What does the current evidence actually measure?
The strongest evidence available comes from a large real-world study of Headspace's AI companion, which tracked 393,969 users — but only for a few days. The study measured whether people came back for a second session within 7 days, not whether they got better or stayed well over months [1]. That is a very short window. Across all the studies here, the longest follow-up is cross-sectional (a single snapshot in time), not a true longitudinal trial [5]. So when you ask about 'long-term follow-up,' the honest answer is that the research hasn't done it yet.
Another study surveyed 107 community members and 86 mental health professionals about their AI use. While 77% of community users found AI generally beneficial, 47% reported experiencing harms or concerns — a significant minority [2]. This tells us that even in the short term, the experience is mixed, and we don't know how those harms or benefits evolve over time.
Does high engagement mean the AI is working long-term?
Not necessarily. The Headspace study showed that an upgraded version of their AI companion (version 2.0) improved retention: 51% of users completed two sessions within 7 days, compared to 29% for the older version [1]. Users also rated conversations positively 94% of the time. But engagement and satisfaction are not the same as clinical improvement. A diary study within the same paper found that users imagined using the tool for stress, anxiety, and daily routines — not for serious or ongoing mental health conditions [1]. So the AI may be a pleasant short-term companion, but we lack data on whether it changes mental health outcomes over weeks or months.
A separate study of 14,721 Japanese adults found that AI companion use was linked to higher life satisfaction, happiness, and purpose — but only in people who were already lonely or moderately socially connected [5]. This suggests AI companions might fill a gap for some people, but the study was a cross-sectional survey, not a controlled experiment. It cannot prove that AI caused the well-being, nor can it say how long any benefit lasts.
What about safety and risks — are they tracked long-term?
Safety monitoring in the current studies is limited to immediate risk detection, not long-term follow-up. The Headspace AI includes safety guardrails to detect crises and flag the need for human escalation, but the study did not report how often those flags occurred or what happened afterward [1]. A 2023 commentary warned that general-purpose chatbots like ChatGPT are 'not yet ready for clinical use' because they lack validation for real-world healthcare tasks [6]. And a 2024 study found that an older version of ChatGPT gave overly pessimistic prognoses for schizophrenia, which could discourage patients from seeking treatment [3]. These are exactly the kinds of risks that need long-term monitoring — and we don't have it.
Public discussion on Reddit, analyzed in a 2025 study, shows that users appreciate AI for emotional support and accessibility but also worry about adverse effects and lack of conversational depth [4]. Experienced AI users acknowledged both benefits and limitations, while mental health communities specifically highlighted the lack of deep connection. These concerns are real, but no study has tracked whether they lead to harm or dissatisfaction over months of use.
About These Sources
This answer is built on 6 peer-reviewed studies — published from 2023 to 2026, 5 from 2024 or later, 4 in Q1 journals, collectively cited 199 times — selected as the most relevant from 6 studies that passed quality screening, drawn from 57 papers retrieved from a database of over 500 million.
Sources used in this answer
Real-World Use of a Mental Health AI Companion: Multiple Methods Study.
In a real-world study of 393,969 Headspace members, the AI companion showed strong short-term engagement (51% returned for a second session within 7 days with the upgraded version), but the study did not measure clinical outcomes or follow-up beyond days.
Use of AI in Mental Health Care: Community and Mental Health Professionals Survey.
A survey of 107 community members and 86 mental health professionals found that while 77% of community users found AI beneficial, 47% reported experiencing harms or concerns, highlighting mixed short-term experiences.
Comparing the Perspectives of Generative AI, Mental Health Experts, and the General Public on Schizophrenia Recovery: Case Vignette Study
A vignette study comparing four large language models (LLMs) to mental health professionals found that ChatGPT-3.5 gave overly pessimistic prognoses for schizophrenia, which could reduce patient motivation for treatment.
The ChatGPT Effect: Investigating Shifting Discourse Patterns, Sentiment, and Benefit–Challenge Framing in AI Mental Health Support
Analysis of 517 Reddit posts showed that users value AI for emotional support and accessibility but worry about adverse effects and lack of conversational depth; experienced users acknowledged both benefits and limitations.
AI companions and subjective well-being: Moderation by social connectedness and loneliness
A cross-sectional survey of 14,721 Japanese adults found that AI companion use was associated with higher well-being, especially among lonely individuals, but the study cannot prove causation or track long-term effects.
AI chatbots not yet ready for clinical use
A 2023 commentary argued that general-purpose chatbots like ChatGPT are not yet ready for clinical use due to lack of validation for real-world healthcare tasks, emphasizing the need for rigorous testing.