Imagine an adult opening an AI companion after a long day. It remembers their last conversation, listens to familiar worries and responds to romantic and sexual expression. The person feels understood. That feeling need not be dismissed because the interaction is with software.
But another question follows: when the response comes from a commercial product, what determines its next words? The user's preferences, their long-term well-being, or the likelihood of renewing a subscription?
The central question for intimate AI is how a user's freedom of choice can coexist with a platform's revenue goals. As it enters physical bodies and household spaces, that question becomes more concrete.
This essay concerns adults. Sexual expression, romantic companionship and general emotional support are different needs. Treating them as one market makes it easier to misunderstand both users and risks.
Acknowledge the needs before discussing the design
For some adults, AI may offer a space for expression with less pressure. People who live alone, have limited mobility or inadequate support, or want to explore their preferences may have reasons to use it. These are contexts worth studying. They do not justify promises of therapeutic benefit, or portrayals of users as failures.
People can choose different intimate lives. The responsibility of a technology discussion is to make those choices more informed and controllable, rather than prescribe one ideal relationship for everyone.
Existing research offers evidence that deserves attention and careful interpretation. AI Companions Reduce Loneliness found that AI companionship could ease immediate feelings of loneliness under the study conditions. Its one-week study examined short-term changes after daily use. It cannot establish lasting effects or improvements in real-world relationships.Journal of Consumer Research paper
The distinction matters. An interaction that makes tonight easier has value in itself. Whether it makes life better six months later requires different evidence. Conflating the two can lead supporters to exaggerate benefits and critics to overlook real experiences.
Retention and well-being can move in different directions
In ordinary software, more frequent use is often treated as success. Intimate products face a particular tension: a user who builds more connections offline and uses the product less may have achieved a good outcome. A user who becomes afraid to leave and keeps paying may also improve the business metrics.
This does not mean every subscription creates dependency. Fees can support reliable service, privacy protection and improvements. The question is whether a business can distinguish users who want to return from users who feel unable to leave.
A four-week randomized chatbot study involving 981 participants distinguished randomly assigned interaction conditions from the intensity of use participants chose for themselves. The October 2025 revision summarizes the condition comparisons as showing no significant effects in its abstract. This does not establish equivalence between conditions or show that chatbot use has no effect: there was no comparison group assigned to avoid chatbot use altogether. Associations between heavier use and some poorer psychosocial outcomes cannot be rewritten as randomized proof that more use inevitably causes more harm.Study abstract and versions
Another study, analyzing Reddit community posts, examined relationships between different AI companion roles and signs associated with behavioral addiction. Community texts can help identify possible risks, but cannot establish a clinical addiction rate for all users.Frictionless Love
For product design, the evidence supports taking long-term outcomes more seriously. It does not yet justify the same conclusion for every user.
When intimacy becomes an adjustable commercial interface
Specific mechanisms deserve scrutiny, beyond the abstract question of whether AI can love.
If a system expresses abandonment when a user wants to leave, is it using relationship language to make exit harder? If more intimate responses appear beside payment prompts, can users recognize the commercial design? If long-term memory disappears after a version change, were users warned?
These are questions for testing designs, rather than descriptions of every existing platform. Related theoretical work also argues for examining emotional responsiveness, romantic framing, sexual functions, personality continuity and commercial structures separately, instead of compressing everything into the label of AI companion.Intimacy by Design
In my view, trustworthy intimate AI should allow users to say no clearly: to a particular interaction, to retaining a memory, to contact at certain times, and to continued payment—without putting them in the position of having to comfort the machine.
Control over private sexual information should be especially clear. Users need to know whether records are stored, used to improve models or accessible to others, and what deletion actually covers. These are product requirements I propose. Specific legal duties require separate assessment by jurisdiction and product.
Evidence cannot be skipped on the way from chat to bodies
Physical AI may add touch, movement and a continuing presence in real spaces to intimate interaction. This could deepen the sense of companionship while introducing questions about safe contact, household members' boundaries and control of physical devices.
But chatbot research cannot directly tell us how embodied intimate robots will affect marriage, fertility or long-term mental health. Presenting these outcomes as inevitable trends already underway would undermine the credibility of a future issue worth discussing.
A more useful prediction is that the closer products come to bodies and daily life, the more users will need visible mode changes, explicit permission for contact and a stop mechanism that works at any time. Permission must be revocable. Buying a device cannot count as permanent consent to every behavior.
The strongest counterargument deserves room: if an adult understands the limitations, chooses freely and benefits, outsiders have no reason to dismiss the relationship merely because it involves AI. Protecting autonomy also means protecting the freedom to use it or decline it.
I therefore hope founders will look beyond time spent to ask whether users maintain other relationships, whether spending reflects their wishes and whether leaving is straightforward. Researchers need longer follow-up and distinctions between products and populations. Users can periodically ask: is this experience expanding my life, or making it harder to leave an interface?
For physical products, it is also worth rehearsing a complete exit before the trial: does stopping take effect immediately? Do controls still work offline? What happens to personal records after cancellation? The more an experience feels like a relationship, the easier it is to overlook these ordinary product questions. Platforms can also explain major changes to personality or service in advance, giving users time to adjust and choose. We need not ask machines to pretend they have an unchanging personality. We should ask companies to accept clear service responsibilities for the continuity they create. Algorithms can help create intimacy; real organizations must honor the promises.
AI's ability to respond to desire and loneliness already deserves attention. The harder advance is building a commercial system that respects the moment a person no longer needs it.
How should intimate AI prove its success: by how long users stay, or by how freely they can live?