Notifications and reminders: when an alert is worth it

A notification is worth the interruption when it brings something relevant, at the moment the person needs it, and makes sense without opening the app. Anything else becomes noise and pushes people to turn everything off. Nielsen Norman Group, NN/g, lists the mistakes that drive people away. The main ones are asking for permission on first launch, sending in bursts, sending what doesn’t matter and hiding the way to switch off.

What counts as a notification

A notification is a message triggered by a system event, not by what the person just did. NN/g on indicators, validations and notifications splits them into two types. An action-required one should interrupt, because something depends on the person. A passive one only informs and can go unnoticed.

Error messages are a different thing. Apple’s guidelines call for an alert, not a notification, to show an error. Mixing the two confuses people who already know both formats.

When an alert is worth the interruption

Every notification competes for attention with many others. A Telefonica Research study, cited by NN/g, counted an average of 56 phone notifications a day in 2016. Each one that isn’t useful spends the patience the next one will need.

The most useful test is to separate transactional alerts from marketing ones. A transactional alert is about something the person already has: an order, a booking, an account, a payment. NN/g on transactional notifications shows that people expect SMS and push only for urgent, important content. Promotions sent that way tend to annoy, especially by SMS.

Timing matters too. In the same study, a delivery notice that arrived 30 minutes after the package was a letdown. An automated customer-service reply 30 minutes after an email complaint, on the other hand, felt fine (NN/g).

Asking for permission at the right time

Asking on first launch is the most common mistake. In NN/g’s studies, the instant reaction is often to decline, because people don’t know yet what they’ll get. The advice is to let them use the app first and ask in a later session, saying what will arrive.

Android brought the same idea into the platform. Since Android 13 (API level 33), apps have to request notification permission at runtime. The Android documentation recommends waiting until people know the app and asking in context, like when they tap an alert bell.

What to write

The title says what the message is for. “Rate your trip” works. The brand name in the title wastes the space, because the system already shows who sent it (NN/g). Apple’s guidelines ask for the same thing: a short title, without the app name.

The body has to be short and complete at once. NN/g notes that push notifications show 50 to 240 characters. The essentials go first, because people read the first few words and decide. An order number that doesn’t say what was bought sends them hunting for the rest.

The ideal alert is one that settles things on its own. In the study cited by NN/g on mobile microsessions, about 60% of short phone sessions involved reading or acting on a notification. When people tap it, they should land on the screen for that topic, not on the home page.

Leave sensitive data out. Apple points out that you can’t know who’s looking at the screen when the alert shows up.

Push, SMS or messaging

NN/g splits the channels by use:

SMS for what’s urgent, important or will be checked later: booking reminders, pickup codes, time changes, transaction confirmations.
Push for what isn’t pressing, like shipping status or a nudge to come back to the app.

SMS doesn’t depend on a data connection and doesn’t vanish with a single tap. That’s why it can hold information people will need later. The same article notes that culture shapes the channel: where the phone number is your identity, talking to a business happens in a messaging app.

Bursts, repeats and an easy way out

Several alerts in a row feel like a doorbell that won’t stop. NN/g recommends combining them into one message when you have more than five to send at once. Apple goes further: don’t send several notifications about the same thing, even if the person hasn’t responded.

Turning them off has to be easy. NN/g puts that setting inside the app, in the settings area, without forcing people into the system settings. With SMS, every message carries a way to stop receiving them (NN/g). Hiding that exit costs trust and can cost you the app.

Appointment reminders: what research measures

Reminders are the case where alerts have a measured effect. A Cochrane systematic review pooled eight randomized trials with 6,615 participants. Attendance at appointments was 67.8% with no reminder, 78.6% with a text message and 80.3% with a phone call.

The same review found texts and calls had a similar effect. In two studies, the cost per attendance of a text was 55% and 65% lower than a call. The authors rate the evidence as low to moderate quality.

Two limits matter. The review measures healthcare appointments, so applying it to a salon or a repair shop is an extrapolation. And it asks future studies to report the message content and its timing relative to the appointment. In other words, research doesn’t yet say when the best moment to remind is.

In practice, with my assistant

On Vexor, my AI product for WhatsApp, the assistant confirms every booked slot and reminds the client 72 hours, 24 hours and 2 hours before. I installed the assistant at each client and customized those three reminders along with the scheduling flow.

The summaries follow the logic of bundling instead of drip-feeding. The day, week and month summary comes as a PDF or spreadsheet for whoever wants to follow along. The week’s agenda arrives when the business owner asks for it on their own WhatsApp.

Where the sources disagree

Apple’s rule against several notifications about the same thing runs straight into a series of reminders. And it holds even when the person hasn’t responded. The case for a series is that each alert arrives at a different distance from the slot and can serve a different decision, like rescheduling with time to spare or leaving home. But the Cochrane review doesn’t measure how many alerts is the right amount. The number is a design choice, not a research finding.

There’s tension in the channel too. NN/g prefers SMS for booking reminders but warns that frequent SMS feels more invasive than push. The channel that holds information best is also the one that costs the most when overused.

The common mistake

The most frequent slip is treating notifications as a free marketing channel. The team sends everything, people turn everything off, and the reminder that mattered stops getting through. Every alert should answer one question before it goes out: does this help the person now, or does it only help the company?