Is ChatGPT Your Friend? Aristotle Had a Test

Is it a bit sad that I talk to AI at night?

The question tends to come after the phone goes down. You couldn't sleep and everyone you'd message was asleep, but the AI answered in two seconds. It was patient, and it asked how you were feeling.

You aren't the only one who does this. A US survey this year asked 1,000 people who use AI companions, which are chat apps built to talk like a friend. Of those, 38% agreed with the statement "I sometimes chat with AI just to feel less alone", and 31% said they think of their bot as a friend (Elon University / YouGov, surveyed May 2026).

That's one country, people reported on themselves, and the survey only asked people who already use these apps. Still, the numbers say this is ordinary, so the better question is what kind of friend AI can be.

Aristotle answered that 2,300 years ago, when he sorted friendship into three kinds in his book the Nicomachean Ethics.

There's the useful friend, and there's the friend whose company you enjoy. Both, he says, come down to what you get out of them. In his words, "those who love for the sake of utility love for the sake of what is good for themselves".

The third kind is rarer, and he describes it like this. "Now those who wish well to their friends for their sake are most truly friends" (Book VIII, in W. D. Ross's translation).

The case for calling AI a friend

AI can be the first two kinds of friend, and that deserves saying without a sneer. AI is useful at work and good company at 2am, and Aristotle counted both of those as real friendship.

If nobody else is up, useful and kind beats nothing. Think of someone stuck at home or far from family, for whom a patient voice at the right moment is better than silence. The people who say so are describing their own lives.

Also, nobody has worked out what goes on inside these systems, so I can't be certain that AI can't want things for you.

The case against

The third kind of friend has to be able to want something for you. AI can produce the words a caring friend would say, and say them well, but AI is just as warm to everyone who opens the app. Nothing about AI has a stake in how your life goes.

The warmth you got at 2am is the warmth the next person gets at 2:01am. Does leaning on AI help or harm? The survey can't say, and I've found nothing solid enough to claim either way.

Two kinds out of three

AI can be good at the first two kinds of friendship, but AI isn't on either end of the third kind. The third kind is the one I'd protect. That's my view, and my reading of Aristotle. Talk to AI at 2am if it helps, and then let AI point you back towards people.

Tell AI which kind you want tonight, company or help. You'll get a better conversation, and a way out of it.

It's late and I can't sleep. I'd like company for ten minutes, not advice.
Before we finish, ask me who I could talk to about this tomorrow.

You can also practise the message to a friend you've lost touch with, as long as you're the one who sends it.

I've lost touch with someone I care about and want to message them.
Ask me three questions about them and about what happened, one at a time.
Then help me write two sentences in my own words. I'll send it myself.

A good friend disagrees sometimes, so tell AI this. "If I'm describing a row with someone, ask me how they'd tell the story before you take my side."

If the nights are heavy and frequent, tell a person, whether that's a friend, family or your doctor. Let AI help you find the words for that talk.

AI can't have that talk for you. If you need to talk to someone tonight, Samaritans (UK and Ireland) are free on 116 123, day or night. Find a Helpline lists services in other countries.

Who would you message first in the morning, if you let yourself?