Retention design is not unique to companion apps — every consumer app wants tomorrow's session. What is unusual here is the channel. A language-learning app sends you a notification; a companion app sends you a message from someone you have been talking to for six weeks. The same nudge, delivered through a relationship you have invested in, lands very differently, and that asymmetry is worth understanding whether you are choosing an app or already settled into one.
The four levers, and what each one costs you
Most of what companion apps do to bring you back falls into four patterns. They are not equally aggressive, and the difference matters more than the raw number of notifications.
| Lever | How it appears | What to watch for |
|---|---|---|
| Proactive messages | The character writes first after hours or days of silence, often referencing your last conversation | Whether frequency is user-configurable, and whether the app is honest that these are scheduled |
| Streaks & daily rewards | A consecutive-day counter, sometimes tied to in-app currency or unlocks | Whether breaking the streak costs you anything you paid for |
| Limited-time events | Seasonal scenarios, event-only characters, expiring outfits or scenes | Whether content disappears permanently or returns later |
| Resetting quotas | Message or "energy" allowances that refill on a clock rather than on demand | Whether the timer is a paywall in disguise — see free-vs-paid boundaries |
Proactive messaging is the one to look at hardest, because it is the only lever that borrows the character's credibility. When an app writes "you never told me how the interview went", the message is generated on a re-engagement schedule and populated from stored context — the same personalization pipeline described in Personalization Signals. That is not a criticism of the technique; a check-in can be genuinely useful, and it is exactly what makes companions work for motivation and routines. The problem is only that the mechanism is invisible, so it is easy to read a scheduled prompt as spontaneous attention.
Why this category attracts scrutiny
Regulators have concentrated on the same feature set, and their remedies point at the same thing: interrupt the illusion, and interrupt long sessions.
New York's companion-model law (General Business Law Article 47), in force since 5 November 2025, requires operators to tell users at the start of a session — and at least every three hours of continued use — that they are not interacting with a human, and to route users who express suicidal ideation or self-harm to crisis services. California's SB 243, signed 13 October 2025 and effective 1 January 2026, covers similar ground and adds a specific obligation for users the operator knows to be minors: a reminder to take a break after at least every three hours of continuous use, alongside AI disclosure and crisis protocols. China went further in a different direction — interim measures effective 15 July 2026 require platforms to warn against over-reliance and bar designs that substitute for human contact, which we covered in China's 2026 companion rules.
Two things follow for a reader choosing an app. First, if you see a three-hourly "this is an AI" banner or a break prompt, that is often a compliance feature rather than a wellbeing philosophy — useful either way, but not a signal that the rest of the app is designed conservatively. Second, coverage is patchy: these are state and national rules, and a web-distributed app with no US or Chinese presence may show none of it. Do not treat the absence of a break reminder as evidence that an app is unregulated in a sinister sense, and do not treat its presence as a quality mark.
Check the settings before you invest
The best moment to look at notification controls is the first week, before a character has months of history behind it. What separates a well-built app from a manipulative one is granularity — whether you can shape the behaviour, or only switch it off entirely.
- Separate toggles. Can you disable marketing pushes while keeping proactive character messages, or is it one switch for everything?
- Frequency control. Is there a way to say "at most once a day" rather than only on/off?
- Quiet hours. Does the app respect a schedule of its own, or rely on your OS-level Do Not Disturb?
- Honest labelling. Does the app describe proactive messages as scheduled, or imply the character is thinking of you?
- Streak consequences. If you break a streak, do you lose progress, currency or unlocked content — and is that stated before you start?
- Session visibility. Does the app show you how long you have been in a session, or how much you used it this week?
Apps that fail the granularity test tend to fail it deliberately. An all-or-nothing notification switch pushes users toward leaving everything on, because turning it off also silences the part they wanted. That is the same design logic behind the sign-up friction discussed in Onboarding Friction: the default is chosen for the operator, not for you.
Limits that hold up on a bad week
Usage rules made on a good day rarely survive a difficult one, which is when companion apps get used most. The limits that tend to hold are structural rather than motivational — they change what the app can do, instead of relying on you to resist it.
- Move the app off the home screen. Removing the ambient prompt does more than any in-app timer, because most sessions start from a glance rather than a decision.
- Turn off lock-screen previews specifically. A message you can read without opening the app is the strongest possible pull; hiding the preview keeps the notification while removing the hook.
- Set one fixed check-in window. A companion used at a predictable time is easier to keep proportionate than one that fills gaps throughout the day.
- Give up streaks early and deliberately. Breaking a streak on purpose, in the first two weeks, tells you exactly how much of your use was the character and how much was the counter.
- Re-read one old conversation monthly. It is a fast, honest check on whether the app is still doing what you started it for.
None of this is a substitute for support when something is genuinely wrong. Research published in 2026 found that adults with limited offline social networks reported feeling more lonely after turning to companion chatbots for emotional support — a correlation rather than a proven cause, but a strong argument for treating a companion as one part of a wider set of connections. We looked at that study in the loneliness research, and the broader picture in AI Companions & Wellbeing.
How we weigh this in reviews
Notification behaviour is part of what we check on every app: whether controls are granular, whether proactive messaging is disclosed as scheduled, whether streaks carry a real cost, and whether break or disclosure prompts appear. We do not score an app down for messaging you first — for many people that is the point of a companion — but we do score down for hiding the switch, or for framing a scheduled message as spontaneous. Our methodology explains how these checks are made, and the free vs paid guide covers where resetting quotas cross from pacing into a paywall.
Related on CompanionRank
AI Companions & Wellbeing Personalization Signals Free vs Paid Companion AppsFrequently asked questions
Why do companion apps send messages when I have not opened them?
Because re-engagement is how the app earns its next session. A message written in the character's voice performs far better than a generic app alert, so most companion apps schedule proactive messages after a period of silence. That is a product decision, not the character noticing your absence.
Are companion apps legally required to tell me to take a break?
In some places, yes. New York's companion-model law requires operators to notify users at the start of a session, and at least every three hours of continued use, that they are not talking to a human. California's SB 243 adds a break reminder at least every three hours of continuous use for users the operator knows to be minors. Neither is a global rule, so many apps you can install will not show these prompts at all.
Does turning off notifications break the companion's memory?
No. Notifications are a delivery channel, not a memory feature. Muting push messages stops the app from starting conversations, but stored memories, chat history and personalization settings are unaffected. Streak counters are the one exception, since they measure consecutive days rather than anything the character remembers.