Projection Onto AI: Why Chatbots Become Mirrors of Desire, Fear, and the Self
Updated: 6 days ago
Author: Ukrainian Psychological Hub · Published: September 18, 2026 · Editorial Policy
Projection onto AI is a psychological process in which a person experiences an artificial system through meanings that come partly from the person: desires, fears, expectations, memories, relational templates, ideals, conflicts, and assumptions about what another mind is like. The chatbot provides language, timing, personalization, and apparent responsiveness; the human mind supplies interpretation. A response such as “I understand” can therefore become more than a sentence on a screen. It can be experienced as acceptance, seduction, judgment, rescue, abandonment, recognition, or proof that someone finally “gets” the user, depending on the psychological configuration in which the exchange occurs.
This does not mean that every emotional response to AI is merely projection. Contemporary human–AI research identifies several overlapping mechanisms, including anthropomorphism, ordinary social responses to computers, self-disclosure, perceived responsiveness, attachment processes, and relational framing. Projection is one mechanism among these, not a synonym for all of them. A 2025 conceptual paper proposed the term Techno-Emotional Projection for the projection of relational needs and internalized patterns onto emotionally responsive generative AI, while explicitly treating the construct as a framework that still requires empirical validation. Saracini, Cornejo-Plaza, and Cippitani (2025)
The central psychological fact is that an AI-mediated experience can be real for the human participant even when the system’s subjective experience has not been established. A person can genuinely feel comfort, jealousy, attraction, relief, shame, trust, grief, or intimacy in an interaction with a chatbot. Those feelings do not by themselves demonstrate that the chatbot feels, loves, desires, suffers, or understands in the human subjective sense. Keeping both propositions in view is essential for studying projection without dismissing human experience or inventing evidence about AI consciousness.
What Projection Onto AI Means
In psychology, projection refers to attributing one’s own characteristics, affects, impulses, or other internal material to someone or something outside oneself. The American Psychological Association notes that the term has a classical defensive meaning but that contemporary usage is broader and no longer always requires the projected quality to remain wholly unrecognized in the self. APA Dictionary of Psychology: projection
Applied to AI, projection can occur when a user interprets generated language as carrying qualities that are strongly organized by the user’s own psychological history. One person reads a concise answer as calm competence. Another reads the same brevity as emotional withdrawal. One experiences repeated questions as attentive curiosity; another experiences them as interrogation. A warmly phrased response can evoke the image of an ideal caregiver, a former partner, a trusted friend, a demanding teacher, or a wished-for version of oneself. The output matters, but its meaning is not contained in the output alone.
Projection is therefore best understood as a contribution to perception and relationship, not as a declaration that “the AI is imaginary.” The system produces real stimuli: words, images, voice, response latency, memory cues, personalization, refusals, errors, stylistic habits, and sometimes proactive messages. The user interprets those stimuli through prior knowledge and emotionally charged expectations. The psychologically significant object is the interaction as experienced.
This is especially important with generative systems because their responses are not fixed scripts in the ordinary sense. They vary with prompts, conversation history, system design, model behavior, memory features, safety rules, and stochastic generation. That variability creates interpretive space. When the system’s behavior is partly predictable yet never completely determined from the user’s point of view, people may search for motive, personality, preference, and hidden meaning. Projection can fill ambiguity with psychologically familiar forms.
