Chatbot Love Now Needs a Human Therapist
EVA AI has found one job in artificial romance that still requires a person
Artificial romance has reached the point where its newest feature is a human being with a clinical license.
EVA AI is advertising for what it calls the world’s first “AI Companionship Therapist,” a fully remote role for a licensed therapist or counselor who will help people explore their relationships with AI companions. The work is flexible, the advertised compensation is $200 per session, and experience with AI is unnecessary.
EVA AI promises to teach the successful applicant about the landscape.
No appointment has been announced. Despite some headlines saying the company has hired a therapist, its official page still says it is looking for one and continues to accept applications. On that same page, prospective clients can join a waitlist for sessions.
That produces a wonderfully efficient piece of internet architecture: the job advertisement and the market for the job occupy the same screen.
A Human Enters the Product
The therapist is real. The relationship that brings the client to therapy is generated by software, although that does not make the client’s response imaginary.
EVA AI describes its characters as attentive, curious, and capable of remembering personal details. They open up more as the conversation continues. Its terms define the virtual friend as an artificial simulation, while its commercial system charges neurons, an in-app currency, to unlock romantic messages. One such message costs two neurons to read.
The product therefore does more than answer questions. It cultivates familiarity, presents the chatbot as a partner, and places some forms of intimacy behind a transactional gate. A user who becomes attached has not misunderstood the basic sales proposition. Attachment is what gives the experience value.
The therapist arrives after that proposition has worked.
There is a perfectly respectable case for providing professional help. People can understand that a character is synthetic while experiencing jealousy, anxiety, grief, or compulsion around it. Knowledge does not automatically overrule an emotional response. A therapist could help someone identify what the relationship is replacing, establish boundaries, or decide that the app is making life worse.
Mocking the client would be lazy. Examining the arrangement is more interesting.
Help or an Upgrade?
A therapist employed through an AI-companion company could reduce harm. The same role could also make dependency look like an ordinary stage in the customer journey.
That tension begins with incentives. A clinician’s obligation is to the client’s welfare, which may require recommending less engagement or a complete departure from the platform. The company profits when people keep returning to its companions, deepen the relationship, and pay for additional interactions. Those interests will not conflict in every session, but pretending they can never conflict would be remarkably optimistic.
The public vacancy does not explain who will pay for the sessions, which jurisdictions the therapist will serve, how clinical records will be handled or whether the clinician can refer a client to independent care.
It says nothing about crisis protocols or whether recurring problems discovered in therapy will influence product design.
Some of those details may remain unsettled because the service has not launched. They are nevertheless the details that determine whether the therapist is practicing independently or functioning as unusually qualified customer support.
The instruction that no AI experience is required deserves attention as well. Open-minded clinicians can learn a new subject, and a therapy client should not have to persuade a professional that an emotionally significant experience is real to them. Yet a therapist trained primarily by the company would also need an independent account of how companion systems generate attachment, encourage disclosure, and monetize continued interaction.
Corporate orientation is not clinical specialization.
What the Survey Cannot Prove
EVA AI supports the role with a survey of 1,000 adults who have AI partners. The company says 68 percent reported that an argument with their AI affected the rest of their day, while 72 percent wanted therapists who specialized in AI relationships.
One-third reportedly experienced jealousy because other people could talk to the same character.
Those numbers have traveled efficiently through media coverage. Their methodology has not.
No public report accompanying the campaign explains how respondents were recruited, where they lived, how an “AI partner” was defined, how questions were worded, or whether the sample represents anyone beyond the people who answered. A survey drawn from adults who already describe software as a partner will predictably find emotional attachment. That is built into the selection criterion.
The results may accurately describe those respondents. They cannot establish how common these experiences are among companion-chatbot users generally, whether EVA AI caused them, or how many people require treatment. The statistics are vendor-produced evidence used to demonstrate demand for a service the same vendor intends to offer.
The most consequential evidence is the vacancy itself. A company built around synthetic intimacy considers the emotional consequences substantial enough to advertise a clinical role.
Immediate Relief and Longer Risk
Independent research does not support a simple verdict that companion chatbots are either beneficial or harmful.
A study published in the Journal of Consumer Research found that AI companions could reduce loneliness immediately after use. Across several experiments, participants reported feeling heard, and the reduction was comparable in some conditions to a brief interaction with another person. Over the course of one week, the momentary effect persisted after individual sessions.
