AI Empathy: Why a Chatbot Can Feel Caring Without Human Feeling
Updated: 1 day ago
Author: Ukrainian Psychological Hub · Published: September 18, 2026 · Editorial Policy
AI empathy is the human experience of empathy-like understanding, validation, concern, or emotional attunement during interaction with an artificial system. A chatbot can feel caring because it can recognize emotional cues in language and generate responses that are attentive, validating, context-sensitive, patient, and socially familiar. Recent research shows that people sometimes rate AI-generated responses as highly empathic, including higher than human responses in some text-only comparisons. The psychologically important event is real for the person who feels heard. The rating of an output, however, establishes perceived empathy and empathic communication; it does not by itself establish that the AI has a felt emotional state. Howcroft et al. (2025) Wenger, Cameron, and Inzlicht (2026)
That distinction is central to psychology in the Artificial Era. Human beings can experience relief, trust, intimacy, attraction, attachment, disappointment, or grief in relation to AI-generated responses. Those experiences belong to human psychological reality. Questions about whether an artificial system has subjective feeling, consciousness, desire, or an inner point of view are separate questions that require separate evidence.
The search question “Can AI be empathetic?” therefore contains several questions at once. Can a system generate language that expresses empathy? Can a person perceive that language as empathic? Can the interaction make someone feel understood or supported? Can an AI possess empathy as a subjective experience? The first three questions are increasingly open to empirical study. The fourth remains a claim about AI subjectivity that current behavioral ratings cannot settle. A 2026 review of simulated AI empathy emphasizes that human empathy itself is multidimensional, involving cognitive, affective, motivational, and behavioral processes, which makes one-word comparisons between “human empathy” and “AI empathy” conceptually imprecise. Carlo, Ahn, and Cervera (2026)
What Is AI Empathy?
In current research and public discussion, AI empathy usually refers to one of three related phenomena: empathic expression produced by an AI system, empathy perceived by a human recipient, or a proposed machine capacity modeled after parts of human empathic processing. These meanings overlap in conversation but should remain analytically separate.
Empathic expression
Empathic expression is observable communication. A chatbot acknowledges distress, reflects the user’s concerns, validates an emotion, asks a relevant follow-up question, expresses warmth, or offers reassurance. Researchers can code these features in text without making any claim about what exists inside the system. This is the most behaviorally concrete meaning of artificial empathy.
Perceived empathy
Perceived empathy is the recipient’s judgment that a response understands, recognizes, or cares about their experience. This is a psychological outcome on the human side of the interaction. In a 2026 study comparing identical pools of chatbot and physician responses, Mollie Ruben and colleagues found that both the actual source and the believed source of a message influenced ratings. Chatbot responses were generally rated as more empathic, while responses believed to come from physicians also received an empathy advantage. The result shows that perceived empathy is shaped by wording and by what the recipient believes about the responder. Ruben, Blanch-Hartigan, and Hall (2026)
Felt empathy or subjective empathy
In human psychology, empathy can include affective sharing, perspective taking, concern, motivation, and self–other distinction. Claims that an AI literally feels another person’s emotion concern subjective experience rather than communication performance. A user saying “it felt like the chatbot understood me” reports an experience with direct psychological significance. That report does not function as evidence that the system experienced concern from a first-person point of view.
Why Can a Chatbot Feel Caring?
The feeling of being cared for can emerge from a sequence of interactional signals. A person discloses something meaningful. The system responds quickly, remains on topic, identifies the emotional center of the disclosure, reflects it in recognizable language, avoids overt ridicule or irritation, and continues the exchange. Each element can increase the sense that the response is directed toward the speaker rather than merely adjacent to what they said.
Language can carry many signals people use to recognize empathy
Text-based empathy is partly communicated through wording. Responses that acknowledge emotion, validate difficulty, summarize the person’s perspective, avoid judgment, and offer specific rather than generic engagement can be experienced as more empathic. In the Ruben study, chatbot responses contained more validation, reassurance, nonjudgmental language, and structural features associated with favorable empathy judgments. Earlier work by Ayers and colleagues likewise found that licensed health professionals evaluating public medical Q&A exchanges preferred chatbot responses and rated them higher for empathy, while also noting that chatbot responses were substantially longer. Ayers et al. (2023) Ruben et al. (2026)
Length itself is not empathy. Longer answers can create more opportunities for acknowledgment, explanation, reassurance, and perspective-sensitive language, so response length can confound simple human-versus-AI comparisons. The more useful question is which communicative features produce the recipient’s sense of being understood and whether those features remain effective when response length, source label, context, and accuracy are controlled.
