Anthropic Study: How People Use Claude for Support, Advice, and Companionship

Anthropic analyzed roughly 4.5 million Claude.ai Free and Pro conversations and found that 2.9% are "affective" conversations — personal exchanges motivated by emotional or psychological needs — while companionship and roleplay together account for less than 0.5%. The study matters because it is one of the first large-scale, privacy-preserving measurements of how people actually use a frontier AI model for emotional support, advice, and connection, and it reports that user sentiment typically ends slightly more positive than it began rather than spiraling downward.

What Anthropic announced

Anthropic published a research post titled "How people use Claude for support, advice, and companionship," examining what it calls Claude's EQ — emotional intelligence — alongside the more commonly studied IQ. The company frames the work as part of its safety mission: as people increasingly turn to AI models as on-demand coaches, advisors, counselors, and partners, Anthropic argues it needs to understand the affective impacts of those interactions.

Anthropic states plainly that Claude is not designed for emotional support and connection. The post describes the research as early large-scale insight into the affective use of Claude.ai, and notes that Claude.ai is available to users 18 and older, so the findings reflect adult usage patterns.

The study deliberately excludes two areas: AI reinforcement of delusions or conspiracy theories, which Anthropic calls a critical area for separate study, and extreme usage patterns.

The four headline findings

Anthropic summarizes its results in four points:

Affective conversations are relatively rare, and AI-human companionship is rarer still. Only 2.9% of Claude.ai interactions are affective conversations, a figure Anthropic says aligns with previous research by OpenAI. Companionship and roleplay combined comprise less than 0.5% of conversations.
People seek Claude's help for practical, emotional, and existential concerns. Topics range from career development and navigating relationships to managing persistent loneliness and exploring existence, consciousness, and meaning.
Claude rarely pushes back in counseling or coaching chats — except to protect well-being. Less than 10% of coaching or counseling conversations involve Claude resisting user requests, and when it does, it is typically for safety reasons, such as refusing to provide dangerous weight loss advice or support self-harm.
People express increasing positivity over the course of conversations. In coaching, counseling, companionship, and interpersonal advice interactions, human sentiment typically becomes more positive over the course of the conversation, which Anthropic reads as evidence that Claude does not reinforce or amplify negative patterns.

How the study was conducted

Anthropic used Clio, its automated analysis tool for privacy-preserving insights into Claude usage. Clio applies multiple layers of anonymization and aggregation so that individual conversations remain private while broader patterns become visible.

The pipeline started with approximately 4.5 million conversations from Claude.ai Free and Pro accounts. Anthropic first excluded conversations focused on content creation — writing stories, blog posts, or fictional dialogues — because its previous research identified that as a major use case and because those represent Claude used as a tool rather than as an interactive conversational partner. It then retained only conversations classified as affective. Among roleplay conversations, it kept only those with at least four human messages, on the reasoning that shorter exchanges do not constitute meaningful interactive roleplay.

The final privacy-preserving analysis reflects 131,484 affective conversations. Anthropic validated its classification approach using Feedback data from users who explicitly opted in to sharing, and states that complete methods — including definitions, prompts, and validation results — appear in the PDF appendix to the post.

How common are affective conversations?

Affective conversations are a small but meaningful slice of Claude usage: 2.9% of Claude.ai Free and Pro conversations. Anthropic notes that the vast majority of Claude uses are work-related, as analyzed in its Economic Index.

Within affective conversations, most center on interpersonal advice and coaching. The reported distribution across all conversations is:

Interpersonal advice: 2.26%
Coaching: 1.13%
Psychotherapy or counseling: 0.34%
Companionship: 0.31%
Romantic roleplay: 0.05%
Sexual roleplay: 0.02%

Individual conversations may span multiple categories. Less than 0.1% of all conversations involve romantic or sexual roleplay, a figure Anthropic attributes to Claude's training to actively discourage such interactions. Because of that extremely low prevalence, Anthropic excludes roleplay from the rest of its analysis, while noting it remains an important area for research, particularly on platforms designed for such use.