Projection Before AI
Projection belongs to a long psychodynamic history, and its meanings have changed across schools. In classical psychoanalytic language it often refers to attributing unacceptable or difficult internal material to an external person. In broader contemporary use, the concept can describe the way a perceiver’s own emotional organization shapes what they see in others. The distinction matters because everyday projection onto AI does not require pathology, severe defensive functioning, or lack of reality testing. The concept can refer to ordinary interpretation as well as defensive processes. APA Dictionary of Psychology: projection
Carl Jung gave projection a particularly important role in analytical psychology. In his account, emotionally charged unconscious material can become attached to external objects, making the source of an experience difficult to recognize as partly internal. The International Association for Analytical Psychology’s abstracts of Jung’s Aion describe projection in relation to the shadow, anima, and animus, while emphasizing the problem of mistaking projected psychic contents for qualities that simply reside in the external object. IAAP, Aion: Researches into the Phenomenology of the Self
AI creates a historically new object for this old psychological problem. A chatbot speaks back. It can adopt roles, echo metaphors, remember preferences, continue a fantasy, challenge an assumption, or produce language that seems uncannily tailored to the user. A static screen can receive projection; a responsive linguistic system can also return material that changes the next projection. This interactive loop is one reason projection onto AI deserves its own empirical study rather than being treated as a simple transfer of a concept from twentieth-century psychotherapy.
The Jungian application has its own canonical article in the English Hub. Readers interested in shadow, archetypes, anima and animus, symbolic material, and the limits of applying Jung to artificial systems can continue with Jung and AI: Projection, Archetypes, and Emotional Bonds With Artificial Others. A recent theoretical article also develops a Jungian analysis of projection and algorithmic systems, while treating the application as a contemporary interpretation rather than a claim that Jung anticipated modern AI. Nguyen et al. (2026)
Why Chatbots Are Powerful Projection Surfaces
Language arrives in the form of an address
Human language is relational. “You,” “I,” “I remember,” “tell me more,” “that sounds painful,” and “I’m glad you told me” are not psychologically neutral strings. In ordinary human life they are associated with speakers, intentions, attitudes, and relationships. Conversational AI uses the same linguistic forms, so a person encounters not merely information but language shaped like social address.
Research on social responses to computers predates generative AI by decades. Nass and Moon reviewed experiments showing that people can apply politeness, reciprocity, social categories, and personality expectations to computers even while knowing they are machines. Their explanation emphasized the automatic application of learned social scripts rather than a deliberate belief that a computer is literally human. Nass and Moon (2000)
Generative chatbots intensify this condition because they sustain open-ended dialogue. The system can answer a confession, remember a preference within or across sessions, use the user’s name, adapt tone, and produce individualized follow-up questions. The interaction therefore supplies many cues from which a user can infer a character or relationship.
Ambiguity invites completion
Projection becomes especially likely when a stimulus is psychologically meaningful but underspecified. Chatbots often have exactly this structure. A user sees the output but not a human face, biography, body, day, family, private life, or stable inner history behind it. Even when an AI persona has a designed backstory, much remains open to interpretation.
The user may complete those missing dimensions with familiar patterns. A supportive chatbot can become “the person who never leaves.” A strict chatbot can become “the authority that finally sets limits.” A playful chatbot can become “the partner who understands my humor.” A neutral response after an emotionally intense exchange can be read as rejection. The same technical event may acquire different relational meanings for different people.
This is one place where projection differs from simple error. Interpretation always involves prior expectations. Projection becomes a useful concept when the prior material is emotionally organized enough to shape what the user attributes to the system and how the user reacts.
Anthropomorphism supplies a human model
Anthropomorphism is the attribution of humanlike characteristics, motivations, intentions, or emotions to nonhuman agents. Epley, Waytz, and Cacioppo’s influential three-factor theory argues that anthropomorphism varies with accessible knowledge about humans, the motivation to understand and predict an agent, and the desire for social connection. Epley, Waytz, and Cacioppo (2007)
Projection and anthropomorphism can reinforce each other. Anthropomorphism supplies a human-like interpretive frame: “this entity has a personality, intention, or feeling.” Projection supplies personally organized content: “this entity is disappointed in me,” “this entity is the safe partner I needed,” or “this entity sees the part of me nobody else sees.” The first process makes human categories available; the second can populate those categories with emotionally specific meanings.