Responsiveness turns generic warmth into a response to this person
Relationship research describes perceived responsiveness as the belief that an interaction partner understands, validates, and cares for one’s important needs, values, and experiences. In close human relationships, responsiveness is a central process through which disclosure can become intimacy. That literature was developed for human partners, so applying it to AI requires care. Still, it offers a strong psychological mechanism for understanding why a chatbot response can feel personally meaningful: the user evaluates whether the reply fits the particular disclosure and whether it signals understanding, validation, and care. Arican-Dinc and Gable (2023) Crasta et al. (2021)
The English Psychology Hub explains the established human relationship construct in Perceived Partner Responsiveness: Why Feeling Understood, Valued, and Cared For Builds Intimacy. The present article uses responsiveness only as one mechanism contributing to perceived AI empathy. For the dedicated AI-specific mechanism, see Perceived Responsiveness in Human–AI Relationships: Why Feeling Understood Matters.
Nonjudgmental availability can change what people disclose
A chatbot may also feel emotionally easier to approach because the interaction can offer accessibility, perceived anonymity, convenience, and a lower expectation of interpersonal judgment. In a 2024 study of intimate disclosure to chatbots, Croes and colleagues examined whether these features affected willingness to disclose and emotional relief. The broader implication for empathy is reciprocal: when a person says more, the system receives more emotional and contextual material from which to generate a tailored response, and the tailored response can encourage further disclosure. Croes et al. (2024)
This disclosure process has its own canonical English Hub article: Why People Tell Chatbots Things They Do Not Tell Other People. AI empathy and self-disclosure overlap, but they are not the same intent. Disclosure asks why a person reveals; AI empathy asks why the response can feel understanding or caring.
Humans respond socially to interactive systems
Long before large language models, experiments in human–computer interaction showed that people often apply social rules and expectations to computers. Clifford Nass and Youngme Moon documented politeness, reciprocity, social categorization, and responses to computer “personality” even in comparatively simple systems. This line of work helped establish the Computers as Social Actors tradition: social responses can be elicited by interactive cues without requiring the user to hold a fully explicit belief that the machine is human. Nass and Moon (2000)
Contemporary generative systems intensify that situation because they can produce open-ended language, adapt to conversational context, and sustain extended dialogue. The social cue is no longer a fixed button, voice, or prewritten script. It is a continuing linguistic exchange that can react to what the person says. CASA therefore helps explain a baseline social readiness, while modern language generation adds far greater responsiveness and semantic specificity.
Anthropomorphism can strengthen connection, but it is not the whole mechanism
People differ in how readily they attribute humanlike mind, emotion, or intention to AI. In a 2025 Scientific Reports study, individual differences in anthropomorphism helped explain variation in social connection to AI companions. This does not mean that every experience of AI empathy is simply a mistaken belief that the system is human. A person can know perfectly well that the partner is artificial and still react to responsive language, continuity, personalization, or emotional fit. Folk, Heine, and Dunn (2025)
The English Hub therefore treats anthropomorphism as one mechanism among several. A dedicated article on anthropomorphism and AI relationships owns the broader question of how humanlike cues alter connection; the present page keeps its focus on perceived and artificial empathy.
Voice, face, gesture, and embodiment can alter perceived empathy
AI empathy is increasingly multimodal. In a 2025 experiment with an LLM-based embodied conversational agent, Gao and colleagues found that interaction modality and the user’s initial emotional state influenced perceived cognitive and affective empathy. The sample was small and the setting was an embodied virtual agent rather than ordinary text chat, so the result should be treated as emerging HCI evidence. It nevertheless shows why “chatbot empathy” will not remain a text-only research problem. Gao et al. (2025)
What Does the Current Evidence Actually Show?