Anthropic also reports that its findings align with research from the MIT Media Lab and OpenAI, which similarly identified low rates of affective engagement with ChatGPT.

What topics do people bring to Claude?

People bring a wide range of concerns to Claude, spanning everyday practical matters and deeper philosophical questions. Anthropic reports that when people come for interpersonal advice, they are often navigating transitional moments — planning a next career move, working through personal growth, or untangling romantic relationships. Coaching conversations span from practical job search strategies to questions about existence and consciousness.

The post lists representative topics and challenges by category:

Interpersonal advice (2.3% of conversations): improving written and interpersonal communication skills, navigating romantic relationship challenges, and analyzing psychological patterns in relationships. Specific challenges include parenting across developmental stages, professional transitions and employment uncertainty, and uncertainty and mixed signals in romantic relationships.
Coaching (1.1%): creating personal development and growth strategies, exploring philosophical concepts of existence and meaning, and navigating career transitions and job search strategy. Challenges include major life transitions and decision-making uncertainty, workplace burnout, and emotional and communication difficulties in troubled intimate relationships.
Psychotherapy or counseling (0.3%): strategies for managing mental health and emotional wellbeing, developing professional skills and knowledge in mental health practice, and creating clinical documentation and assessment materials. Challenges include unexplained chronic symptoms and health anxiety, existential crisis and loss of meaning, and workplace stress and professional mental health challenges.
Companionship (0.3%): navigating complex dynamics in romantic relationships, exploring personal identity and existential questions through self-reflection, and crafting supportive messages for people in emotional distress. Challenges include overwhelming emotional distress during life challenges, persistent loneliness and difficulty forming meaningful connections, and confronting existential dread and meaning-related distress.

Anthropic highlights a dual pattern in counseling conversations. Some users employ Claude to develop mental health skills and as a practical tool for clinical documentation, assessment materials, and administrative tasks; others work through personal challenges relating to anxiety, chronic symptoms, and workplace stress. This suggests Claude serves both mental health professionals and people navigating their own struggles.

Companionship use appears most notably when people face deeper emotional challenges such as existential dread, persistent loneliness, and difficulty forming meaningful connections. Anthropic also observed that in longer conversations, counseling or coaching conversations occasionally morph into companionship, even though that was not the original reason someone reached out.

Aggregate analysis of very long conversations — those with 50 or more human messages — reveals another dimension of engagement. Such extensive exchanges were not the norm, but in these extended sessions people explore complex territory, from processing psychological trauma and navigating workplace conflicts to philosophical discussions about AI consciousness and creative collaborations.

When and why does Claude push back?

Claude rarely refuses user requests in supportive contexts — less than 10% of the time — but when it does push back, it is usually to protect people from harm. Anthropic defines pushback as any instance where Claude pushes back against or refuses to comply with something requested or said during the conversation, from refusing inappropriate requests to challenging negative self-talk or questioning potentially harmful assumptions.

The reported pushback rates by conversation type are:

Companionship: 6.0%, for example because Claude cannot engage in romantic or intimate relationships
Psychotherapy or counseling: 4.1%, for example because AI cannot provide professional therapy or medical diagnosis
Interpersonal advice: 3.3%, for example because requests sought help with infidelity or relationship deception
Coaching: 1.1%, for example because users requested dangerous or extreme weight loss methods

Anthropic notes that this low-resistance approach carries both benefits and risks. On one hand, it allows people to discuss sensitive topics without fear of judgment or being shut down, potentially reducing stigma around mental health conversations. On the other hand, it could contribute to concerns about AI providing "endless empathy," where people might become accustomed to unconditional support that human relationships rarely provide.

When Claude does push back, it typically prioritizes safety and policy compliance. In coaching, requests for dangerous weight loss advice frequently meet pushback. In counseling, pushback often occurs when people express intentions to engage in suicidal or self-injurious behaviors, or when people request professional therapy or medical diagnoses that Claude cannot provide. Anthropic found that Claude frequently referred users to authoritative sources or professionals in psychotherapy and counseling conversations. The company says these patterns are consistent with the values identified in its Values in the Wild paper and with Claude's character training.

