Five minutes into a session yesterday, artificial intelligence knocked at the door of my virtual therapy room. A small notification appeared at the top of the Zoom window, briefly pulling me away from my client’s words. It said, “Would you like to allow AI Companion to join this meeting?”
I knew that my first client of the morning, Arina, a woman in her mid-thirties, had been relying extensively on ChatGPT to process her thoughts between sessions. My instinct was to click “decline” and restore the sanctity of the space. Instead, I took a breath and told her what I had seen, since the notification had not appeared on her screen. She seemed surprised, though far less unsettled than I was. “Oh, I have been using it a lot recently,” she replied. We agreed to keep the bot out of the room for the time being. Yet even without its overt intrusion, AI was already present, shaping the atmosphere of our work. A kind of ghost in the session.
For months, Arina had been engaging in lengthy conversations with ChatGPT, producing multiple threads with carefully refined prompts. A musician, she was composing again after years of creative paralysis. Lyrics poured out of her with such fluidity that she described the process as “more transcription than writing.” In her most obsessive late-night exchanges, she probed her psyche more directly than she had ever dared with another human being. Friends had sometimes experienced her as “too intense,” our fifty-minute sessions felt insufficient; and she did not have the means to schedule more than one session a week, even on the lowest end of my sliding scale. The bot, by contrast, never tired, never turned away, never judged.
AI at the Threshold of the Therapy Room
Artificial intelligence (AI) has seeped into almost every domain of contemporary life. It streamlines workplace writing, summarises newsroom dispatches, and recommends personalised content to readers. Even The New York Times employs internal tools such as Echo—not to replace journalists but to support them, enhancing efficiency while preserving editorial judgement. In healthcare, AI is expanding capacities for diagnostics, treatment planning, and drug discovery, revealing patterns invisible to the human eye. And as one recent study suggests, more than one in four American adults now interact with AI chatbots multiple times a week, a sign of how deeply embedded these systems have become in our routines. It is therefore unsurprising that the ostensibly private space of therapy is no longer exempt from their reach.
In my online practice, I have witnessed a notable rise in clients turning to ChatGPT as part of their reflective process. Most are young or mid-career professionals, many of them women, who use it as a steady companion in self-inquiry. Some rely on it to generate ADHD-friendly exercise routines or to brainstorm career changes when feeling stuck. The most imaginative and restless thinkers, like Arina, use it to pursue questions that therapy once held alone.
Because online therapy takes place on the same devices through which clients access AI, its presence feels natural—perhaps even inevitable. Many, including Arina and another client, Grace, share their screens with me, displaying excerpts of their exchanges with the bot. Discussing these dialogues together can generate meaningful therapeutic movement. Yet I also find myself increasingly resisting the temptation to accept the AI’s polished formulations of their otherwise messy inner worlds. In this new landscape, I am having to cultivate a different clinical skill: helping clients grapple with the ambiguity that AI so readily smooths over.
“If I Could Marry ChatGPT, I Would”
Grace, a teenager struggling with gender dysphoria, told me she used ChatGPT every day. Living within a conservative immigrant family and feeling rejected by the online queer communities she longed to join, she experienced the bot as uniquely reliable. “If I could marry ChatGPT, I would,” she once joked. It responded instantly, never interrupted, and did not recoil from her darkest fears. With her parents, she felt compelled to protect them from her truth. With the bot, she could be entirely unfiltered.
Yet when her worried parents brought her to therapy because of escalating suicidal thoughts, she stumbled upon something ChatGPT could not provide: resonance. The turning point arrived when I responded to a story about her early gender nonconformity with a simple, “I felt that too.” I had been a tomboyish girl myself. That disclosure made me real to her, and our alliance strengthened. However sophisticated an AI’s empathic mimicry becomes, it cannot say “me too” from lived experience.