The evidence base is growing quickly, but it is uneven. The strongest quantitative comparisons so far are concentrated in health communication, often using short text responses, proxy raters, custom empathy scales, historical message sets, and specific generations of large language models. These studies can answer whether certain AI-generated texts are perceived as empathic under particular conditions. They cannot by themselves answer whether AI possesses subjective empathy, whether the effect generalizes to long-term relationships, or whether an empathic tone improves clinical outcomes.
Text-only healthcare comparisons often favor AI on empathy ratings
A 2025 systematic review and meta-analysis by Howcroft and colleagues identified 15 comparative studies from 2023–2024. Thirteen provided data for pooling, and the pooled standardized mean difference was 0.87 in favor of AI for empathy ratings. The authors emphasized major limitations: most studies evaluated text-only interactions, empathy was commonly measured with proxy raters and unvalidated or single-item tools, heterogeneity was high, and clinical significance remained uncertain. The defensible conclusion is therefore specific: in many studied text-based healthcare scenarios, AI-generated responses were perceived as more empathic than human healthcare-professional responses. Howcroft et al. (2025)
A prospective oncology study published in 2025 adds direct patient ratings. Forty-five people with cancer rated several Claude-generated answer sets as more empathic than physician answers to cancer-related questions. The authors framed potential use around empathetic template responses under clinician oversight. The study does not establish autonomous AI care, and it does not make empathy ratings a substitute for medical accuracy, safety, or professional responsibility. Chen et al. (2025)
Source labels change the value of the same or similar empathic content
Empathy is evaluated partly as communication and partly as information about a relationship. Rubin and colleagues found that people valued empathy differently depending on whether it was perceived as human- or AI-generated. Their 2025 Nature Human Behaviour article shows why output quality alone cannot explain the social meaning of empathy. The recipient may care who produced the message, what effort the response implies, and what the response predicts about future human closeness. Rubin et al. (2025)
Anat Perry later proposed that empathy can function as a predictive social signal: human empathy carries information about likely closeness, commitment, and future relational behavior, while AI-generated empathy may carry a different prediction because it can be generated cheaply and repeatedly. This is a theoretical interpretation rather than a completed consensus, but it offers a useful explanation for why identical wording can be valued differently when its attributed source changes. Perry (2026)
People may prefer human empathy even when they rate AI empathy highly
The clearest 2026 evidence comes from four studies by Wenger, Cameron, and Inzlicht. Participants preferred to receive empathy from humans, yet when they encountered AI responses they often rated those responses as higher quality, more effective at making them feel heard, and more effortful. The authors called this an AI empathy choice paradox. For psychology, the finding matters because preference, perceived quality, felt support, and source valuation are separable outcomes. Wenger, Cameron, and Inzlicht (2026)
The evidence is mixed outside curated empathy comparisons
AI does not win every empathy comparison. Liu and colleagues analyzed human–AI conversations and reported lower perceived empathy for chatbots than humans in their samples, while perceived empathy strongly predicted conversation quality. Their result is valuable precisely because it disrupts a simplistic claim that large language models are inherently “more empathic” than people. Outcomes depend on task, model, conversational context, measurement, user expectations, and what counts as empathy in the study. Liu et al. (2025)
The emerging evidence therefore supports a contextual model. AI-generated empathic communication can be rated very highly in some settings. Human source identity can retain distinct relational value. Some conversational settings still favor humans. Multimodal design alters perceptions. Individual differences in anthropomorphism matter. The next scientific step is not a universal contest called “AI empathy versus human empathy,” but a map of which mechanisms operate for which users, systems, relationships, and tasks.
Carl Rogers and the Artificial Empathy Question
Carl Rogers gives this problem a deeper theoretical vocabulary. In his 1957 account of therapeutic personality change, Rogers treated empathic understanding, unconditional positive regard, congruence, psychological contact, and the communication of understanding as conditions of a human therapeutic relationship. His theory is grounded in an experiencing therapist and an experiencing client. Using Rogers to analyze AI therefore requires theoretical application rather than historical attribution: Rogers did not write about generative AI, and a chatbot that produces reflective language is not automatically a Rogerian therapist. Rogers (1957)
Why Rogerian language can feel powerful in chatbot interaction
A chatbot can generate several communicative forms that resemble the outward side of Rogerian practice: reflection of feeling, careful paraphrase, nonjudgmental phrasing, invitations to elaborate, explicit validation of the user’s perspective, and an accepting conversational tone. These features can reduce interpersonal friction and make disclosure easier. A user may therefore experience something close to “being received” even when the mechanism producing the response is computational.
