What AI Companion Apps Don't Put in the App Store Screenshot

We make one of these apps, so take the framing for what it is. But everything specific below is public record: regulatory filings, a named nonprofit's published research, and the companies' own blog posts, not opinion. Where we couldn't verify a number against a real source, we left it out rather than round it up to sound more dramatic. Sources are linked at the bottom.

The category (AI chat companions, AI friends, whatever you want to call them) is genuinely useful to a lot of people, and genuinely crowded. Every app's App Store screenshots look roughly the same: a warm face, a chat bubble, "she remembers everything." What doesn't make the screenshot is what happens to your data, what it actually costs to keep the character feeling personalized, and what "remembers" holds up to under scrutiny. Here's what's actually on the record.

Privacy: the part nobody screenshots

In May 2025, Italy's data protection authority (the Garante) fined Luka Inc., the company behind Replika, €5 million, finding the company had processed personal data without a valid legal basis under GDPR, and had failed to put in place any meaningful age verification, which meant minors, including children under 13, could have emotionally intense, romantically-toned conversations with the chatbot. The authority also opened a separate investigation into how the underlying model was trained.

Separately, Mozilla Foundation's *Privacy Not Included* review, a long-running, well-regarded consumer privacy research project, found 210 third-party trackers firing within five minutes of using the Replika app, including trackers sending data to Facebook and AppsFlyer. Mozilla's reviewers called it one of the worst-scoring apps they'd tested for privacy, two years running.

Safety: what changed in late 2025

Character.AI, one of the largest players in the category, announced in October 2025 that it would remove open-ended chat for users under 18 entirely, with the change fully in effect by November 25, 2025. It replaced it with a more limited, creativity-focused experience for teens, and rolled out age-assurance tooling, including third-party verification via Persona. The company's own blog post framed this as a response to regulator questions about what open-ended AI chat, even with content controls working perfectly, does to teenagers. Texas Attorney General Ken Paxton had opened an investigation into the company in August 2025, reportedly examining whether its chatbots were presented as qualified mental-health tools without the credentials to back that up.

That's a real, meaningful safety change, and it's fair to give credit for it. It's also a reminder that "AI companion" as a category grew fast enough that basic guardrails, age verification chief among them, got bolted on after the fact, under regulatory pressure, rather than built in from day one.

Money: what "free" actually costs

A lot of these apps are free to download and then monetize through consumable currencies (gems, coins, "charms") layered on top of a subscription rather than instead of one. On Replika, personality traits, outfits, and room items still cost gems even for PRO subscribers, with gem packs publicly listed from $9.99 for 50 gems up to $49.99 for 1,000. On Character.AI, multiple independent reports through early-to-mid 2026 describe a free tier with a daily message cap and full-screen ads that interrupt conversations mid-chat, alongside a newer in-app currency, a shift several outlets tied to a change in company leadership and pressure to grow revenue after user numbers had declined from an earlier peak.

To be fair to those models: consumable currencies fund real infrastructure costs, and plenty of successful products use them well. The point isn't that monetization is bad. It's that "free" and "no coins, no ads" are different promises, and it's worth knowing which one you're actually signing up for before you're three months and forty dollars of gems into a relationship with a character.

Memory: the thing every screenshot promises

"She remembers everything" is on nearly every AI companion app's marketing page. In practice, memory quality varies enormously and is genuinely hard to verify from the outside. Reviewers and users across several of these apps have reported conversations that loop, or a companion that seems to forget recent context within days, which tracks with what you'd expect from an app relying mainly on a growing raw transcript rather than a separate memory-extraction step (we wrote about how that actually works in more detail). The honest test is the one we'd suggest for any app, including ours: give it one specific, odd detail, leave for a week, and see if it survives being asked about sideways.

We're not claiming BunnyHop is flawless. It's a small team, and no memory system is perfect. What we're claiming is specific and checkable: no coins, no ads, structured memory extraction instead of a raw transcript, and a privacy policy you can read in one sitting. Go verify it, and verify the same things about whatever else you're considering.

What we're actually claiming, and not claiming

The actual question worth asking

Not "which app has the prettiest character." Most of them clear that bar. Ask: what happens to what I tell it, what does staying customized actually cost over six months, and does it forget me the moment the free option runs out. Those three questions cut through the screenshots faster than any review can.