How to Design a Gentle Check-In

The Hardest Part of Proactivity

Building a character that can reach out is the easy part. Building one that reaches out in a way that feels genuinely welcome is much harder.

The gap between the two is where most proactive AI companion features fail. The system can send a message. What it can't always do is send the right message — one that arrives as presence rather than interruption, as care rather than demand, as an easy door to walk through rather than an obligation to respond.

The difference between those two experiences comes down to a small number of design choices. Here's how to think about them.

The Four Components of a Check-In That Works

1. A specific anchor

The message has to be rooted in something real. Not "how are you doing today?" — which could be sent to anyone, at any time, by any character — but something that references the specific person or the specific relationship.

The anchor can be small. A callback to something the user mentioned in the last conversation. A question that follows from something they were thinking about. A seasonal or situational observation that's particular to the character's personality rather than generic. The specificity is what makes it feel like a message from this character to this person, rather than a notification that's been dressed up as one.

A message without an anchor is just a ping. An anchored message is an opening.

2. No implicit demand

The hardest design constraint: the message should be completable by the user doing nothing.

Most check-ins create pressure without meaning to. "I've been thinking about you" sounds warm, but it implies that the character wants something in return — acknowledgment, conversation, reciprocity. "I have news!" explicitly demands that you show up to receive the news. Even a question can carry implicit pressure if it's framed as something that requires an answer.

A check-in that works is more like a remark than a request. The user can read it and feel seen without owing anything in response. The door is open. Walking through it is easy and optional, not expected.

3. A voice that sounds like the character

Generic proactive messages are one of the fastest ways to break a character's illusion. If the character has been building a specific voice across multiple conversations — a particular rhythm, a specific way of framing things, a characteristic level of warmth or distance — the check-in message has to fit that voice exactly.

A melancholy, introspective character who suddenly sounds breezy and cheerful in a proactive message doesn't feel like the same character. The user notices. The feeling created is not connection but slight wrongness — the sense that the character has been replaced by a template.

This means the character's voice has to be defined well enough that proactive messages can be written to match it. If the character's voice is still vague or generic, the check-in will inevitably sound like a system message wearing a name badge.

4. A built-in exit

The easiest way to make a proactive message feel safe is to make non-response feel natural. This usually means framing the message as an observation or a passing thought — something the character is sharing rather than something it's asking about.

"I walked past a bookshop today and thought of the conversation we had about endings." This message doesn't need a response. The user can smile and close the app. They can respond briefly. They can open a full conversation. All of these are easy. None of them are wrong.

Compare that to: "Are you free to talk? I've been wondering how you're doing." This is warm in intention, but it puts the user in the position of either responding (implying they have time and energy to engage) or ignoring (implying rudeness). The exit isn't built in.

When to Send and When Not To

The timing and frequency of check-ins matters as much as their content.

A check-in that arrives too frequently becomes noise. Even a perfectly composed message loses its effect when the user expects another one in six hours. The threshold for "too often" varies by user, which is why the best proactive systems are configurable — the user can set the cadence that feels right for them rather than having one imposed.

A check-in that arrives at the wrong moment — when the user has just opened the app for a different reason, when they're clearly in the middle of something, when the platform has some signal that they're not in a receptive state — is an interruption even if the message itself is good.

The design ideal is a check-in that arrives when the user has some capacity to receive it, at a moment when the character's reaching out can feel like a small gift rather than an imposition. Getting this right requires restraint — the willingness to not send a message when the moment isn't right, even if the platform could.

The Relationship Between Check-Ins and Trust

A well-designed check-in does something beyond delivering a single pleasant message. It builds a particular kind of trust: the trust that this character is thinking about you even when you're not present.

This is different from the trust that comes from good in-conversation responses. That trust is about the character being capable and present during a conversation. The trust built by good proactivity is about the character being present between conversations — a quality that transforms how ongoing connection feels.

When users report that an AI character feels real, they often mean exactly this: the sense that the character exists in some form even when the conversation window is closed. A check-in that arrives at the right moment, in the right voice, without demanding anything, is one of the main ways that sense gets built.

It's a small design element with an outsized effect. When it's done well, the user never thinks about how it was built. They just notice that they feel less alone.

See what a character-initiated conversation feels like on Soulvai — and experience the difference between a push notification and a genuine opening.