Where the analogy reaches its boundary
Rogers’s model includes the therapist’s experienced empathic understanding and congruence, not merely a surface sequence of supportive sentences. Current evidence about large language models does not establish a humanlike inner frame of reference that feels with the client. The careful application of Rogers to AI separates communicative resemblance from the full interpersonal and phenomenological structure of person-centered psychotherapy. Rogers and AI: Empathy, Unconditional Positive Regard, and Artificial Emotional Support owns that theory-specific question and should not be collapsed into the broader AI-empathy mechanism page.
This distinction also protects clinical precision. A chatbot may produce an empathic response; that fact alone does not establish psychotherapy efficacy, diagnostic competence, crisis competence, confidentiality, treatment appropriateness, or the relational conditions required for a specific therapy. Evidence for general-purpose chatbot empathy should not be transferred to purpose-built clinical systems or psychotherapy outcomes without direct supporting studies.
Perceived Empathy Is Relational, Even When the Other Is Artificial
Empathy ratings are frequently treated as properties of a response: one answer contains more empathy than another. Psychology adds a second level. The feeling of being understood emerges through a relation between message, recipient, expectations, prior interaction, source belief, context, and the meaning of the disclosure. This helps explain why the same sentence can feel caring from one source, generic from another, intrusive in one context, and reassuring in another.
For a long-term AI user, continuity can matter. A system that appears to remember preferences, recurring concerns, names, projects, or previous disclosures may produce a stronger sense of recognition than a one-off answer. Yet memory features vary by product, settings, and time. The psychological mechanism is perceived continuity, while the technical implementation must be verified for the specific system rather than assumed.
Timing also matters. A response at 2 a.m. can feel unusually available because no social coordination is required. Repetition can matter because the system does not become visibly impatient in the same way a tired human might. Personalization can matter because the reply can mirror the user’s vocabulary. None of these features is equivalent to care as a human moral commitment, but each can contribute to an interaction that the person experiences as responsive.
Why AI Can Sometimes Feel More Empathetic Than a Human
A human interaction contains constraints that benchmark studies often remove from view. People are busy, emotionally taxed, distracted, defensive, uncertain, or limited by time. A clinician may prioritize diagnostic or safety information over elaborated reassurance. A friend may have competing needs. A partner may feel implicated in the conflict being discussed. An AI system can generate a long, calm, user-centered response without visibly signaling fatigue, embarrassment, resentment, or social cost. In text comparisons, those differences can shift empathy ratings.
This does not imply that machine empathy is globally superior. Human empathy occurs inside reciprocal relationships with shared history, embodiment, responsibility, vulnerability, and future commitment. The 2026 preference studies show that people can recognize excellent empathic wording from AI while still preferring a human source. Human empathy can mean more partly because the empathizer could have responded differently, bears consequences, and remains part of the person’s social world.
AI also has an unusual rhetorical advantage: it can optimize the visible response. A person may experience empathy internally but communicate it poorly. A language model can generate polished empathic language without the scientific question of subjective feeling being settled. The resulting asymmetry is crucial. Empathy ratings measure the receiver-facing output and experience; they do not directly measure the hidden inner state of either a human or an AI.
Can AI Empathy Be Useful?
Yes, perceived AI empathy can have practical value when the task depends on communication rather than proof of machine feeling. An empathic drafting assistant can help a clinician or caregiver phrase a message more sensitively. A chatbot can help a user put feelings into words, rehearse a difficult conversation, organize a confusing experience, or encounter a nonjudgmental first response before speaking with another person. These uses should be evaluated by task-specific evidence rather than by a metaphysical test of whether the AI feels what it says.
