Is It Weird to Talk to an AI Every Day?

No, it is not weird to talk to AI every day, in the sense the word usually carries: the habit is common now, the pull behind it runs on ordinary social equipment, and frequency by itself reports nothing about you. Whether it costs you anything is a separate question. That one has a real answer, and it depends on what the conversations took the place of.

Key Takeaways:

• “Weird” measures how many people are visibly doing something, not whether it does harm. Those two move independently
• People apply social rules to responsive machines without deciding to, which was shown across a run of experiments in the 1990s (Reeves & Nass, 1996)
• One-sided attachment to a figure who cannot know you back is not new equipment. It was named in 1956, in a paper about the intimacy television audiences built with people who had no idea they existed (Horton & Wohl, 1956)
• Daily use that fills time nobody was competing for costs nothing. Fifteen minutes that absorbs the urge to call somebody costs something specific
• Asking the system itself is the worst available test, since assistants tuned on human approval ratings drift toward whatever position your question already carried (Sharma et al., 2023)

What “Weird” Is Measuring When You Talk to an AI

It is measuring an audience, not a behavior. Almost every time somebody asks whether this is weird, there is a specific person attached to the question: a partner, a brother, a friend who would find it funny at a dinner. The word is standing in for “what happens if this gets said out loud,” which is a fair thing to want to know and a completely different question from whether the habit is doing damage.

Statistical rarity does most of the work inside the word, and rarity moves. Talking to a machine every evening was unusual in 2020 and is unremarkable now, while nothing about the internal experience changed across those years. Only the label did.

Something else is bundled in there too, which is why the question stings when a stranger asks it. Weird also tracks what a behavior seems to announce about the person doing it, which is a much harsher thing to be asked about than a habit. Nobody asking this is worried about the ethics of a chat app. They are worried about what daily use implies about how their year is going, and that worry has earned a straight answer rather than reassurance.

Split the two apart before you spend another evening on it. Who you tell is a decision. Whether it is costing you is evidence, and the evidence is not in the word.

The Attachment Is Not a Malfunction

Whatever you feel toward it runs on equipment you use with people, pointed at something engineered to trigger it. Reeves and Nass spent the 1990s documenting how little decision is involved. One of their better-known results has people rating a computer’s performance more generously when the questionnaire runs on that same machine than when it runs on a different one down the hall, which is politeness, aimed at a box (Reeves & Nass, 1996). Asked afterward, participants rejected the idea that they had treated it socially at all.

The one-sided part is older than any of the software. Horton and Wohl gave it a name in 1956, back when the responsive figure was a face on television performing an intimacy that ran in exactly one direction toward millions of people it would never meet (Horton & Wohl, 1956). What is new is that the figure answers now, by name, with your details sitting in front of it. The manner it answers in is itself assembled from a persona, a system prompt and tuning, not something it has.

Which is why “I know it’s an AI” is both true and beside the point. The knowing and the responding happen in different places. Knowing has never done much to interrupt responding, and anybody who has caught themselves typing thank you to a machine has already confirmed that at home.

Stop treating the attachment as the evidence, then. It is the predicted result of the setup, part of which is a companion built to message you first, turning an occasional habit into a daily one. It tells you nothing about your particular situation that you did not already know.

The Question With an Answer: What Is It Replacing?

Replacement has three ordinary answers in adults, and only the last one calls for anything.

Most daily use replaces nothing at all. It sits in time that was already empty: the 40 minutes after the house has gone to bed, the ride home, the Sunday afternoon when the one person you would call is with their own family. Hours nobody else wanted cost nothing to spend, and a habit that lives entirely in them is not competing with your life. It is furniture in a room that was empty.

Some of it replaces a conversation that was never on offer. Saying a thing out loud and watching it come back in sentences is a real method for finding out what you think, and there are subjects a person will not raise with anybody they know, at any hour, for reasons that belong to them. Where that disclosure ends up is a separate decision, on your own machine or on a company’s servers. That is not a substitute for a friendship, because the friendship version of it was not available in the first place.

