Keyword Alerts

Most parents start by writing far too many words. Forty terms covering drugs, swearing, dating, gaming and everything they half remember from a news article. That list generates alerts every day, the alerts become background noise within a fortnight, and the one that mattered arrives to a parent who has already stopped looking. A list of eight to twelve words that you would genuinely act on is worth more than a list of fifty you will learn to swipe away.

Four categories cover almost everything worth watching. The first is self harm and crisis language, which is the reason many families install a tool like this at all. The second is bullying and threat: the words a child would be called, or would be told to do to themselves, by someone trying to hurt them. The third is secrecy and meeting: the vocabulary of an adult persuading a child to keep something quiet or to come somewhere alone. The fourth is money, because requests for cash, gift cards or transfers are the fingerprint of both scams and sextortion.

Leave swearing off the list. It is the single most common mistake we see. Teenagers swear, in volume, and putting profanity on a watch list guarantees a stream of alerts that tell you nothing about safety while training you to ignore the ones that do. The same goes for ordinary slang and for brand names of things you merely disapprove of.

Write the words the way people actually type them. Add the abbreviation as well as the full phrase, add the common misspelling, and remember that matching is case insensitive so there is no need to add capitalised variants. Review the list every few months, because both the risks and the vocabulary change as a child gets older.

An alert reads roughly like this: the word “hurt myself” matched, in a messaging app, at 22:14 last night. That is the whole payload. You are not sent the notification text, the message body, the name or handle of whoever sent it, the surrounding sentence, or a screenshot. None of that is collected, so none of it can be delivered to you, leaked in a breach, or handed to anybody who asks.

This is a deliberate design decision rather than a technical shortcut. The purpose of an alert is to tell you that a conversation may need to happen, and the three facts do that job completely. Adding the message content would change the product into something that reads private conversations, which is exactly the thing we have built it not to do, and it would give a child a genuine reason to work around the tool.

It also changes what you can do with an alert, and that is a feature. You cannot confront someone with a quote, because you do not have one. You can only go and ask. In our experience that constraint produces better conversations, because it removes the temptation to arrive already convinced and lets your child explain something you may well have misread.

Alerts group in the same way the rest of the app groups things. Four matches on the same word within a short window arrive as one line with a count, and repeated matches from one app do not turn into a stream of separate buzzes. Everything is kept in the alert history whether or not it notified you, so a category you have muted still builds a record you can read on purpose.

Matching runs on notification previews only. When a message arrives, Android may show a short line of text on the lock screen or in the shade, and that line is what is compared against your list. It is never the conversation itself. The app is not opened, the chat is not read, and nothing inside an encrypted messenger is accessible to us or to anybody outside the conversation.

That has consequences you should know before you rely on it. If your child has previews turned off, if an app shows only “New message”, or if the preview truncates before your word appears, nothing will match. Some apps deliberately hide content in notifications. A conversation held entirely in a video call, a voice note, or an image will produce no text to match at all.

So treat keyword alerts as one layer among several rather than as a net. They will catch a word sitting in plain sight on a lock screen, which is more often than you might think, and they will catch nothing at all in a determined conversation that has moved somewhere quiet. Pair them with the app list, so you notice a new messaging app appearing, and with a household habit of actually talking.

A false positive is a match that is technically correct and completely uninteresting. A song title, a game mechanic, a group chat quoting a film, a news headline in a headlines app. They are unavoidable, because a word list has no idea what a sentence means, and they are the main reason people abandon this feature.

Deal with them one at a time as they arrive. If a word fires three times in a week and every match was harmless, either delete it or make it more specific. “kill” is hopeless in a house with a teenager who plays games. “kill myself” is a different word entirely, and it will sit quietly for months. Longer phrases are almost always better than single words.

You can also exclude an app. If a news reader or a football app is producing every match, muting that app for keyword purposes removes the noise without weakening the list anywhere it matters. And you can mute a specific match if it is a running joke in a group chat, which stops the same harmless line arriving every evening.

What you should not do is turn the whole feature off in frustration. A list that produces one alert a month is doing its job, and the quiet is the point. If your list is producing several a day after tuning, it is still too broad, and shortening it will make it more useful rather than less.

The child app already discloses that keyword matching is active, and that only the word, the app and the time are ever reported. So the question is not whether your child will find out; it is whether they hear it from you first or read it on a settings screen and wonder what else they have not been told.

How much detail to share depends on age. With a younger child, reading the list aloud is often the best approach: these are the words that would make me worried, and if one comes up I will ask you about it. With a teenager, the categories usually work better than the exact terms, partly because a precise list invites creative spelling and partly because the categories are the honest description anyway.

Frame it as a smoke alarm and not a listening device. A smoke alarm does not know what you are cooking, it does not report your dinner to anybody, and nobody in the house resents it, because everyone understands what it does and why it is there. Keyword alerts should occupy exactly the same emotional space in a home.

And say what you will do with a match. “If one of these comes up I will ask you about it, that same evening, and I am not going to be able to read your messages” is a promise you can actually keep, because the product makes it impossible to break. Being able to make a promise like that is worth more than any amount of extra data.