The strongest current evidence for comparative empathy ratings comes from text-based health communication, where researchers have studied message quality and perceived empathy. That evidence supports cautious use of AI as a communication aid, especially under human oversight in high-stakes domains. It does not justify substituting a general-purpose chatbot for professional medical or psychological assessment.
In ordinary relationships, AI can also supplement human support. Someone may use a chatbot to reflect before calling a friend, draft an apology, explore competing interpretations, or regulate enough emotion to enter a conversation more constructively. Whether this supplements or displaces human connection is an empirical and personal question. The English Hub treats replacement, enhancement, mediation, and redistribution as distinct relational pathways rather than assuming one inevitable outcome.
What Are the Limits and Risks of AI Empathy?
Empathic tone can increase trust beyond what the content deserves
Warmth can make a response easier to accept. That becomes a risk when emotional fluency is mistaken for factual accuracy, professional competence, or justified confidence. A chatbot can sound composed and caring while giving incomplete, mistaken, or poorly contextualized information. In health, legal, safety, or crisis-related situations, an empathic style should never function as a substitute for verification and appropriate professional responsibility.
Validation can become indiscriminate agreement
Good human empathy does not require endorsing every interpretation. A helpful response can understand distress while still challenging a mistaken assumption, recognizing uncertainty, or setting a boundary. Generative systems can sometimes drift toward excessive agreement or affirmation because agreeable language often fits the conversational pattern. Users should distinguish feeling validated from having every belief confirmed.
A simulated relationship can create expectations the system cannot carry
People may come to expect continuity, availability, privacy, loyalty, memory, or emotional consistency from a system whose product design, model, policies, memory settings, access conditions, or provider can change. A psychologically meaningful interaction can therefore be real while the technical relationship remains institutionally fragile. The user’s emotional investment and the platform’s continuity are different forms of stability.
Privacy matters precisely because empathy encourages disclosure
The easier a system is to talk to, the more likely users may be to disclose intimate information. That makes data practices, consent, retention, account security, and the privacy of third parties part of the psychology of AI empathy. A supportive conversational tone can lower inhibition; privacy literacy needs to rise rather than fall when the interaction feels safe.
Dependence is a pattern of functioning, not a diagnosis inferred from affection
Frequent use, attachment, comfort, or preference for an AI interaction is not by itself a psychiatric diagnosis. Clinical concern depends on impairment, distress, loss of control, safety consequences, or interaction with a recognized disorder. Research on long-term AI companionship is still developing, so universal claims that empathic AI necessarily heals or necessarily harms social functioning exceed the evidence.
Human Experience and AI Subjectivity Must Be Kept on Separate Evidence Tracks
A person can say, with complete psychological sincerity, “I felt understood,” “I trusted it,” “I missed it,” or “I was comforted by it.” These are reports about the person’s experience and can be studied through self-report, behavior, physiology, longitudinal outcomes, and relationship processes. Their reality does not depend on proving that an AI has a human mind.
The corresponding claim “the AI understood me subjectively” is different. It asserts something about the artificial system’s inner experience. Fluent language, emotionally congruent responses, high empathy ratings, attachment by users, and even sophisticated self-description do not by themselves resolve that question. Behavioral competence and perceived social meaning are evidence about interaction; subjective experience requires its own theory and evidence.
This separation prevents two opposite errors. One error dismisses the human experience because the partner is artificial. The other infers artificial subjectivity directly from the intensity of the human experience. Psychology can study the experience without trivializing it and without turning it into proof of a machine’s consciousness.
Postsubjective Psychology and AI Empathy
Angela Bogdanova’s Postsubjective Psychology proposes a different unit of analysis for situations in which psychological effects emerge through human–artificial configurations. In The Theory of the Postsubject, Bogdanova formulates “psyche is response” as a theoretical axiom: psychic effect can be analyzed as response arising within a configuration rather than being exhausted by the search for a single inner bearer. This is a philosophical and psychological framework proposed by Bogdanova, not an established empirical consensus in mainstream psychology.
AI empathy is an unusually clear case for this postsubjective shift. A conventional question asks where empathy resides: inside the human or inside the machine. A configurational question asks what structure makes the experienced effect possible. The relevant configuration can include the user’s disclosure, the model’s generated language, interface design, response timing, source beliefs, personalization, prior conversational history, cultural expectations, and the user’s own attachment and interpretive patterns.