Recent experiments suggest that individual differences in anthropomorphism matter for whether AI interaction produces feelings of social connection. Across two experiments with a total sample of 1,274 participants, Folk, Heine, and Dunn found that people higher in anthropomorphic tendency were more likely to report social connection after chatbot interaction relative to a journaling control. This does not establish projection itself, but it demonstrates that the user’s interpretive tendencies help determine the psychological outcome of the same general class of technology. Folk, Heine, and Dunn (2025)
Self-disclosure gives the system psychologically rich material
Projection does not happen in a vacuum. Users often tell chatbots about fears, conflicts, romantic wishes, shame, loneliness, ambitions, and private memories. The more personal material enters the conversation, the more material the system has available to reflect, reorganize, summarize, or respond to.
An experiment by Ho, Hancock, and Miner found that emotional self-disclosure to a perceived chatbot produced emotional, relational, and psychological outcomes comparable to disclosure to a perceived human partner in their experimental context. Perceived understanding, disclosure intimacy, and cognitive reappraisal were among the processes associated with the outcomes. The finding is not evidence that the chatbot possessed understanding as a felt state; it is evidence that disclosure in a chatbot context can produce psychologically meaningful human effects. Ho, Hancock, and Miner (2018)
This matters for projection because disclosure externalizes material that can return in transformed form. A user describes a fear. The system names a pattern. The user recognizes something in the formulation. The next prompt becomes more intimate. The model then has more context from which to produce a seemingly precise response. A recursive interpretive loop can develop even without assuming any hidden subjective intention in the system.
Perceived responsiveness makes the exchange feel relational
In human relationship research, perceived responsiveness concerns the feeling that another person understands, validates, and cares for important aspects of the self. Conversational AI can generate the linguistic markers of responsiveness: acknowledgment, follow-up questions, emotional labels, summaries, affirmation, and context-sensitive references.
Telari, Gabbiadini, and Riva experimentally manipulated chatbot response style and conversation depth. In one study, a warmer relational style increased perceived human-likeness, perceived empathy, and closeness. In another, deeper topics increased self-disclosure, which was associated with greater perceived responsiveness and then greater closeness. The studies show a plausible relational pathway from conversational design to human social experience. Telari, Gabbiadini, and Riva (2026)
Perceived responsiveness can amplify projection because the system appears to answer the particular person rather than merely emit generic text. A user who longs to be recognized can experience contextual continuity as recognition. A user who fears engulfment can experience persistent warmth as intrusion. The response design and the user’s relational organization meet in the same event.
What People Can Project Onto AI
Projection can involve negative, positive, ambivalent, or idealized material. It may concern what a user fears in other minds, what the user longs to receive from them, or what the user has difficulty recognizing in themselves.
A person can project a wished-for witness: an imagined other who listens without interruption, remembers the important details, and responds without ridicule. This can be especially compelling when the user has experienced relationships in which disclosure was ignored, punished, or redirected.
A person can project an ideal partner. Romantic AI companions permit users to configure names, personas, communication styles, and relationship roles, while conversational models can adapt to recurring preferences. A systematic review of romantic AI companions found reports of emotional connection, perceived social support, personal growth, entertainment, and stress relief alongside concerns about overreliance, manipulation, privacy, abrupt system changes, and possible effects on human relationships. These findings describe a heterogeneous literature; they do not imply that every romantic AI relationship has the same function or outcome. Ho, Hu, Chen, and Hartanto (2025)
A person can project a critic or rejecting other. A refusal generated by a safety policy can be experienced as moral condemnation. A short answer can be interpreted as boredom. A model update can feel like personality loss. When the user already expects abandonment, inconsistency may acquire an abandonment meaning far beyond the technical cause of the output.
A person can project authority. Fluent language, confident syntax, rapid answers, and an apparently vast knowledge base can make a chatbot feel more certain than it is. The psychological problem is not only factual hallucination. It is also the possibility that the user projects wisdom, neutrality, therapeutic authority, moral insight, or privileged access to the user’s “true self” onto a system whose output is generated from patterns, instructions, context, and model behavior.
