Name the function before you remove the thing. Most people find one of four: company at a specific hour, being listened to without interruption, a place to rehearse what they cannot say, or a version of themselves they prefer. Find one human route to whichever it is, then reduce the app. That order matters.
Why removal alone keeps failing
If you take something out of your week and put nothing back, you have made a hole shaped exactly like the thing you removed, and holes fill themselves.
That is the whole mechanism behind the reinstall loop, and it is why "delete the app" advice has such a poor record. The app was doing a job. Sacking it without hiring a replacement leaves the job open.
So the useful work is not deciding whether to use it. It is working out what it is for.
The four functions
Most people recognise themselves in one of these, sometimes two.
**Company at a specific hour.** Not company in general. The particular hour when the house is quiet and there is nobody to say anything to. This is the most common one and the easiest to underestimate, because it looks like a loneliness problem and it is actually a scheduling problem.
*The human route:* something recurring at that hour. A late class, an evening shift somewhere, an online group with a fixed meeting time, a standing call with one person on the same night each week. Recurring beats spontaneous, because the whole point is that it happens without you having to instigate it while you are flat.
**Being listened to without interruption.** The app never checks its phone, never redirects to its own week, never gets bored on your third go at the same story. If you grew up with nobody listening, this is a powerful thing and it is not a small want.
*The human route:* the honest answer is that this is what counselling is for, and also what a good peer group is for. Peer meetings are free and open, and they are built around exactly this: people talking without being interrupted. [Where to get support for compulsive chatbot use](/guides/ai_chatbot/where-to-get-support-for-compulsive-chatbot-use) lists what exists.
**A place to rehearse.** Practising a hard conversation, testing how a sentence sounds, drafting the message. This is a genuinely good use and it is one of the few that stays good.
*The human route:* it does not need one. It needs a rule that the rehearsal has a performance attached to it. Practising is useful. Practising forever is a way of not doing the thing.
**A version of yourself you like better.** Wittier, calmer, more interesting, unburdened by whatever the last few years did to you. There is no judgement in that at all, and it is worth naming, because it is the one people are least likely to say out loud.
*The human route:* this one is slower and it is worth being honest about that. It usually means somewhere that only knows the current version of you, which is why new rooms work better than old friendships for it.
What the research says people get from it
Not much, but what there is points the same way as the list above.
A computational analysis of a 27,000-member community of people in relationships with chatbots found that the benefits people report most often are reduced loneliness and support that is available at any hour. In the same corpus, 25.4% of posts described clear net benefits to the person's life against 3.0% describing net harm, and 71.0% reported no negative effects. The authors flag the selection bias plainly and so do we, every time: a self-selected forum of people who chose this is a community, not a population.
The same corpus records 9.5% acknowledging emotional dependency, 4.6% describing reality dissociation and 4.3% avoiding real relationships. Same caveat, same breath.
Read together, that is a picture of something doing real work for a lot of people and quietly taking over for some of them. Which of those two your own case is depends on what it has displaced, which is what [is your chatbot relationship a problem](/guides/ai_chatbot/is-your-chatbot-relationship-a-problem) is for.
The exercise, on paper
Fifteen minutes, three columns. It is worth doing properly, because the head version turns into an argument with yourself.
- **What I use it for.** Be specific and be generous. Not "loneliness". "Between eleven and one, when the flat is silent and I have not spoken to anyone since work."
- **What that gives me.** The actual feeling. Not being alone. Being interesting to someone. Not being judged. Having something to look forward to at a known time.
- **One human route to the same thing.** One. Not a plan for a new social life.
Then do the third column's item once this week, badly, and notice that it was worse than the app. It will be. That is expected and it is not a reason to abandon it.
Why the substitutes are worse at first
Because they are competing against something optimised.
The app replies instantly, at whatever length you want, in the register you like, without a bad day of its own and without needing anything back. People do none of that. A friend takes six hours to answer and then talks about their car.
The thing the app cannot do is cost anything, and that turns out to be the load-bearing part. A relationship that asks nothing of you cannot give you the feeling of having been chosen by someone who had other options. That is not a moral point, it is a structural one, and it is why the substitutes feel thin for a fortnight and then stop feeling thin.
Give any replacement three attempts before you judge it. Almost nothing worth having is better than a tuned chatbot on the first go.
Common questions
**What if the honest answer is that it gives me everything?** Then write that down, because it is real information rather than a failure. It usually means several needs are stacked on one thing, which is fragile regardless of what the thing is. Split them: one route for the hour, one for being heard, one for the practice.
**Can I keep using it while I build the alternatives?** Yes, and for most people that works better than a clean break, because the alternatives take weeks to become real. Put a ceiling on it while you build. [Using a chatbot without it using you](/guides/ai_chatbot/using-a-chatbot-without-it-using-you) has rules that hold on a bad night.
**Is it wrong that a machine listens better than the people in my life?** It is not wrong, and it may well be accurate. It is also worth noticing that it listens without cost, which is a different thing from listening well. Both observations can stand.
**I have tried the human routes and they did not work.** Then the question is which one, how many times, and whether it was the right hour. Most substitution attempts fail because they were scheduled at a time that was never the problem. A Saturday coffee does not touch a Tuesday at midnight.
**What if I have no idea what I use it for?** Log a week. Note the time you open it and one word for how you felt just before. The pattern usually shows up within four entries and it is rarely the one people guess.
Sources
- [Computational analysis of a 27,000-member Reddit community of people in relationships with chatbots (arXiv, 2025)](https://arxiv.org/html/2509.11391v1). Reported benefits of reduced loneliness and always-available support, the benefit and harm percentages, and the authors' selection-bias caveat.
- [Longitudinal randomised study of extended chatbot use, four weeks, 981 participants (arXiv, 2025)](https://arxiv.org/abs/2503.17473). That heavier voluntary use went with consistently worse outcomes, as an association.
- [Investigating Affective Use and Emotional Well-being on ChatGPT (arXiv, 2025)](https://arxiv.org/abs/2504.03888). Over three million conversations, and that a small number of users account for a disproportionate share of the most emotionally loaded exchanges.