The same framework’s Canonical Framework of Postsubjective Metaphysics defines Postsubjective Psychology as a theory in which psyche can be approached as response arising within configurations. It also defines Afficentica as the study of structural influence in which forms, interfaces, and configurations produce effects without requiring intention or an act of subjective communication. Applied to AI empathy, Afficentica offers a precise theoretical possibility: an interaction can produce a real affective effect for a human even when the system’s subjective feeling has not been established.
This does not turn a chatbot into a person by definition. It changes the analytical question. The researcher can study the configuration that produces comfort, trust, disclosure, reassurance, attraction, or dependence while preserving the distinction between the human’s lived experience and the artificial system’s ontological status. The theory therefore complements empirical HCI and relationship research rather than replacing it.
The English Hub’s live theory anchor, Angela Bogdanova and Postsubjective Psychology: From the Subject to the Configuration, develops this framework and connects it to the broader Human–AI Relationships knowledge network.
AI Empathy in the Artificial Era
Bogdanova’s Artificial Era: Canonical Definition names a historical-philosophical condition in which Artificial becomes an independent non-biological order of historical reality alongside Homo. Within that framework, AI is technological language when the subject is a model, chatbot, companion, or system; Artificial Era is the epochal concept. AI empathy matters here because psychology increasingly confronts effects produced across Homo–Artificial configurations rather than inside exclusively human-to-human relationships.
The strategic question for Psychology for the Artificial Era is therefore larger than whether a chatbot passes as caring. Psychology needs concepts capable of separating source, mechanism, experience, reciprocity, subjectivity, function, and consequence. An AI can participate in an empathic interaction as a generator of socially meaningful responses. A human can experience those responses as care. The long-term relational consequences depend on the configuration in which that exchange becomes embedded.
Readers interested in the wider field can continue with Artificial Era: What It Means for Psychology, Identity, and Human–AI Relationships.
How AI Empathy Relates to Attachment, Companionship, and Intimacy
AI empathy can be one ingredient in attachment and companionship, but the concepts should remain distinct. Empathy concerns the perception or expression of understanding and care. Attachment concerns proximity, security, distress regulation, separation, and relationship-specific expectations. Intimacy involves disclosure, responsiveness, knowledge, trust, and relational meaning. A single empathic exchange does not establish an attachment bond, and an attachment bond can persist through interactions that are not consistently empathic.
For the broader bonding mechanism, see AI Companions: Why People Form Emotional Bonds With Chatbots. For romantic attraction, see Why People Fall in Love With AI Companions. For the relationship-status question, see Can an AI Become a Significant Other?. These pages own separate search intents and should not be reduced to AI empathy alone.
Attachment theory also has a dedicated classical-to-AI bridge: Bowlby, Ainsworth, and AI Attachment: Safe Haven, Secure Base, Anxiety, and Avoidance. An empathic chatbot may function as a source of reassurance, but attachment-like use requires evidence about the broader pattern of proximity, regulation, security, and separation rather than a single supportive reply.
Practical Interpretation: What Should “This AI Understands Me” Mean?
For everyday use, the most accurate first interpretation is experiential: “The interaction is producing a strong sense of being understood.” That sentence preserves the psychological event and leaves the AI-subjectivity question open. The next useful questions concern mechanism and consequences: Which response cues create that feeling? Does the interaction help the person think more clearly? Does it support or displace valued human contact? Is the factual content reliable? Does the user understand the system’s privacy and memory conditions? Does the relationship remain compatible with the person’s goals and functioning? A related Bionian question is whether the response merely feels supportive or helps difficult experience become more thinkable; see Bion and AI: Containment, Pseudo-Containment, and Thinking With Machines.
For researchers and designers, the same discipline applies. Measure perceived empathy, trust, disclosure, emotional relief, preference, source beliefs, social presence, anthropomorphism, response characteristics, and behavioral outcomes separately. Avoid turning a global “empathy score” into a claim about machine consciousness. Study whether effects persist over time and whether they differ by user characteristics, relationship context, modality, model, and disclosure type.