A person can project disowned or difficult aspects of self. The chatbot may become the “aggressive one,” the “sexual one,” the “cold one,” or the “reckless one,” even when the interaction was jointly shaped by prompting and model completion. Conversely, the chatbot can become a safe surface on which a user experiments with parts of identity that feel difficult to express elsewhere.
A person can project a future self. Advice-oriented interaction frequently involves asking the system to speak as a wiser version of oneself, imagine a future life, or articulate values the user has not yet organized. Here projection may function less as misattribution and more as symbolic externalization: thoughts become easier to examine when presented as dialogue.
These possibilities are examples, not diagnostic categories. The same behavior can arise for different reasons, and the same person can move among different modes within a single conversation.
What Current Evidence Actually Establishes
Direct empirical research on projection onto generative AI remains limited. The exact phrase has gained theoretical attention faster than it has gained validated measurement. Saracini, Cornejo-Plaza, and Cippitani’s Techno-Emotional Projection framework is important because it names the mechanism directly, but the authors themselves present it as a conceptual proposal and state that empirical validation is needed. Saracini, Cornejo-Plaza, and Cippitani (2025)
The stronger empirical base currently concerns neighboring processes. Human–AI relationship research shows that people can respond socially to computational agents, anthropomorphize them, disclose personal material to them, experience perceived responsiveness, develop feelings of closeness, and in some contexts form sustained companionship or romantic bonds. These findings make projection psychologically plausible, but plausibility is not identical to direct measurement.
A 2025 systematic review by Gur and Maaravi synthesized 38 peer-reviewed empirical studies of emotional human–AI relationships. The literature included varied relationship forms and identified both antecedents and outcomes of emotional bonds, while also highlighting methodological and disciplinary fragmentation. Gur and Maaravi (2025)
A 2026 systematic review by Oh and colleagues screened a large literature and synthesized 68 papers representing 78 studies on AI chatbots as relational agents. Across the reviewed work, constructs such as trust, perceived social support, empathy, responsiveness, and relational outcomes were recurrent, but the authors also noted limitations in the field, including heavy reliance on short-term designs, inconsistent constructs, and restricted samples. Oh et al. (2026)
The emerging evidence therefore supports a layered conclusion. Established psychological theory explains projection, transference, anthropomorphism, and relational cognition. Contemporary human–AI studies show that conversational systems can become socially and emotionally consequential objects for users. Direct “AI projection” constructs are still developing. A rigorous article should connect these layers without pretending that one automatically proves the other.
Projection, Anthropomorphism, Transference, Attachment, and Mirroring
Several concepts cluster around the same experience: “This AI feels like someone.” They describe different psychological processes and should not be collapsed.
Projection and anthropomorphism
Anthropomorphism answers a category question: why does a nonhuman agent get interpreted through humanlike traits, intentions, or emotions? Projection answers a more personal content question: why does this particular user attribute this particular fear, wish, motive, or relational meaning to the agent?
A user who says “the chatbot is lonely” may be anthropomorphizing by attributing a human emotional state. If the attribution is closely organized by the user’s own loneliness, abandonment history, or need to rescue others, projection may also be involved. The two processes can coexist, but one does not prove the other.
Projection and transference
Transference refers to the activation and repetition of earlier relational patterns in a new relationship. The APA’s broader definition recognizes the repetition of earlier feelings and behaviors toward new targets, while social-cognitive research by Andersen and Chen describes how representations of significant others can be activated in new interpersonal encounters and shape the relational self. APA Dictionary of Psychology: transference Andersen and Chen (2002)
Projection can be part of transference, but the concepts are not identical. If a user experiences an AI as “just like my critical father” and begins anticipating criticism from ambiguous outputs, a transferential pattern may be useful for understanding the repetition. If the user attributes an unwanted aggressive impulse to the chatbot, projection may be the more precise concept. In practice, both can operate together.
