How strong is the placebo effect in AI mental health tools?
The placebo effect in AI interactions is surprisingly powerful and stubborn. In a 2024 study, participants performed a letter discrimination task while believing an AI was adapting the interface to help or hinder them—but in reality, no AI was ever present. The result: participants performed descriptively better regardless of whether the AI was described as helpful or harmful, and their expectations remained high even after negative descriptions [2]. This suggests that once people believe an AI is involved, their performance expectations are biased upward and resistant to correction, meaning the mere presence of an AI label can boost outcomes through expectancy alone.
Do users expect too much from AI companions?
Yes, users consistently expect more from human doctors than from chatbots, but this gap creates a surprising opportunity for AI. In a 2024 experiment with 364 participants, researchers found that people had higher expectations of human doctors across almost all aspects of mental health communication, but those expectations were more likely to be violated when interacting with a human doctor [3]. In contrast, chatbots that used an emotionally warm (affective) approach actually met or exceeded users' lower expectations, making participants more willing to switch to the chatbot for mental health support [3]. This means that while AI companions may not match human doctors in absolute performance, they can outperform expectations—and that expectancy violation is a key driver of user satisfaction and behavioral change.
Can AI mental health tools also cause harm through expectations?
Yes, the same expectancy mechanisms that create placebo benefits can also produce nocebo effects—negative outcomes driven by negative expectations. A 2025 viewpoint paper argues that generative AI in mental health care could amplify both placebo and nocebo effects through subtle cues in tone, accuracy, or loss of human nuance in automated responses [1]. For example, if an AI companion's language is slightly off or impersonal, it might inadvertently heighten a user's anxiety or distrust, undermining the therapeutic benefit. The paper emphasizes that these effects are especially potent in mental health, where outcomes are tightly linked to patients' perceptions of provider competence and empathy [1]. This means that the same expectancy bias that can make AI seem more effective can also make it seem less effective—or even harmful—if the user's expectations are mismatched with the AI's actual performance.
About These Sources
This answer is built on 3 peer-reviewed studies — published from 2024 to 2025, 3 from 2024 or later — selected as the most relevant from 3 studies that passed quality screening, drawn from 54 papers retrieved from a database of over 500 million.
Sources used in this answer
Placebo, Nocebo, and Machine Learning: How Generative AI Could Shape Patient Perception in Mental Health Care.
This 2025 viewpoint paper argues that generative AI in clinical communication can amplify both placebo and nocebo effects in mental health care, especially through tone, assurance, and response speed, and calls for transparency and ethical oversight as these technologies evolve.
"AI enhances our performance, I have no doubt this one will do the same": The Placebo effect is robust to negative descriptions of AI
In a 2024 experiment with a letter discrimination task, participants performed better when they believed an AI was present (even though none was), and this placebo effect persisted even when the AI was described negatively, showing that performance expectations with AI are biased and robust to negative verbal descriptions.
AI-powered mental health communication: Examining the effects of affection expectations on health behavioral intentions.
In a 2024 online experiment with 364 participants, people had higher expectations of human doctors than chatbots, but chatbots using an affective (warm) approach were more likely to meet expectations and increase users' willingness to switch to the chatbot for mental health support, compared to neutral-toned human doctors.