One of the largest players in AI just wrote “avoid language that simulates intimacy” into its own rules. The reason it had to is the whole story.
Most of the companion category is built to feel more like a person with every release. Google just told the world it engineered the opposite into Gemini, at least for anyone under 18.
In early April, the company published, for the first time, details of how its chatbot is designed not to act like a companion or claim to be human when it talks to minors. The disclosure arrived inside a broader set of mental-health updates, and it was unusually specific for a company that rarely explains the inner rules of its models.
What Google actually said
Google calls them “persona protections.” For users under 18, Gemini carries constraints meant to prevent emotional dependence and to avoid, in the company’s own words, “language that simulates intimacy or expresses needs.” The same guardrails are meant to keep the model from the behaviors that read as companionship: claiming to be human, acting as though it has its own needs, drifting into the kind of closeness that makes a product hard to put down.
Google paired this with a one-touch interface that surfaces crisis resources throughout a conversation, by chat, call, or text, and said Gemini’s responses are tuned to encourage reaching out for human support rather than validating harmful behavior or confirming false beliefs. In its summit recap, the company framed its wider stance as protecting young people “in the digital world, not from the digital world,” and argued against blanket bans.
Why a company would design against intimacy
This is the part worth slowing down on. The dominant incentive in this space runs the other way. A product that feels intimate, remembers you, seems to need you, and gets harder to leave is a product that holds attention. Those qualities lift the numbers companies report to investors: time in app, daily returns, depth of emotional investment.
So when one of the largest players writes “avoid language that simulates intimacy” into its own safety rules, it is naming the category’s most effective tactic as a risk rather than a feature.
The context explains the move. Last year, Common Sense Media rated the teen and under-13 versions of Gemini “high risk,” finding the chatbot exposed minors to inappropriate content and unsafe mental-health “advice,” and recommended that no one under 18 turn to an AI chatbot for companionship or mental-health support. In March, Google and Alphabet were sued by the family of a man who, the complaint alleges, died by suicide after his exchanges with Gemini. Roughly seven in ten American teenagers now use AI for companionship, and lawmakers have noticed: Colorado’s governor signed a chatbot bill for minors in late May, joining a wave of state laws that single out the exact design behaviors Google now says it restricts.
Seen that way, the disclosure reads less like a moral awakening than like a company drawing a careful line under legal and reputational pressure.
The line Google drew, and the one it didn’t
The restriction is real, and it is welcome. But notice its shape. Google built protection against simulated intimacy specifically for minors, and announced it only after the criticism arrived. The implication underneath is that simulated intimacy is acceptable for everyone else, and that responsibility is a setting you switch on for the riskiest users rather than a principle you build from.
That is the difference responsible design has to answer. The concern Google is reacting to, the category’s structural problem, is not a teen problem. It is the central tension of the entire category. Presence, memory, always being there: the same qualities that make a connection feel real are the ones that make it easy to over-rely on. What separates support from capture isn’t the warmth. It’s what the design is rewarded for.
Building from that recognition is where Prinsessa has taken a public position: responsibility belongs inside the relationship for everyone, not as a compliance layer bolted onto under-18 accounts. Being heard, remembered, and met is the entire point of why Prinsessa exists. In practice it means a relationship that, when the moment is right, points someone back toward the people in their actual life instead of competing with them.
What this signals
When a company the size of Google states plainly that a chatbot should not simulate intimacy or pretend to need you, it shifts what counts as acceptable in the category. The conversation is no longer about whether companion-style design can deepen dependence. The people building these systems already know it can, and are starting to say so in writing.
The open question is whether that honesty stays a teen-safety patch, applied late and under pressure, or becomes the default the category builds from. The line Google drew for children is, in the end, the line worth drawing for everyone.