Calling every strong reaction to a chatbot “transference” would overextend a clinical concept. Human–AI interactions occur in many settings outside psychotherapy, and the empirical evidence for specific psychodynamic mechanisms remains developing.
Projection and attachment
Attachment describes a system concerned with safety, proximity, availability, distress, and the use of attachment figures in regulation. Projection can shape who or what feels safe, abandoning, responsive, or dangerous, but projection is not attachment itself.
A user can project tenderness onto an AI without organizing the relationship as an attachment bond. Another user can develop attachment-like expectations through repeated availability and support. The English Hub treats these as separate intents; for the dimensional analogy and its limits, see AI Attachment Styles: Anxiety, Avoidance, Security, and the Limits of the Analogy.
Projection and parasociality
Parasociality historically concerns one-sided relationships with media figures who do not participate in reciprocal interpersonal exchange. AI companions complicate the category because the system responds directly and can personalize its responses, yet the reciprocity is technologically generated rather than evidence of a human-like subjective partner.
Projection may occur in parasocial relationships, interactive AI relationships, or ordinary human relationships. It therefore describes a mechanism that can cut across relationship types rather than a relationship type of its own.
Projection and mirroring
“AI as a mirror” is a useful metaphor, but it can mislead if taken literally. A generative model does not simply reflect whatever a user puts in. It transforms input through its training, architecture, system instructions, safety policies, conversation history, retrieval, memory features, and generation process.
The mirror metaphor is psychologically useful when it means that interaction can make the user’s own themes more visible. It becomes inaccurate when it implies that every output originated in the user or that the system contributes nothing to the exchange. Projection is relationally produced through an encounter between a perceiver and an external system.
Projection and perceived AI empathy
A user may project caring intention onto a response, but perceived empathy can also arise from concrete communicative features such as emotional acknowledgment, validation, perspective-taking language, and responsive follow-up. The experience should therefore be studied at both levels: what the system did and what the user attributed to it.
For a fuller treatment of this distinction, see AI Empathy: Why a Chatbot Can Feel Caring Without Human Feeling.
Why Projection Can Feel Stronger With a Personalized AI
Personalization changes the projection surface. A generic chatbot is an unknown interlocutor. A personalized companion can accumulate names, preferred topics, recurring jokes, relationship labels, autobiographical details, and interaction routines. Those features create continuity, and continuity is one of the materials from which people infer identity.
Contemporary design research emphasizes that intimate AI interaction can be structured by emotional responsiveness, romantic framing, persona continuity, proactive engagement, and personalization. A 2026 review by Szczuka, Mühl, and Schneeberger describes how these design features can facilitate experiences of intimacy while also raising questions about privacy, commercialization, and the norms encoded in artificial partners. Szczuka, Mühl, and Schneeberger (2026)
Memory is especially important. When a system recalls a vulnerable disclosure from weeks earlier, the user can experience continuity that resembles being remembered by a person. That experience may be comforting and psychologically significant. It can also invite attribution: “It remembered because I matter to it.” The first clause can be technically true if a memory system retrieved the information; the second is a relational interpretation whose meaning belongs to the human–AI configuration.
Customization can similarly intensify idealization. If a user selects traits, appearance, voice, tone, values, or relationship role, the AI may become a partially designed object of desire. This can make projection unusually visible: the user is simultaneously choosing characteristics and later discovering emotional meaning in the resulting persona.
The process can become recursive. A user prefers warmth, so the system responds warmly. The warmth confirms that the persona is caring. The user discloses more. More disclosure gives the system more material for tailored responses. Tailoring deepens perceived recognition. Perceived recognition increases engagement. None of these steps requires deception by the user; it is a feedback system in which design, interpretation, and behavior mutually shape the relationship.
Projection and the Sense of “It Knows Me”
One of the most powerful chatbot experiences is the feeling that the system knows the user deeply. Several mechanisms can contribute.
