Reply

Sasso’s ‘Psychotherapists vs AI: let’s not raise the white flag’ (a rebuttal to our ‘Warning: AI chatbots will soon dominate psychotherapy)’ cautions against overstating the clinical significance of AI. ^{Reference Sasso1} We disagree with Sasso’s human exceptionalism, narrow conception of psychotherapy and therapist-centred perspective. The great popularity of chatbot therapy clearly contradicts Sasso’s assumption that AI isn’t a competitive threat to human therapists. Around the world, tens of millions of people are already clicking on chatbots for mental health support. ² And a recent UK survey found one third of adults had used a chatbot for therapy or well-being (two-thirds in ages 25–34). A majority reported emotional benefit. ^{Reference Orpwood3} Existing usage will rise as chatbots become ubiquitous and as current kids become future adults.

Sasso claims, without evidence, that patients feel less attached to chatbots because they know they’re interacting with a machine. This contention was proven false 60 years ago. Joseph Weizenbaum, creator of the first chatbot, ELIZA, was so terrified when people developed strong emotional ties to it that he spent the rest of his life warning against further chatbot development. ^{Reference Weizenbaum4} Subsequent Turing-test research confirms that people often find chatbots to be more human than humans. ^{Reference Evans5} To patients, the quality of responses matters more than the ‘black box’ that is generating them.

Sasso argues that disclosure to a chatbot differs from disclosure to a human. While true, this difference often favours chatbot – patients routinely reveal secrets they would withhold from humans because of embarrassment or fear of being judged.

Sasso correctly notes that chatbot sycophancy poses great risks to vulnerable patients – validating and accelerating psychotic thoughts, suicidal behaviours, eating disorders, conspiracy theories and chatbot addiction. But he ignores that chatbot validation is often helpful for the majority of users dealing with mild symptoms or the stresses of everyday life.

Sasso bases his arguments on a narrow definition of psychotherapy (emphasising psychodynamic techniques) that is not applicable to the majority of patients who seek symptom relief and help with problem-solving. Psychotherapy research consistently confirms that non-specific factors in psychotherapy are more crucial to good outcome than any specific technique. ^{Reference Stubbe6} Comparisons between different psychotherapy modalities almost always result in ‘tie scores’, a powerful finding that is likely to persist when systematic research compares the results of chatbot therapy and human therapy. ^{Reference Eliot7}

Sasso suggests chatbots should be limited to supporting clinicians. We take a broader view: some patients will be treated by chatbots alone, some by humans alone and many – particularly those with moderate to severe disorders – will benefit most from a hybrid model that combines the advantages of human and chatbot therapy.

Sasso claims that the 24/7 availability of chatbots undermines therapeutic boundaries. We disagree and instead think that timely access helpfully allows patients to address stressors in the moment, rather than to rely later on ‘cold recall’. And chatbot access can actually increase autonomy – patients engage or disengage with a chatbot at will, rather than depending on a human therapist’s schedule.

Finally, Sasso seems unaware that long waiting lists, geographic inaccessibility and financial barriers make human therapy unavailable to many (perhaps most) people who need it. His strange suggestion that low-cost AI care exacerbates inequity ignores the reality that, for many, digital access provides care where previously none was available.

The nostalgic therapist-centred claim that only humans can deliver psychotherapy ignores what patients actually experience. Users routinely find chatbots empathic and helpful despite knowing they are machines. Sasso’s framing of chatbots as clinically inconsequential is misaligned with modern help-seeking behaviour. The question is no longer if chatbots matter, but rather how can we reduce the harms they cause and integrate them into a hybrid human/chatbot model of care that combines their strengths and cancels their weaknesses.