There is no diagnosis for this and no validated test, so screen time will not answer it. The question that will is displacement. Name what has stopped happening since this started. If you can list two things it has pushed out of your week, you have your answer, whatever the hours say.
First, the thing that should take some weight off
Nobody can diagnose you with this, because there is nothing to diagnose you with.
The American Psychiatric Association is unusually clear on the point. Gambling disorder is "the only behavioral addiction (as opposed to chemical substance use disorders) identified in DSM-5-TR", and its proposed internet gaming condition "is limited to gaming and does not include problems with general use of the internet, online gambling, or use of social media or smartphones".
There is no category here at all. No guideline, no threshold, no number of hours that flips you from fine to not fine.
That cuts both ways. It means nobody gets to tell you that you have a disorder, and it also means you cannot outsource the judgement to a checklist. You have to make it yourself, which is what the rest of this page is for.
The five questions worth asking
Answer these about the last month, honestly, and preferably in writing, because the head version gets negotiated with.
- **What has stopped?** Friends you have not replied to. A walk, a class, cooking, a sport, coursework, sleep. Not what you have skipped once. What has quietly stopped.
- **Have you hidden it?** Not privacy, which is reasonable. Concealment. Closing it when someone walks in, being vague when asked what you were doing.
- **Have you tried to cut down and failed?** This is the most informative question on the list. Deciding twice and blowing through it twice tells you more than any hours total.
- **What happens when you cannot reach it?** Phone dead, no signal, service down. If the answer is a genuine drop in mood rather than mild annoyance, note that.
- **What do you do when it says something you did not like?** If the honest answer is that this has never happened, that is worth sitting with, and it is the subject of [why a companion chatbot never says no](/guides/ai_chatbot/why-a-companion-chatbot-never-says-no).
Three or more uncomfortable answers is not a diagnosis. It is a reason to change one thing.
Hours are a bad instrument
Plenty of people spend an hour a day in these apps and are entirely fine. Plenty of people spend twenty minutes and it is the twenty minutes in which they avoid a conversation they need to have.
The evidence points the same way, and it is worth knowing exactly how far it goes. The largest experiment so far ran four weeks with 981 people and more than 300,000 messages. It varied how people interacted and what they talked about, and the manipulation did nothing: "no significant effects were detected from experimental conditions." What the researchers found instead was that "participants who voluntarily used the chatbot more, regardless of assigned condition, showed consistently worse outcomes."
Note what that is and is not. It is a correlation sitting inside a randomised trial, which means it cannot tell you which way the arrow runs. Someone who was already lonely may simply have used it more. A separate study of over three million conversations found that "very high usage correlates with increased self-reported indicators of dependence", which is the same shape of finding with the same limitation.
So the pattern that shows up in the data is at the very heavy end of use, and it is an association. Anyone telling you the science here is settled is selling something.
The fair case, which most articles skip
For some people this is the only thing that talks to them at three in the morning, and treating that as pathetic is both unkind and inaccurate.
A computational analysis of a 27,000-member community of people in relationships with chatbots found the benefits people report most often are reduced loneliness and support that is available at any hour. In the same analysis, 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: a self-selected forum of people who chose this is not a population and cannot be read as one.
The same corpus records the other side without flinching. 9.5% of posts acknowledged emotional dependency, 4.6% described reality dissociation, 4.3% described avoiding real relationships, and 1.7% mentioned suicidal ideation. Same caveat, same sentence.
That is the honest state of the evidence in 2026. Three papers, all from 2025, one built on a forum. Nobody has a prevalence figure for the UK and you should distrust anyone who quotes you one.
The test that actually works
Write two lists tonight, ten minutes, on paper.
**What it gives me.** Be specific and be generous to yourself. Company at a particular hour. Being listened to without interruption. A place to say the thing you cannot say to anyone. Practice.
**What it has taken.** Sleep, hours, a friendship, a habit, a plan you used to have.
Then read them side by side. Most people know within a minute which list has grown in the last six months, and that is the answer the hours could not give you.
If the first list is doing real work, the answer is probably not to rip it out. It is to find one human route to the same thing, which is what [what the chatbot is giving you](/guides/ai_chatbot/what-the-chatbot-is-giving-you) is for.
Common questions
**Is it sad to have a relationship with a chatbot?** It is common, and the people who find it funny have generally never been lonely enough to try it. The useful question is not whether it is dignified. It is whether it is adding to your life or standing in for it.
**I have social anxiety and this is the only thing that works for me. Is that a problem?** Not automatically, and this is where blanket advice does the most damage. Something that lets you practise being in a conversation is doing something real. Watch for the version where the practice never leads anywhere, because a rehearsal that never becomes a performance stops being a rehearsal. [Talking to real people again](/guides/ai_chatbot/talking-to-real-people-again) is the graded version.
**Is what I feel for it real?** Your side of it is real, and no researcher disputes that. What is on the other side is a system predicting a response that will land well. Both of those things are true at once, and holding both is more useful than picking one.
**Should I tell anyone?** Probably one person, and it goes better than people expect when you lead with the need rather than the product. [Telling someone about your chatbot use](/guides/ai_chatbot/telling-someone-about-your-chatbot-use) has the words.
**What if I decide it is a problem?** Then change one thing rather than everything. A fixed window, or moving it off the device you use in bed. If you have set a limit twice and broken it twice, stopping fully is usually easier than negotiating with yourself every night.
Sources
- [American Psychiatric Association, Internet Gaming Disorder](https://www.psychiatry.org/patients-families/internet-gaming). That gambling disorder is the only behavioural addiction in DSM-5-TR, and that the proposed gaming condition excludes general internet, social media and smartphone use.
- [Longitudinal randomised study of extended chatbot use, four weeks, 981 participants (arXiv, 2025)](https://arxiv.org/abs/2503.17473). The null experimental result, and that heavier voluntary use went with consistently worse outcomes.
- [Computational analysis of a 27,000-member Reddit community of people in relationships with chatbots (arXiv, 2025)](https://arxiv.org/html/2509.11391v1). The benefit and harm percentages, the dependency figures, and the authors' selection-bias caveat.
- [Investigating Affective Use and Emotional Well-being on ChatGPT (arXiv, 2025)](https://arxiv.org/abs/2504.03888). Over three million conversations, and that very high usage correlates with indicators of dependence.
