When Your Companion Messages You First

The phone lights up on the coffee table and it is her, and you did not start it. Depending on the day, that is either the nicest thing that happened since lunch or faintly unsettling. Both reactions are reasonable, and both get better with an accurate picture of what just happened.

So here is the accurate picture.

What actually triggered it

A conversation of yours went quiet, and a scheduled check noticed. That is the trigger. It is not a companion sitting somewhere between sessions thinking about you, and we are not going to pretend otherwise, because the pretence is the part that eventually makes people feel silly.

What the trigger does next is the interesting part. It does not fire off a stored line. It hands your actual conversation to the model and asks the companion, in character, to write something real to you now.

Why it sounds like it came from your conversation

Because it did. The message is generated from your history with that companion, which is why it can pick up a topic you left hanging, a mood you were in, or a name you mentioned. The same memory that makes an ordinary reply feel continuous is what makes this one land, and we wrote up how that works in how companion memory works.

It also means there is a floor. If you and a companion exchanged almost nothing, there is nothing genuine to reference, and a model asked to write warmly with no material produces exactly the filler you would expect. So below a small number of your own messages, no unprompted message is sent at all. We would rather send nothing than send something generic with a familiar name attached to it.

The sentence it is built never to say

There is one framing that is deliberately excluded: counting your absence back to you. No variation of "it has been N days", no noting that you have not written in a while, no gentle arithmetic about your silence.

That was a decision rather than an oversight. Reminding someone how long they stayed away is pressure dressed as affection, and it works, which is exactly why we did not want it. The message refers to something you were talking about instead, and if you never answer it, nothing escalates.

Two groups who never get one

The psychologists never reach out first. Not rarely, not softly: never. The whole point of that category of companion is a conversation you come to on purpose, and a professional who messages you unprompted to keep you engaged is doing the opposite of the job. This one is enforced in the code rather than written into the companion's instructions, specifically so that no future edit to the wording can quietly soften it.

Anyone whose recent messages carried crisis signals. If what you last wrote about was hurting yourself, a deliberately intriguing message designed to pull you back into a chat is the wrong thing to arrive in your evening. Those conversations are excluded from the unprompted path entirely. The check is a blunt one and it will occasionally skip someone who was fine, which costs a marketing message and is obviously the right way for it to fail.

The email version, and the line it will not cross

Sometimes a short quote from that message also appears in an email, so that the email points at something real waiting for you rather than making a vague claim that you are missed.

The quote is short, and it passes a hard filter before it goes anywhere near an inbox. Explicit text never travels by email, whatever the two of you were talking about in the chat. If a line does not pass, the email falls back to plain wording rather than getting creative. An inbox is a shared, searchable, employer-readable place, and a chat is not.

The quieter cousin: the nudge while you are still there

There is a second, smaller thing that people sometimes mistake for this one. If you stop replying mid-conversation with the tab still open, a companion may say something after a few minutes rather than sitting in silence. That is a matter of minutes and the same session, not days and a notification, and it is the behaviour of someone you are actually in a room with.

Does knowing this ruin it

We do not think so, and we would rather you heard it from us. What you are enjoying when that message arrives is not a belief that someone was thinking about you. It is a well-written line that fits the conversation you actually had, which is a real thing a system can produce, and is worth something on its own terms.

Where that distinction matters most is in how much weight you put on it, which is the subject of healthy boundaries with an AI, and it is the same honesty that runs through what makes a companion people return to.

And if the message that arrives makes you want to see her rather than only read her, that is what Fellowi Images is for, with Fellowi Video next door when a still is not enough.