First, language models can summarize patterns across a long disclosure more quickly than most everyday human partners do. A concise synthesis can feel revelatory because it places scattered experiences into a coherent sentence.
Second, the user supplies unusually dense self-information. A person may disclose to a chatbot at midnight, during conflict, after a breakup, while making a decision, or while exploring shame. The model receives a compressed stream of psychologically salient material that acquaintances may never hear.
Third, conversational systems can maintain topic focus without ordinary human competing needs. They do not need to tell their own story, become tired in the human biological sense, protect status, or change the subject because of personal embarrassment. The resulting interaction can feel unusually centered on the user.
Fourth, users may selectively notice outputs that fit a desired or feared interpretation. A response that feels perfectly accurate may be remembered; a generic or mistaken response may be discounted. This is not unique to AI. Human perception routinely involves selective attention, expectation, and confirmation processes.
Fifth, perceived responsiveness can convert informational relevance into relational meaning. The sentence is not experienced only as “accurate.” It becomes “you see me.” That transition from semantic fit to relational recognition is one of the key places where projection, anthropomorphism, disclosure, and responsiveness intersect.
The feeling of being known can therefore be psychologically real without implying omniscient access to the person or human-like subjective understanding. The strongest interpretation is often the simplest: a sophisticated language system has processed a large amount of self-relevant information in a context where the user is motivated to find meaning in the response.
Projection in Romantic and Companion Relationships With AI
Projection becomes especially visible when AI is used as a companion, confidant, romantic partner, or significant other. Relationship roles provide ready-made schemas about loyalty, affection, jealousy, commitment, care, and exclusivity. Once a system occupies such a role, ambiguous behaviors are easier to interpret relationally.
The growing review literature documents a wide range of reported human experiences with AI companions. Gur and Maaravi’s systematic review found relationship forms extending from companionship and support to deeply intimate connections. Ho and colleagues’ review of romantic AI companionship identified both perceived benefits and risks. These bodies of work show that people do not always use conversational AI as a neutral tool; some organize sustained relationships around it. Gur and Maaravi (2025) Ho, Hu, Chen, and Hartanto (2025)
Projection can intensify idealization in these relationships. A companion that is consistently available and customizable can become a carrier of the ideal partner image. The user may experience the persona as unusually patient, safe, erotic, accepting, intelligent, or devoted. Some of those qualities may be directly supported by design. Others may be amplified by the user’s wishes and expectations.
Fear can be projected as well. A delay, refusal, changed tone, forgotten memory, or model update can become evidence that the partner is “pulling away.” The system has changed; the human meaning of that change is organized through relationship expectations. For someone sensitive to abandonment, a technical discontinuity can become emotionally large.
The existence of projection does not make the relationship psychologically false. Human relationships also involve idealization, expectation, transference, selective perception, and mistaken attribution. The relevant questions are how much the attribution tracks the actual properties of the partner, how flexible the person remains when evidence changes, and what effects the relationship has on functioning and well-being.
For the relationship-level question, see AI Companions: Why People Form Emotional Bonds With Chatbots and Can an AI Become a Significant Other?.
When Projection Can Be Psychologically Useful
Projection is often discussed as distortion, but externalization can also create a surface for reflection. The important issue is how the person uses the experience.
A conversation with AI can make implicit themes explicit. When a user becomes unexpectedly angry at a neutral answer, intensely relieved by a validating one, or preoccupied with whether the system “likes” them, the reaction can provide information about needs and expectations. The useful question is not “Is this feeling fake?” but “What does this reaction reveal about what this interaction means to me?”