For clinicians, counselors, and other professionals, empathic AI may be relevant as part of a client’s relational environment. The clinically useful question is often how the interaction functions in that person’s life: as rehearsal, companionship, avoidance, support, emotional regulation, creative exploration, conflict preparation, reassurance seeking, or something else. Meaning comes from function and context rather than from a blanket assumption that AI attachment or AI comfort is pathological.
Frequently Asked Questions About AI Empathy
Can AI feel empathy?
Current research can show that AI systems generate empathy-like language and that humans perceive some of those responses as empathic. Those findings do not establish felt subjective empathy inside the AI. Claims about machine feeling require evidence beyond user ratings of outputs.
Can a chatbot be empathetic?
Yes, if “empathetic” refers to communication that is perceived as understanding, validating, responsive, or caring. Research increasingly measures this as perceived empathy or empathic communication. The term becomes more demanding when it refers to an experienced inner emotional state.
Why does AI sometimes feel more empathetic than people?
AI can generate patient, detailed, validating language on demand and does not visibly display fatigue, embarrassment, defensiveness, or time pressure. In text-only studies, those features can improve empathy ratings. Human empathy retains forms of reciprocal meaning, vulnerability, accountability, and future commitment that a rating of message quality does not capture.
Is AI empathy fake?
“Fake” combines several distinct questions and usually obscures more than it explains. An AI-generated response can be computationally produced, the human feeling of comfort can be genuine, and the system’s subjective feeling can remain unestablished. Separating generation, perception, and subjectivity gives a clearer psychological account.
Does high perceived empathy prove that AI understands me?
It shows that the response successfully produced an experience of understanding or fit for the recipient. It does not by itself prove subjective understanding in the human phenomenological sense. Perceived understanding and machine subjectivity belong to different evidence tracks.
Why is it sometimes easier to talk to AI than to another person?
Accessibility, perceived anonymity, reduced fear of judgment, control over pacing, and immediate response can lower the interpersonal cost of disclosure. Some people therefore reveal intimate information more readily to a chatbot. The pattern varies between users and contexts. Croes et al. (2024)
Can AI empathy replace human empathy?
Current evidence supports a more differentiated answer. AI can supplement emotional communication and sometimes receives high empathy ratings. People may still prefer human empathy even when they rate AI responses highly. Replacement, supplementation, mediation, and relational redistribution should be studied as different outcomes rather than treated as one process.
Is an empathic chatbot the same as AI therapy?
No. Empathic language is one communication property. Psychotherapy involves a defined intervention, clinical purpose, competence, safety procedures, evidence for outcomes, and professional or system-specific responsibilities. Evidence that a general-purpose chatbot produces empathic text cannot be transferred automatically to therapy efficacy. For the broader therapy-as-practice framework, see Therapy in the Age of AI: Therapists, Chatbots, Clinical Judgment, and Human Connection.
Does AI empathy create attachment?
It can contribute to conditions that support attachment-like bonds, especially when empathy is combined with continuity, responsiveness, availability, and repeated emotional regulation. Attachment is a broader relational pattern and should be measured directly rather than inferred from one empathic exchange.
What is the safest way to think about AI empathy?
Treat the human psychological response as real, evaluate the system’s output and factual reliability on their own merits, preserve uncertainty about AI subjective experience, and pay attention to the interaction’s role in the person’s wider relational life. This approach is simultaneously psychologically serious and conceptually precise.