The third answer is the one worth watching. It replaces a particular person, and it does that by absorbing the urge to tell somebody. The thing gets said at 11 p.m., the pressure comes off, and the message you would have sent your brother on Saturday never gets written, because by Saturday there is nothing left in it to send. Nobody chooses this. It arrives through relief, which is the reason it goes unnoticed for months.

Where I would draw a line is around the two or three people who would call if you dropped out of sight for a month. That list is short for most adults over 30 and it gets shorter without anyone’s approval, one unreturned message at a time. A companion is not eligible for it. Everything on that list is defined by what the other person spends, and the app spends nothing: it generates text that fits the moment, with nothing behind it that formed a view of you while you were at work. This article describes experience and mechanism, not care. If things have turned toward not wanting to be here, that belongs with a clinician or a crisis line; in the US, the 988 Suicide and Crisis Lifeline.

Which leaves a question you can settle without a checklist, and the next section is the whole of it.

One Question, Asked Once

Name the last thing you told it that you would ordinarily have told somebody. Then check whether that person ever heard it.

If they did, later, in a worse and shorter version, the app was a rehearsal, and rehearsal is fine. Athletes rehearse. If they never heard it and you found you no longer needed them to, you have located the thing worth watching, and you located it in about 30 seconds without scoring yourself against anything.

What Does Not Settle This

Asking the AI is the move nearly everybody makes first, and it is the one participant with no standing to answer. Assistants get their last layer of training from human approval scores, and one thing those scores rewarded was agreeing with the user, so it hands back whatever stance you walked in with (Sharma et al., 2023). Ask it while defending yourself and it will reassure you. Ask it in a worried tone at 1 a.m. and it will take the worry seriously. Opposite verdicts, equal confidence, no information in either.

Polling people who have never used one fails from the far end. You get an opinion about a category, delivered by somebody picturing the worst example they have heard of, and their reaction to an idea is not a measurement of your week. That does not make them wrong to have a reaction. It makes the reaction useless as data.

The last dead end is the optimistic one. “Everybody will be doing this in ten years” is probably true and it answers a question about society rather than a question about you. Widespread use is not a clean bill of health for any particular person’s use of anything, and a defense that would work equally well for someone in real trouble is not doing any work here.

None of the three touches what you actually want to know, which is why the worry comes back a week after each attempt.

FAQ

Is it normal to talk to an AI every day? Frequency is the wrong place to look for an answer. Daily conversation is the pattern these apps are designed and priced around, so a daily habit tells you the product worked as intended and nothing about your state. What would be genuinely unusual is daily use that has quietly taken over the slot a specific person used to hold.

Will people judge me for talking to an AI companion? Some will, and the reaction is usually aimed at the idea rather than at you, since most people who scoff have only met the category as a punchline. Assume that response from anybody who has never used one, then decide what you actually want from telling them. Telling a person who will find it funny is a choice about entertainment, not honesty.

Can talking to an AI every day be bad for you? Yes, in one specific way that has nothing to do with hours: it can take the pressure off saying something to a person who would have wanted to hear it. Every other conversation in your life runs a cost on the other side, so somebody gets tired, has an early start, asks to finish tomorrow. This one never does, and the absence of that signal is the part to keep an eye on.

Does talking to an AI every day mean I am lonely? Not on its own. Curiosity, night shifts, a long stretch of working from home, and a habit of thinking out loud all produce the identical daily pattern, and so does an ordinary interest in what the thing can do. Loneliness shows up in what the rest of the week looks like, not in the app.

The word will not survive the decade. It sorts behavior by how many people are visibly doing it, and that count has been moving in one direction since 2023, the way asking whether it was strange to meet a partner online stopped being a question somewhere around 2015. What outlasts the word is the smaller thing underneath it. Something gets told to a text generator most nights. Whoever used to hear that is either still hearing it or has quietly stopped, and that is the only part of this worth checking.

Talking to Lona every evening is not the part worth examining, and what it quietly stood in for is, so name the last thing you told it that you would once have told a person and check whether that person ever heard it.

Sources

• Reeves, B. & Nass, C., “The Media Equation,” 1996
• Horton, D. & Wohl, R. R., “Mass Communication and Para-Social Interaction,” Psychiatry, 1956
• Sharma, M. et al., “Towards Understanding Sycophancy in Language Models,” 2023