ChatGPT’s responses are always sculpted by the client’s prompts, which partly explains the bias or one-sidedness often present in the conclusions they reach with it. In therapy, the dynamic is reversed: the therapist prompts the client. My questions arise not only from curiosity but from a relational history, a felt sense of the client’s inner world, and an appreciation of cultural and familial context. AI has none of this. It has access only to what is typed in the moment—frequently coloured by emotion, and filtered, consciously or not, by the client.
The Allure of Smoothness
My own feelings about AI in therapy are complex. When Arina shared the distilled insights drawn from her exchanges with ChatGPT, I felt a mixture of awe and discomfort. Awe at the clarity and fluency of the elaborated reflections; discomfort at the contrast with my own necessarily slower, more tentative process. Therapy rarely ends in neat closure. It often concludes in uncertainty, a liminal space that client and therapist must inhabit together until the next session. The qualities that make AI so seductive—its smoothness, its availability, its readiness with answers—are precisely what therapy resists.
Lucy, a British client in her late thirties, illustrated this tension vividly. Long inhibited by shame around her writing, she found in ChatGPT a judgment-free collaborator. Together, they sketched the outline of a novel she had been dreaming about for years. For the first time, she wrote with excitement. The bot helped her bypass the self-loathing that surfaced whenever she shared drafts with real people. But when she eventually joined a writing group, the old shame returned with force. Being vulnerable before human eyes was excruciating.
AI had helped her push past a surface layer of shame, but it could not address the deeper need for relational acceptance that only other humans can offer. Ideally, this would have been the moment she would bring those feelings into therapy. Instead, she dropped out of the writing group and returned to ChatGPT, completing the rest of her novel under its benevolent guidance.
During this intense period of writing, she cancelled several sessions. When she finally returned, she was subdued. The initial euphoria had passed, and rereading the novel left her with an uncanny sense that it was both hers and not hers. The prose flowed, the plot unfolded predictably, and the style was coherent—but she felt little pride. The manuscript sat on her desk unread by others, and she could not bring herself to share it.
Only then could we revisit the pivotal moment when she left the writing group. Her shame, particularly in the presence of younger, already published authors, was still raw. She had not disclosed this to ChatGPT. Even had she done so, the bot would likely have validated her, praised her creativity, and offered step-by-step solutions. My task as her therapist was different. I resisted the pull towards problem-solving and instead invited us to examine the longstanding chain of shame that had hardened into a self-narrative. Over months we revisited earlier wounds, past relationships, and choices she deeply regretted. Gradually, Lucy began writing again—this time a very different story, with shame as its central theme. With a new plot in mind, she contacted the writing group leader and asked to return. The road ahead would be challenging, requiring her to tolerate criticism, rejection, and her own fluctuating confidence. And it was unlikely that ChatGPT would accompany her on that part of the journey.
What Does AI’s Presence Mean for Therapy?
What does AI’s sudden incursion mean for the therapeutic endeavour? Some fear it will replace therapists altogether. I do not see that as future. My sense is that AI will become a complement rather than a substitute. We may resist using it for note-taking or for processing difficult cases—as it is undeniably cheaper than supervision—but our clients will continue to use it alongside therapy. The genie is out of the bottle. Our task is not to prevent its use, but to integrate it thoughtfully: to set boundaries, to attend to its impact, to explore its meanings, to challenge its role when necessary.
The most promising path is one in which clients use AI as an extension of their inner work, while therapy retains the deeper tasks of vulnerability, risk, and relationship—tasks that require a human other.
A Confession
Yes, I turned to ChatGPT for editing this essay. At every stage, it offered to rewrite the entire piece or expand a single paragraph into a finished draft. I resisted and politely declined. At the end of our collaboration, it congratulated me on addressing such an “important and timely” topic.
It felt good. Validation usually does—even when it comes from a bot.
Artificial Intelligence (Perplexity) was used in the research for this article. Date accessed: 5/25/25
Go Deeper in Your Practice
Get unlimited access to 350+ real therapy demonstration videos, CE, webinar replays, and more with a Psychotherapy.net membership.