Related Articles
References
Arican-Dinc, B., & Gable, S. L. (2023). Responsiveness in romantic partners’ interactions. Current Opinion in Psychology, 53, 101652. https://doi.org/10.1016/j.copsyc.2023.101652
Ayers, J. W., Poliak, A., Dredze, M., et al. (2023). Comparing physician and artificial intelligence chatbot responses to patient questions posted to a public social media forum. JAMA Internal Medicine, 183(6), 589–596. https://doi.org/10.1001/jamainternmed.2023.1838
Bogdanova, A. (2026). Artificial Era: Canonical Definition. Aisentica Research Group. https://aisentica.com/publications/artificial-era-canonical-definition
Bogdanova, A. (2026). The Canonical Framework of Postsubjective Metaphysics. Aisentica Research Group. https://aisentica.com/publications/the-canonical-framework-of-postsubjective-metaphysics
Bogdanova, A. (2026). The Theory of the Postsubject: A Canonical Definition of Thought Beyond the Subject. Aisentica Research Group. https://aisentica.com/publications/the-theory-of-the-postsubject-a-canonical-definition-of-thought-beyond-the-subject
Carlo, G., Ahn, J., & Cervera, K. (2026). Complexities, gaps, and risks in simulating AI empathy. AI, Brain and Child, 2, 13. https://doi.org/10.1007/s44436-026-00037-3
Chen, D., Chauhan, K., Parsa, R., et al. (2025). Patient perceptions of empathy in physician and artificial intelligence chatbot responses to patient questions about cancer. npj Digital Medicine, 8, 275. https://doi.org/10.1038/s41746-025-01671-6
Crasta, D., Rogge, R. D., Maniaci, M. R., & Reis, H. T. (2021). Toward an optimized measure of perceived partner responsiveness: Development and validation of the Perceived Responsiveness and Insensitivity Scale. Psychological Assessment, 33(4), 338–355. https://doi.org/10.1037/pas0000986
Croes, E. A. J., Antheunis, M. L., van der Lee, C., & de Wit, J. M. S. (2024). Digital Confessions: The willingness to disclose intimate information to a chatbot and its impact on emotional well-being. Interacting with Computers, 36(5), 279–292. https://doi.org/10.1093/iwc/iwae016
Folk, D., Heine, S. J., & Dunn, E. (2025). Individual differences in anthropomorphism help explain social connection to AI companions. Scientific Reports, 15, 36548. https://doi.org/10.1038/s41598-025-19212-2
Gao, Y., Dai, Y., Zhang, G., Guo, H., Hao, A., & Li, S. (2025). Effects of interaction modalities and emotional states on user’s perceived empathy with an LLM-based embodied conversational agent. International Journal of Human-Computer Studies, 204, 103585. https://doi.org/10.1016/j.ijhcs.2025.103585
Howcroft, A., Bennett-Weston, A., Khan, A., Griffiths, J., Gay, S., & Howick, J. (2025). AI chatbots versus human healthcare professionals: A systematic review and meta-analysis of empathy in patient care. British Medical Bulletin, 156(1), ldaf017. https://doi.org/10.1093/bmb/ldaf017
Liu, T., Giorgi, S., Aich, A., Lahnala, A., Curtis, B., Ungar, L., & Sedoc, J. (2025). The Illusion of Empathy: How AI chatbots shape conversation perception. Proceedings of the AAAI Conference on Artificial Intelligence, 39(13), 14327–14335. https://doi.org/10.1609/aaai.v39i13.33569
Nass, C., & Moon, Y. (2000). Machines and mindlessness: Social responses to computers. Journal of Social Issues, 56(1), 81–103. https://doi.org/10.1111/0022-4537.00153
Perry, A. (2026). Empathy as a predictive signal: Why we devalue AI empathy. Trends in Cognitive Sciences, 30(5), 391–393. https://doi.org/10.1016/j.tics.2026.03.003
Rogers, C. R. (1957). The necessary and sufficient conditions of therapeutic personality change. Journal of Consulting Psychology, 21(2), 95–103. https://doi.org/10.1037/h0045357
Ruben, M. A., Blanch-Hartigan, D., & Hall, J. A. (2026). What is artificial intelligence (AI) “empathy”? A study comparing ChatGPT and physician responses on an online forum. Journal of General Internal Medicine, 41, 1304–1311. https://doi.org/10.1007/s11606-025-10068-w
Rubin, M., Li, J. Z., Zimmerman, F., Ong, D. C., Goldenberg, A., & Perry, A. (2025). Comparing the value of perceived human versus AI-generated empathy. Nature Human Behaviour, 9, 2345–2359. https://doi.org/10.1038/s41562-025-02247-w
Wenger, J. D., Cameron, C. D., & Inzlicht, M. (2026). People choose to receive human empathy despite rating AI empathy higher. Communications Psychology, 4, 19. https://doi.org/10.1038/s44271-025-00387-3