send it at a sensible time. A sale alert at 11pm is an intrusion. A delivery update at 11pm is useful. Look at the user’s local time and ask whether the message is actually urgent. If it isn’t, hold it till morning.
Don’t mislead. A clickbait notification works once. The next one gets ignored, and the one after that gets notifications switched off.
Make the tap go somewhere. If the notification is about an order, tapping it should open that order, not the home screen. Landing people on a generic screen after promising them something specific is one of the most common bugs out there. It keeps happening because teams test the send far more carefully than the deep link.
Set up the routing properly while you’re at it. Android App Links and iOS Universal Links, verified through your assetlinks.json and apple-app-site-association files, let one URL work from a notification, an email or the browser. If the user doesn’t have the app yet, deferred deep linking makes sure they still land in the right place after installing it.
How often is too often
There’s no single right number, but a simple rule helps. Send transactional notifications whenever something actually happens. Be stingy with promotional ones.
For most consumer apps, more than two or three promotional pushes a week is too many. And if a user hasn’t opened the app in a month, sending more is not the fix. More messages to someone who’s already drifting away usually just gets you uninstalled sooner.
Build a frequency cap into your system. If a user triggers five notification-worthy events in an hour, they shouldn’t get five notifications. Group them, or drop the less important ones.
On iOS 15 and later, you can mark a notification as time-sensitive so it gets through Focus modes. Keep that for things that genuinely can’t wait. Overuse it and users can turn it off for your app.
Measure what actually matters
Most teams look at open rate and stop there. Open rate is easy to inflate and easy to misread.
Ask better questions. What did people do after they opened it? Did the notification get you the result you wanted, or just a tap? And what happened to opt-outs and uninstalls in the days after the campaign went out?
Track opt-outs for each campaign. If one type of notification causes a jump in people turning push off, it cost you more than it earned, however good the open rate looks. This is the number that tells you whether you’re using up the channel or building it.
Look at delivered numbers too, not just sent. On Android, some phone makers kill background processes aggressively to save battery, and that stops notifications from arriving. Sent and delivered can be quite far apart. If you only look at sends, you’ll blame the gap on users not caring when the real problem is delivery.
Bringing back users who are drifting away
If you use push to win back inactive users, when you send matters more than what you send.
You don’t get long. A user who stopped opening the app a few days ago still remembers why they installed it. A user who stopped two weeks ago mostly doesn’t.
So set your inactivity trigger to fire within the first week, not after a fortnight. Leave it too late and your notification shows up like a message from a stranger, which is usually when people go and disable push altogether.
The engineering bits people skip
Test with the app killed. Notifications behave differently when the app is open, in the background, or fully closed. The fully closed case gets tested the least and matters the most.
Handle token refresh. Push tokens change over time. If you only register the token at install, you’ll slowly lose the ability to reach a chunk of your users without noticing.
Clear out stale tokens. Sending tokens wastes quota and makes your delivery numbers look worse than they are.
Use Android notification channels properly. Since Android 8, users can switch off individual channels instead of all notifications. That’s great if your channels are set up sensibly, and useless if everything sits in one default channel. Give each channel the right importance from the start. Once a channel exists, its importance is controlled by the user, and changing it means creating a new channel ID.
Know the difference between notification and data payloads. With a notification payload, the system shows the message while your app is in the background, and your code never runs. A data payload wakes your app up to handle it. Mixing the two causes the familiar bug where notifications look fine in the foreground and behave oddly in the background. Silent pushes have their own rules and get throttled on both platforms, so don’t use them for anything time-critical.
Keep payloads small and clean. APNs and FCM both limit standard payloads to 4KB, so send an ID and fetch the details inside the app. Notification content also passes through Apple’s and Google’s servers, so keep sensitive data out of it.
None of this is a one-time fix. Tokens rotate, new OS versions change how notifications behave, and phone makers keep adding their own battery rules. That’s why these checks belong in your regular app maintenance routine, not just the launch checklist.
A better way to think about push
Think of push permission as a balance in an account, not a switch that’s either on or off.
Useful transactional notifications add to the balance. Promotional ones take from it. When the balance hits zero, the user turns you off, and there’s very little chance they’ll turn you back on.
Teams that get this send fewer notifications and get more out of them, because a year later their users still have push enabled. Teams that treat push as free broadcasting run through the balance in a few months, then can’t work out why the audience they can reach keeps shrinking.