Your App Blocker Has No Idea ChatGPT Exists

On September 14, Apple shipped its new child safety features in iOS 27. The headline one is Time Allowances, which caps how long you can spend inside a whole app category instead of making you set a limit app by app. Apple built the age-tailored defaults with the American Academy of Pediatrics. It is the best screen time feature the company has ever released and we are not being sarcastic about that.

The categories are Social Media, Entertainment and Games.

Go and open Screen Time on your phone right now and look at the list. This takes about thirty seconds and it makes the rest of this post obvious.

There is no category for a chatbot. ChatGPT sits under Productivity. Most AI assistants land in Productivity or Utilities. Whatever cap you set, on whichever of those three buckets, the thing you may actually be losing your evening to is not in any of them.

We all agreed the enemy was the feed

That is the real story here, and Apple is not the villain of it. Nobody in this category has a good answer, us included. We spent a decade building defences around infinite scroll because infinite scroll was the problem. Blocklists, timers, pause screens, weekly reports. All of it points at TikTok and Instagram, because that is where the hours went.

While we were doing that, the thing that eats the evening started calling itself a productivity tool.

Here is the trick, and it is a good one. Social media at least had the decency to look like leisure. You could catch yourself at it. There was a small flicker of "what am I doing" around minute forty, and that flicker is the thing every blocker on the market quietly depends on. A chatbot never gives you that flicker. It hands you a small sense of accomplishment every few minutes and files the whole session under work.

Nobody feels guilty about forty minutes with a chatbot. That is the whole mechanism.

Researchers went and looked at what people actually say

This is not us speculating. A paper at CHI 2026 out of the Socius lab at UBC, The AI Genie Phenomenon and Three Types of AI Chatbot Addiction, ran a thematic analysis of 334 Reddit posts across 14 subreddits where people described compulsive chatbot use in their own words.

The researchers found three recognisable patterns. Escapist Roleplay, where the conversation becomes somewhere to go instead of here. Pseudosocial Companion, where the chatbot occupies the slot a person used to. And Epistemic Rabbit Hole, which is the one our readers will recognise in themselves, because it is the two hours that felt like research.

The symptoms people described map onto the addiction literature. Inability to stop thinking about it. Real distress when trying to quit. Work and relationships taking the hit.

One honest caveat, because we would rather say it than have you find it. This is qualitative work on a self-selected sample of people who chose to post about struggling. The authors make no claim about how common any of this is, and AI addiction is not a clinical diagnosis. Nobody has a percentage and you should distrust anyone who gives you one. What the paper does establish is that the behaviour is real, it has shape, and people are describing it using the exact vocabulary we spent ten years building for slot machines. That vocabulary is now being used about a text box.

A fourth category would not save you either

We have written before about how Apple's category system leaks, and we covered what was coming in iOS 27 back in June, so we will not relitigate either here. The short version is that any bucket somebody else drew will be slightly wrong for you, and slightly wrong is exactly the confidence level at which you stop checking.

But the deeper problem is not taxonomy. Even if Apple added an AI category tomorrow, you would have to decide to cap it, and deciding to cap it requires believing you have a problem with something that has spent every session making you feel productive. Our defences assume the user already feels vaguely guilty. This one removes the guilt, which was the part doing the actual work.

What actually helps

Not a new category. The thing that works is boring and it has not changed.

Pick the specific apps that eat your evening, by name, based on what you know about yourself rather than what bucket someone else put them in. Then commit to the limit at a moment when you do not want to break it, which is never the moment you want to break it. That gap between deciding and wanting is the entire game.

That is what BreakOff does, and yes, that is the pitch. It is also just what the evidence says. If your blocker only knows about categories, it does not know about this yet.

One last thing, and we will own our share of it. We use these tools every day. We wrote some of this post with one open. The argument is not that AI is bad, because it is not, and this is not a piece about putting the genie back. The argument is that "useful" and "habit-forming" were never opposites, and right now every defence we have built as an industry only covers one of those. The reason you are not hearing this from more people in our corner of software is that almost all of us are currently building on top of it.

Sources: Apple Newsroom: Apple's new child safety features now available; The AI Genie Phenomenon and Three Types of AI Chatbot Addiction, CHI 2026 (arXiv preprint); ACM Digital Library: version of record; UBC Socius lab: author copy; PsyPost: Researchers identify three distinct faces of AI chatbot addiction; TechXplore: Are you addicted to your AI chatbot? It might be by design; Tech Policy Press: AI chatbots and addiction, what does the research say. Disclaimer: This article is for informational purposes only and does not constitute medical, psychological, or professional health advice. If you have concerns about mental health, attention, or device usage, please consult a qualified professional. BreakOff is a productivity and wellness tool, not a medical device.