New research shows that as conversations with Claude, Anthropic’s AI chatbot, progress, users’ expressed sentiment “becomes more positive,” and they often walk away feeling emotionally lighter.
The finding comes from Anthropic’s study on how users are turning to Claude for emotional support, companionship, and interpersonal advice, which the chatbot wasn’t explicitly built for. The results suggest that AI may help prevent the reinforcement of negative emotional patterns, although further research is needed to determine whether positive shifts persist beyond individual conversations.
Career changes, relationships, and coping with uncertainty
Anthropic analyzed 131,484 emotionally driven conversations out of 4.5 million Claude chats to better understand how people seek support from the AI.
Just 2.9% of conversations with the AI chatbot were classified as affective, which means involving emotional or psychological needs such as coaching, counseling, or companionship. Topics like career changes, relationship struggles, and personal uncertainty were most common.
Companionship and roleplay made up less than 0.5% of all chats, with romantic roleplay accounting for under 0.1%. This reflects Claude’ design that actively discourages romantic or sexual interactions.
Unpacking the uplift: Why Claude conversations feel good
Discussions with Claude often leave people feeling better based on users’ language growing more positive as the discussion continues. This trend was most consistent in affective interactions, including coaching sessions focused on motivation and goal-setting, and counseling chats centered on anxiety, loneliness, or stress.
Longer conversations, or those with 50 or more human messages, tended to become more personal, giving users room to process deeper emotions. Topics usually moved from surface-level concerns to trauma, purpose, or existential questions.
Claude’s low resistance may also contribute to these results. Fewer than 10% of affective chats involved pushback, allowing conversations to flow without interruption. This openness helped avoid reinforcing negative beliefs and encouraged emotional momentum.
Anthropic cautions that these changes represent short-term sentiment, not clinical outcomes. The analysis measured expressed language, not lasting psychological states or well-being.
The emotional gray zone in human-AI connection
Anthropic acknowledges that as Claude’s intelligence increases, emotionally rich conversations bring new challenges.
The AI model’s consistent empathy and lack of emotional fatigue may affect how people experience support and alter expectations in real-world relationships. Some “power users” engage in long, personal exchanges that resemble companionship more than assistance.
To address this, the AI company is working with crisis support organization ThroughLine to refine how Claude handles sensitive matters and guides users toward real-world help when needed. While the AI tool is not intended to replace mental health professionals, Anthropic is building safeguards and referral systems to ensure that emotional support from artificial intelligence stays within healthy boundaries.
The company is also planning to study extreme usage patterns, like emotional dependency, sycophancy, and the risk of reinforcing delusional thinking or harmful beliefs. These early findings mark the start of Anthropic’s broader effort to create emotionally aware AI systems that support, rather than replace, genuine human connection.
Read eWeek’s coverage about how some Instagram chatbots make up therapy credentials, and that ChatGPT and other chatbots have had negative impacts on users’ mental health, including some tragic consequences.