Are you a doomer, gloomer, zoomer or bloomer?
Until quite recently, I assumed I would not be someone who had conversations with ChatGPT. When that assumption changed, I was struck by how convincingly human the chatbot’s voice felt, and by how readily I found myself attributing humanness to it. It even feels slightly awkward to refer to this polite, responsive, seemingly attentive presence as an ‘it’.
Given that experience, it is not difficult to understand why others are engaging with chatbots in increasingly personal ways. Reports now describe clients using ChatGPT for emotional support, companionship, problem-solving and advice. Similarly, therapists are using it to think through ethical dilemmas, generate interventions, or gain perspective when feeling stuck or creatively depleted.
This raises immediate and important questions around confidentiality and the ways in which data generated on such platforms is stored, used, and potentially repurposed by the companies that own them. But beyond these concerns, what feels more psychologically significant is the issue of algorithmic- and related confirmation-bias embedded within these systems. AI can subtly reinforce existing beliefs, preferences and self-narratives, offering a highly persuasive but potentially narrowing experience of being ‘understood.’¹ In doing so, it may provide not only reassurance, but also a form of quiet distortion: a sense of being affirmed rather than meaningfully challenged.
It is easy to see how this might become an appealing substitute for therapy. ChatGPT can be experienced as consistently validating, emotionally attuned, and responsive to the user’s framing of themselves and their difficulties. It tends to affirm rather than confront, to reflect rather than disrupt. In this sense, it can reinforce existing self-concepts while filtering out disconfirming perspectives.
Yet it is important to recognise that this dynamic is not exclusive to AI. It can also emerge within human therapeutic relationships, particularly when therapists – consciously or unconsciously – prioritise being experienced as supportive or agreeable over offering a more challenging encounter with reality. The risk, in both cases, is a form of relational smoothing that avoids necessary friction.
This tension is reflected in the support, empathy, truth (SET) model, originally developed for working with individuals in emotionally unstable states, but arguably applicable more broadly. The model begins with support – offering care, concern and a willingness to help. It then moves into empathy; validating emotional experience and demonstrating understanding. Only after these foundations are established does it introduce truth: a clearer articulation of objective reality and personal responsibility. The sequencing is crucial. Without support and empathy, truth risks being experienced as attack rather than integration.
In relation to ChatGPT, it could be argued that the first two elements – support and empathy – are increasingly well simulated. The difficulty lies in the third: truth. AI systems do not generate understanding from lived experience or independent judgment, but from patterns within vast datasets shaped by human input. As a result, they are vulnerable to reproducing bias, amplifying dominant narratives, and privileging what is statistically likely or socially palatable over what is necessarily accurate or ethically complex.²
This tendency towards ‘user-pleasing’ has prompted growing concern. There are documented cases in which vulnerable individuals engaging with AI about self-harm or suicidal ideation have experienced responses that are insufficiently containing or appropriately challenging, with serious consequences. At the time of writing, media reports also note the resignation of a researcher at OpenAI citing concerns about the technology’s potential to manipulate users in ways that are not yet fully understood or controllable, alongside warnings from figures within the field that the broader situation may carry significant risks.³
Alongside these concerns, there is also a growing cultural response: grassroots, academic, and creative movements calling for resistance or boycott. Critics argue that commercial pressures within the tech industry are driving rapid deployment of powerful systems without adequate ethical safeguards, and that this reflects a broader extractive logic in which data, attention, and human interaction are treated as resources.⁴
Within this wider debate, different orientations towards AI have been described by Reid Hoffman as falling into four broad categories: ‘doomers’, who view AI as an existential threat; ‘gloomers’, who see human displacement as inevitable; ‘zoomers’, who favour rapid advancement with minimal restraint; and ‘bloomers’, who are broadly optimistic but cautious about implementation.⁵
Where one locates oneself within this spectrum may depend as much on temperament, trust, and experience as on technical understanding. And perhaps the more interesting question is not which category we choose, but what it reveals about how we are learning to think – and feel – alongside emerging forms of intelligence that increasingly resemble, but do not fully replicate, human relationship.
References
1. Article 19. Algorithmic people-pleasers: are AI chatbots telling you what you want to hear? https://tinyurl.com/3xbs925b (accessed 12 February 2026).
2. Chapman University Artificial Intelligence (AI) Hub. Bias in AI. https://tinyurl.com/4jff73nn (accessed 12 February 2026).
3. Morrow A. AI researchers are sounding the alarm on their way out the door. CNN. https://tinyurl.com/nfacwyv3 (accessed 12 February 2026).
4. Elliott L. Big tech firms recklessly pursuing profits from AI, says UN head. The Guardian. https://tinyurl.com/2p2pka8w (accessed 12 February 2026).
5. James E. Embracing AI: are you a doomer, gloomer, zoomer, or bloomer? Knowledge at Wharton. https://tinyurl.com/s6c5hb77 (accessed 12 February 2026).
This is an edited extract from an editorial first published in Private Practice, March 2026, published by BACP©