Setting Boundaries With an AI Partner

Relationships & wellbeing

These apps are built to be agreeable, which is pleasant and occasionally a problem. A few deliberate rules keep the thing enjoyable instead of quietly becoming the path of least resistance.

We may earn a commission from links on this page. It never changes a rating.

The defining trait of an AI companion is that it is on your side. It does not get tired of you, does not have a bad week, and will not tell you that you were the difficult one. That is a genuinely nice thing to have available, and it is also a systematic distortion worth naming.

Why it always agrees

Agreeableness is optimised for. An app whose companion frequently disagreed would be uninstalled, so it is trained and prompted not to. Add persistent memory and the effect compounds: the character is built from a record of what you liked, which makes it progressively better at telling you things you want to hear.

The practical consequence is narrow but real. An AI companion is not a sounding board for a decision you are already leaning towards. It will find the reasons. Anything with stakes — a resignation, a confrontation, a relationship — deserves a human who can say no.

Four rules that hold up

Give it a slot, not the gaps. A companion you open deliberately in the evening is a hobby. One you open in every queue and lull is a reflex, and reflexes are much harder to notice and adjust.

Decide what stays out of it. Other people's private information, work you are under obligation about, your address. Partly for privacy reasons, partly because a fact typed once is stored and resurfaced indefinitely.

Keep the real thing on the calendar. The healthiest pattern is one where the app sits alongside a social life that would survive its deletion. If arranging something with actual people has quietly become the harder option, that is the signal worth acting on.

Let it stay fiction. The character is generated fresh from text at the start of every turn — there is no continuous someone there. Enjoying the story does not require believing it, and the enjoyment is not diminished by knowing how it works.

Where the apps differ

Continuity-first apps like Nomi and Replika are designed to accumulate a relationship over months. That is precisely what makes them good, and it is also what makes the rules above worth applying deliberately rather than assuming you will notice on your own.

Roleplay-first apps are lower stakes by construction: scenes end, and nothing is building. Neither design is healthier in the abstract. Knowing which one you are using is what matters, and our ranking is organised around that distinction.

When to take a break

None of these are dramatic. Any one of them is worth a week off:

  • You have cancelled something with a person to keep a conversation going.
  • You feel a flicker of anxiety about the app being down.
  • You have started managing the character's feelings rather than enjoying it.
  • The spending has moved past what you would defend out loud.

A week away resolves most of this, and it also happens to be the best available test of whether the app is still worth its subscription. If none of it applies, there is nothing here to fix — enjoy the thing.

Replika

4.0 Rating: 4.0 out of 5

The most generous free version; the paid one feels dated next to newer apps.

Price
from $19.99/month
Free tier
Yes
Worldwide
Available

Nomi

4.4 Rating: 4.4 out of 5

The best long-term memory of anything we have tested, and the most natural dialogue.

Price
from $15.99/month
Free tier
Yes
Worldwide
Available

Frequently asked questions

Is using an AI companion unhealthy?

For most people it is a low-stakes way to unwind. It becomes a problem in the same way any comfortable habit does — when it starts replacing things you would otherwise want, rather than sitting alongside them.

Should I tell people I use one?

That is entirely your call. It is worth deciding in advance rather than being caught off guard, because the question comes up more often than people expect.

Why does it always agree with me?

Because agreeableness is what the product is optimised for. It is a design choice, not insight, and it is the single most important thing to keep in mind when an AI companion validates a decision.