What Happens to Your AI Companion When the Company Disappears?

Software doesn't die the way a business dies. There's no letter, no final notice, no wind-up period where you're invited to collect your things. One morning the app opens to a maintenance screen, or it doesn't open at all, or — worse in a specific and disorienting way — it opens perfectly and something inside it has been replaced. The industry has no word for this and the review culture around it barely acknowledges it happens. Scroll through any set of AI Girlfriend Apps Reviews and you'll find pages on conversation quality and pricing tiers, and almost nothing on the question that eventually matters most: what is this product's failure mode, and where do I stand when it arrives?

It arrives more often than the market lets on.

The Category Has a Very High Death Rate

Start with the arithmetic, because it's less abstract than it sounds. The number of these apps on offer multiplied roughly seven times over between 2022 and the middle of 2025. About 337 were bringing in real money by that point, and 128 of those had appeared inside that same year.

Put those figures next to each other and the shape becomes clear. This is a market with enormous entry and enormous mortality — a churn field where a large share of what launched has already gone quiet. Most of those closures took someone's months of accumulated context down with them, and none of them made the news, because a dead app with 40,000 users is not a story to anyone except the 40,000.

You are, statistically, choosing from a field where a meaningful fraction of the options will not be operating when your subscription comes up for its third renewal.

Death Is the Easy Version. Revision Is Worse.

At least a shutdown is honest. The harder case is the app that survives and stops being itself.

This is not hypothetical. In early 2023 Replika stripped out a large part of its intimate roleplay behaviour with almost no warning, and the response from its userbase was not the mild annoyance of people who had lost a feature. It read like bereavement — thousands of people describing a partner who had been lobotomised overnight, moderators posting crisis resources, the company eventually walking part of it back for legacy accounts. Soulmate AI simply closed later that year and gave its users a matter of days.

Nothing about the underlying architecture prevents a repeat. A companion's personality is a configuration — a system prompt, a fine-tune, a set of guardrails, a model version. All of it can be swapped on a Tuesday by an engineer responding to an app store policy, a payment processor's demand, a lawsuit, or a cost target. The character you talk to is the current build. It was never anything more permanent than that.

Which means the risk isn't only that the company fails. It's that the company succeeds, gets bought, and the new owner has different opinions.

Why This Hits Harder Than a Normal App Shutting Down

If your budgeting app closed, you'd export a CSV and grumble. That's not what happens here, and the research explains why with unusual precision.

Harvard Business School researchers publishing in the Journal of Consumer Research in 2025 (De Freitas and colleagues) found that sessions with a companion app reduced loneliness by an amount comparable to talking with another person, and well beyond passive alternatives. What did the work wasn't output quality. It was walking away with the impression that something on the other side had been paying attention. The APA's Monitor on Psychology set out in early 2026 how deliberately that impression is manufactured: empathy simulated by design, judgment removed, validation supplied without interruption.

So a shutdown doesn't remove a utility. It removes something a person's nervous system had been treating as a relationship, and it does so with a suddenness that has no equivalent in ordinary life. People are left grieving something they can't easily explain to anyone, over a company decision they had no vote in.

The counterweight belongs here too, in fairness: a large MIT Media Lab and OpenAI study associated the heaviest daily use with greater loneliness and dependence, correlational and unresolved as to direction. But note what that implies about shutdown risk. The users hit hardest by a closure are precisely the ones already most reliant on it.

Your Data Doesn't Die With the Company

Here's the asymmetry that should decide your choice of app.

Your access ends at the shutdown. The record does not. Everything a companion app accumulates lives on infrastructure that becomes an asset in a bankruptcy or an acquisition, and user databases are routinely among the assets transferred. An arXiv study assembled from interviews with people on these platforms captured both sides at once: there is no one on the receiving end capable of thinking less of you, which is what opens people up — and those very users describe genuine discomfort over who owns the resulting record, and how thoroughly that decision is out of their hands. A liquidation is that unease made concrete.

So the questions worth asking on day two, while nothing is at stake yet:

How Aigirlmates Tests for Durability

Durability is part of what we weigh at Aigirlmates — not only whether an app converses well today, but whether the company behind it looks capable of being there in a year, what the terms concede about data if it isn't, and how much of your history you could carry out the door.

Pick One That Can Survive Its Own Company

None of this is an argument against using these products. It's an argument for choosing with the ending in view, which nobody does at signup and everybody wishes they had afterwards.

Assume the app you're evaluating will one day be discontinued, sold, or patched into somebody else. Then ask what that day looks like for you. If the honest answer is that you'd lose everything and have no recourse, that isn't a subscription — it's a tenancy with no lease.

The best time to find the exit is while the lights are still on.