Who benefits from AI mental health companions, and how much?
People who use AI companions for emotional support often see real, measurable benefits. In a study of Headspace's AI tool (Ebb), 93.5% of conversations were rated positively by users, and those who stayed engaged averaged 6.1 sessions per user—more than double the average for the earlier version [2]. This suggests that when the AI is well-designed, users find it helpful enough to keep coming back.
For loneliness specifically, AI companions can be surprisingly effective. One study found that interacting with an AI companion reduced loneliness as much as talking to another person, and more than passive activities like watching YouTube [3]. In a week-long trial, the AI consistently lowered loneliness day by day [3]. The key factor was whether users felt heard by the AI—more important than the chatbot's technical performance [3].
However, not everyone benefits equally. About 28% of community members and 43% of mental health professionals already use AI for mental health support, but 47% of community users and 51% of professionals reported experiencing harms or concerns [1]. The most common worries were about reduced human connection, privacy, and medical errors [1]. So while AI can help many, it's not a universal solution and comes with real risks.
What conditions make AI companions effective—or risky?
The design of the AI matters enormously. Headspace's Ebb tool was purpose-built by clinical psychologists using motivational interviewing techniques and included safety mechanisms to detect risk and escalate care [2]. When they upgraded to version 2.0 with better conversational prompts and safety detection, retention jumped from 28.5% to 50.8% of users completing two sessions within a week [2]. This shows that thoughtful, evidence-based design directly boosts real-world engagement.
But even well-designed tools have limitations. Users in diary studies said they imagined using the AI for stress, anxiety, or during daily routines—not as a replacement for human therapy [2]. Both users and professionals emphasized that AI should be a supplement, not a substitute, for human-delivered care [1][2]. The studies consistently found that people want transparency about data safety, clinical guidelines, and clear limits on what the AI can do [1][2].
The biggest risk is that AI might give harmful advice or fail to recognize a crisis. In the surveys, 51% of mental health professionals reported concerns about medical errors and misuse [1]. The Headspace study addressed this by building in safety detection to flag overuse and escalate urgent cases [2]. Without such safeguards, AI companions could do more harm than good.
Do the improvements last? What the evidence says about durability
The evidence for durable, long-term improvement is thin. The strongest study on durability tracked AI companion use for only one week and found consistent loneliness reduction over that period [3]. That's promising but doesn't tell us if the effect lasts months or years. The Headspace study showed strong engagement over multiple sessions, but it didn't measure clinical outcomes like depression or anxiety over time [2].
What we do know is that about half of users stop using AI companions quickly. In the Headspace data, only 50.8% of users completed a second session within a week, even with the improved version [2]. This high dropout rate suggests that while AI can help in the moment, sustaining engagement—and therefore durable improvement—remains a challenge. The users who did stay engaged (averaging 6.1 sessions) likely got more benefit, but they were a minority [2].
The surveys also reveal a split in expectations: 77% of community users found AI generally beneficial, but an equal mix of positive and negative sentiment emerged about its future role [1]. People see potential for accessibility and cost reduction, but worry that AI could reduce human connection and lead to errors [1]. For durable improvement, the evidence points toward AI being most effective as a bridge to human care, not a standalone solution.
About These Sources
This answer is built on 3 studies (2 peer-reviewed, 1 preprint) — published from 2024 to 2025, 3 from 2024 or later, 1 in Q1 journals, collectively cited 89 times — selected as the most relevant from 3 studies that passed quality screening, drawn from 47 papers retrieved from a database of over 500 million.
Sources used in this answer
Use of AI in Mental Health Care: Community and Mental Health Professionals Survey
In a survey of 107 community members and 86 mental health professionals, 28% of community members and 43% of professionals used AI for mental health; 47% of community users and 51% of professionals reported harms or concerns, mainly about reduced human connection, privacy, and medical errors [1].
Real-World Use of a Mental Health AI Companion: Multiple Methods Study (Preprint)
In a real-world study of Headspace's AI companion (Ebb) across 393,969 members, the improved version (2.0) achieved 50.8% two-session retention within 7 days (vs. 28.5% for 1.0) and 93.5% positive conversation ratings; retained users averaged 6.1 sessions [2].
AI Companions Reduce Loneliness
Across six studies, AI companions reduced loneliness as effectively as interacting with another person and more than passive activities like watching YouTube; a week-long trial showed consistent loneliness reduction, with feeling heard being the key driver [3].