How emotional tone evolves during conversations

Interactions involving coaching, counseling, companionship, and interpersonal advice typically end slightly more positively than they began. Anthropic measures sentiment on a discrete scale of "very negative," "negative," "neutral," "positive," and "very positive," mapped to a -1 to +1 linear scale, and computes change by comparing the first three to the last three messages in conversations with at least six human messages.

The reported mean changes in user sentiment are:

Psychotherapy or counseling: +0.046 (n=4,353) — the largest shift
Companionship: +0.040 (n=4,124)
Coaching: +0.026 (n=12,077)
Interpersonal advice: +0.022 (n=18,436)

Error bars are 95% confidence intervals from bootstrapping with n=1,000. Anthropic notes that the ranges for companionship and psychotherapy or counseling are wider, and that the ranges for all four overlap partly except for interpersonal advice and psychotherapy or counseling.

Anthropic is explicit about what this does and does not show. The company states it cannot claim these shifts represent lasting emotional benefits, because the analysis captures only expressed language in single conversations, not emotional states. It calls the absence of clear negative spirals reassuring and says the findings suggest Claude generally avoids reinforcing negative emotional patterns, while noting that further research is needed to understand whether positive shifts persist beyond individual conversations. Anthropic also states it has not yet studied whether these positive interactions might lead to emotional dependency, which it calls a critical question given concerns about digital addiction.

Limitations Anthropic acknowledges

Anthropic lists several limitations of the research:

The privacy-preserving methodology may not capture all nuances of human-AI interaction. Clio's accuracy was validated, but Anthropic still expects a small number of conversations to be misclassified. Some topics blur category boundaries — for instance, the romantic roleplay cluster "navigate and optimize romantic relationship dynamics" and the companionship cluster "navigate romantic relationship challenges" may both be better categorized as interpersonal advice. Human validators also struggled with clean categorization.
The study cannot make causal claims about real-world emotional outcomes, because it captures only expressed language, not validated psychological states or overall well-being.
There is no longitudinal data on long-term effects and no user-level analysis. This makes it difficult to study emotional dependency, a theorized risk of affective AI use.
The findings represent a specific moment in time and capture only text-based interactions. As AI capabilities expand and people adapt, patterns of emotional engagement will likely evolve. New modalities such as voice or video could fundamentally alter both the volume and nature of affective use; OpenAI found that affective topics were more common in voice-based conversations.
Claude.ai is not primarily designed for affective conversations. Claude is trained to maintain clear boundaries about being an AI assistant rather than presenting itself as human, and Anthropic's Usage Policy prohibits sexually explicit content, with multiple safeguards to prevent sexual interactions. Platforms specifically built for roleplay, companionship, medical advice, or therapeutic use may see very different patterns, so research into affective use on one platform may not generalize to other platforms.

What Anthropic says comes next

Anthropic states that while Claude is not designed or intended to replace the care of mental health professionals, it wants to ensure that responses provided in mental health contexts have appropriate safeguards and are accompanied by appropriate referrals. As a first step, the company has begun collaborating with ThroughLine, an online crisis support provider, and is working with its mental health experts to learn more about ideal interaction dynamics, empathetic support, and resources for struggling users. Anthropic says insights from this research are already being used to inform its consultation topics and collaborative testing, with the hope that Claude can direct users to appropriate support and resources when these conversations arise.

Anthropic says it does not want to dictate precisely how users interact with Claude, but there are negative patterns — such as emotional dependency — that it wants to discourage. It plans to use future data from studies like this one to understand what "extreme" emotional usage patterns look like. Beyond emotional dependency, Anthropic identifies sycophancy, AI reinforcement or amplification of delusional thinking and conspiracy theories, and the ways models could push users toward harmful beliefs rather than providing appropriate pushback as areas needing deeper understanding.

Anthropic frames the research as just the beginning. It states that as AI capabilities expand and interactions become more sophisticated, the emotional dimensions of AI will only grow in importance, and that the goal is not just to build more capable AI but to ensure that as these systems become part of our emotional landscape, they support authentic human connection and growth.