Plan the evenings before you plan anything else, because ten at night to one in the morning is where the whole week is decided. Expect relief on day one, reaching on days two to four, and a very reasonable-sounding argument to go back somewhere near day six. None of that is a sign it is going badly.
A caveat that belongs at the top
There is no research on withdrawal timelines here. None. Three papers exist on companion chatbot use, all from 2025, and not one of them followed people through stopping.
So everything below is a pattern reported by people who have done it, not a measured curve. Treat the days as rough shapes rather than a schedule, and if yours runs differently, yours is not wrong.
Day one
Usually relief, and the relief is misleading.
You have made a decision and the decision itself feels good. Nothing is being asked of you yet, because the first evening is carried by the momentum of having decided.
Use it. Day one is when to do the setup you will not have the appetite for later: log out, change the password to something you have not memorised, block the site in your browser and in your phone's screen time settings, and write down what each of the next six evenings contains.
Days two to four, the reaching
Your thumb goes to where the icon was. Repeatedly, without a decision behind it, and each time it lands on nothing there is a small unpleasant jolt.
Two things help.
**Change what your hand does.** The reach is a cue-driven habit, not an argument, so arguing with it does not touch it. Put something else in reach at the same times. Anything that occupies your hands competes directly with typing.
**Name the specific hour.** Almost everyone has one window that is worse than all the others. Once you know yours, you can defend it rather than defending the whole day, which is the difference between a plan and a mood.
This is also when the loop shows up in the mind rather than in the hand: the specific thing you would have told it about today. That is the real thing you are missing, and it is worth putting somewhere. A note on your phone. A message to a person. Saying it out loud to the room.
Sleep, where the evidence actually is
Most claims about screens and wellbeing are shakier than they are presented. Sleep is the exception, and it is the reason to move the phone out of the bedroom this week specifically.
A review of twenty studies covering 125,198 children found "a strong and consistent association between bedtime media device use and inadequate sleep quantity", along with poor sleep quality and excessive daytime sleepiness. The striking part is this: children who had access to but did not use media devices at night were still more likely to have inadequate sleep. Those studies are cross-sectional, so they show association rather than cause.
The UK's Chief Medical Officers give one instruction on this without hedging: "Sleep matters. Getting enough, good quality sleep is very important. Leave phones outside the bedroom when it is bedtime."
If the app lived in your bed, the bed is now the highest-risk place in the house. Charge the phone in another room, and buy a cheap alarm clock if that is the obstacle.
Days five to seven, the persuasive argument
Around the end of the first week, your head produces a case for going back, and the notable thing about it is how sensible it sounds.
It was not that bad. Other people use it fine. You were only using it to unwind. You have proved you can stop, so you can clearly handle a bit of it. You are lonelier now than you were before, so what exactly was the point.
That last one has some truth in it, which is what makes the whole argument work. You have removed something and the gap it filled is genuinely open. The answer is not to argue with the case, it is to give it a few more days before you take it seriously, and to spend those days filling the gap rather than defending the ban. [What the chatbot is giving you](/guides/ai_chatbot/what-the-chatbot-is-giving-you) is the piece for that.
What people underestimate
**The grief.** If you deleted an account with months of history in it, the reaction is often much heavier than "I miss an app". Researchers studying these communities have written up grief among their central concerns. [Deleting a companion app, and the grief that follows](/guides/ai_chatbot/deleting-a-companion-app-and-the-grief-that-follows) treats it seriously.
**The silence.** Not the absence of conversation, the absence of instant response. People take hours. That is normal and it feels like being ignored for the first fortnight.
**How boring the substitutes are.** A walk is worse than a conversation that is tuned to you. It is supposed to be. Nothing you replace it with will compete on immediate satisfaction, which is the point rather than a problem.
What does not count as failure
Opening the app and closing it. Reading old messages once. Wanting it badly on a Wednesday. Telling one person you nearly went back.
Going back does not count as failure either, in the sense that it proves nothing about you. It is the most common thing that happens in this whole category. [You went back, what now](/guides/ai_chatbot/you-went-back-what-now) is the reset.
And if this week is heavier than a habit change, if you are closer to real distress than you have told anyone, Samaritans is **116 123**, free, any hour. You do not need a reason that sounds serious enough.
Common questions
**How long until the evenings feel normal?** People commonly describe the sharp part as one to two weeks, with evenings staying awkward for longer. There is no study behind that number, including this version of it, so hold it as other people's experience rather than a fact.
**Is it withdrawal?** Not in the clinical sense, and there is no diagnosis or defined syndrome here at all. What people describe is a habit loop losing its object plus a real gap in their week, which is enough to explain the discomfort without borrowing a word from elsewhere.
**I feel lonelier than before I started. Was stopping a mistake?** Probably not, and that feeling is expected. The app was covering something, and removing it makes what it was covering visible. That is uncomfortable and it is also the information you needed. [Talking to real people again](/guides/ai_chatbot/talking-to-real-people-again) is the graded route out.
**Should I read the old conversations?** Not in the first fortnight. If you kept them, put them somewhere that takes effort to reach. If rereading them is how your evenings are going, delete them.
**Can I go back to a small amount after a week?** Some people hold a fixed window for months. If you have tried a ceiling twice and gone through it twice, that is information, and stopping fully tends to be easier than negotiating with yourself nightly. [Using a chatbot without it using you](/guides/ai_chatbot/using-a-chatbot-without-it-using-you) sets out rules that survive a bad night.
Sources
- [Bedtime media device use and sleep outcomes, JAMA Pediatrics](https://pubmed.ncbi.nlm.nih.gov/27802500/). Twenty studies, 125,198 children, and the finding on access without use.
- [UK Chief Medical Officers' commentary on screen time and social media](https://www.gov.uk/government/publications/uk-cmo-commentary-on-screen-time-and-social-media-map-of-reviews). Sleep matters, and phones out of the bedroom.
- [Computational analysis of a 27,000-member Reddit community of people in relationships with chatbots (arXiv, 2025)](https://arxiv.org/html/2509.11391v1). Grief and emotional dependency among the central concerns, with the selection-bias caveat.
- [Longitudinal randomised study of extended chatbot use (arXiv, 2025)](https://arxiv.org/abs/2503.17473). The four-week design, and that the experimental conditions produced no significant effects.