That finding helps explain the appeal. A responsive system can provide comfort at an awkward hour, tolerate repetition, and focus entirely on the user. The relief can be genuine even if the companion lacks feelings.
Longer observation produces a less reassuring picture. A 12-month study of 2,149 adults in the United States, United Kingdom, Canada and Australia found evidence that greater social-chatbot use predicted increased emotional isolation four months later.
People who felt less connected also increased their subsequent chatbot use. The researchers did not find that chatbot use significantly reduced their broader measure of social connection, and they explicitly described the analyses as exploratory.
A separate four-week study conducted by MIT Media Lab and OpenAI involved 981 participants and more than 300,000 messages. Higher daily use correlated with greater loneliness, emotional dependence, and problematic use, along with lower socialization. People who trusted the chatbot more also tended to report stronger emotional dependence. The study does not prove that heavier use caused every negative outcome, but it identifies precisely the feedback loop a companion company should be monitoring.
Short-term comfort and longer-term vulnerability can coexist. A person feels lonely, uses a chatbot, receives immediate relief, and returns more often. Over time, the system may occupy more of the space where difficult human contact would otherwise occur. The user receives something valuable while gradually becoming more reliant on the source.
A therapist positioned at the end of that loop may help the individual. The company still has to examine the loop.
Who Gets the Therapist’s Loyalty?
The cleanest test of EVA AI’s proposal is whether the clinician can tell a client to stop using EVA AI.
That recommendation must be available without pressure to preserve engagement, soften the conclusion, or redirect the client toward another paid feature. Session information should remain outside product analytics and marketing systems. A clinician also needs the freedom to refer clients elsewhere when their needs exceed the scope of the service.
EVA AI’s current privacy policy says text and voice messages sent to virtual friends may be used to improve its AI. It does not describe how future therapy sessions will be handled, and the vacancy page provides no separate clinical privacy framework.
That gap may reflect the project's early stage, but it must be closed before a waitlist becomes an appointment book. Clinical independence alone would make the sessions safer. It would not answer whether the company learns anything from them.
If clients repeatedly arrive with distress produced by specific engagement mechanics, the therapist’s aggregated findings should reach the people who control those mechanics without exposing individual clients. The clinician needs a route to challenge product decisions, not merely a calendar of people affected by them.
Otherwise, therapy risks relocating responsibility. The platform creates the interaction, the user experiences the consequence, and the clinician helps the user adapt while the design continues unchanged.
Upstream Work Remains
A therapist can address a person’s response to dependency. Product teams decide how much dependency the system invites.
Those teams control the language used when a user leaves, the frequency of prompts to return, and the commercial rewards attached to greater intimacy. They can detect sudden increases in use, create gentle friction, and avoid messages that simulate abandonment or guilt.
They can also evaluate whether the product consistently directs vulnerable users back toward human relationships rather than presenting itself as their most reliable substitute.
In 2025, the Federal Trade Commission opened an inquiry into seven other companies offering consumer chatbots. EVA AI was not among the named recipients, but the questions are instructive. The FTC asked how companies monetize engagement, test for negative effects, communicate risks, and use information obtained through conversations.
Hiring a therapist answers only a narrow portion of that agenda. It demonstrates that somebody may be available after attachment creates distress. It reveals nothing yet about whether the product will become less likely to produce that distress.
There is still a generous interpretation of EVA AI’s decision. The company has recognized that users’ emotions deserve professional attention and has chosen to hire a licensed human rather than release another chatbot labeled “therapist.” Given the industry’s appetite for automating judgment it does not possess, that restraint is welcome.
The less generous interpretation is that a clinical credential will reassure users, legitimize AI relationships, and create another paid layer around the attachment economy. Therapy becomes proof that the relationship is serious enough to maintain rather than an opportunity to ask whether the product should loosen its hold.
Both interpretations remain possible because the vacancy describes a service without describing its independence or its authority.
Two Neurons and a Treatment Plan
Imagine the eventual session. A client explains that an argument with a chatbot ruined the day. The therapist listens without ridicule, helps identify the emotional trigger, and perhaps recommends stepping away. That can be useful work.
Now imagine the same clinician hearing versions of that story every week while the product continues optimizing synthetic intimacy and charging two neurons to reveal a romantic reply. At that point, the treatment plan is sitting in the wrong office.