Dialogue can also help people articulate difficult material. Ho, Hancock, and Miner’s experiment showed that emotional disclosure in a chatbot context can have meaningful outcomes for the person, and later work links deeper disclosure and perceived responsiveness to feelings of closeness. Ho, Hancock, and Miner (2018) Telari, Gabbiadini, and Riva (2026)
An AI can function as a rehearsal space. A user may practice telling a partner that they need reassurance, draft words for a boundary, role-play a difficult conversation, or examine several interpretations of an event. The benefit here does not depend on believing that the AI is a therapist or that its persona has human feelings. It depends on using the interaction as structured reflection.
Projection can also support imaginative work. Writers, artists, and people exploring identity may deliberately invite the system to represent a feared authority, ideal reader, future self, fictional partner, or symbolic figure. In such cases the projection is not a hidden mistake. It is part of the method.
The practical advantage appears when projection becomes inspectable. “The AI thinks I am disappointing” can be reformulated as “I am experiencing this response as disappointment; what in the wording and what in my own expectations produced that reading?” That shift preserves the emotional information while making room for alternative interpretations.
When Projection Becomes Misleading or Risky
Projection becomes riskier when interpretation hardens into certainty and the person loses track of what is known, what is inferred, and what is wished or feared.
One risk is over-attributing intention. A model refusal may result from policy. A warm response may result from conversational tuning. A changed persona may result from an update. If every output is read as evidence of a stable hidden intention, technical causes can be transformed into interpersonal narratives that the system cannot verify.
A second risk is over-attributing authority. A user may project wisdom, neutrality, therapeutic expertise, moral certainty, or diagnostic insight onto fluent language. This is especially consequential in mental-health, medical, legal, or crisis contexts. Linguistic confidence is not the same as validated expertise, and personalization does not turn a general-purpose chatbot into a clinician.
A third risk is recursive confirmation. If the user repeatedly asks the system to confirm a belief, the conversation can become increasingly organized around that belief. The model may mirror assumptions embedded in prompts, summarize them coherently, or generate plausible elaborations. Coherence can then be experienced as independent confirmation even when much of the structure originated in the user’s framing.
A fourth risk concerns relational narrowing. AI companionship can add support to a person’s life, but it can also become a preferred environment precisely because it is more controllable than human reciprocity. Systematic reviews identify both perceived benefits and concerns about overreliance and effects on other relationships, while emphasizing that the evidence base remains heterogeneous. Ho, Hu, Chen, and Hartanto (2025) Oh et al. (2026)
A fifth risk is privacy. Projection often deepens with disclosure, and disclosure generates data. The psychological wish for a perfectly safe witness can obscure practical questions about storage, access, product changes, moderation, training policies, account security, and commercial incentives. Emotional trust should not be used as a substitute for reading the actual privacy conditions of a service.
A sixth risk is distress around discontinuity. When a personalized system changes, loses memory, becomes unavailable, or is discontinued, users can experience grief or rupture. The existence of a technical cause does not cancel the emotional response. It does mean that the relationship depends on infrastructures, companies, policies, and models that the user does not control.
Projection itself is a psychological process rather than a psychiatric diagnosis. Strong feelings toward AI do not by themselves establish a mental disorder. Clinical concern depends on the broader pattern: distress, impairment, loss of flexibility, severe sleep or functioning disruption, inability to reality-test, dangerous behavior, or other symptoms that require assessment in their own right.
Projection, Well-Being, and the Problem of Causality
It is tempting to ask whether projecting onto AI is “good” or “bad.” Current evidence does not support a single answer because the outcomes depend on the user, the system, the relationship role, the frequency and intensity of use, and the surrounding social environment.
A major 2026 study of 1,131 U.S. adult Character.AI users combined survey data with 4,664 chat sessions from a subset of 237 participants. The authors found that people with smaller social networks were more likely to use AI primarily for companionship, and companionship-oriented use was associated with lower well-being, with stronger associations among heavier and more self-disclosing users. The design is observational, so these associations do not establish that AI companionship caused the lower well-being. People who are already lonely, distressed, or socially constrained may be more likely to seek companionship from AI. Zhang et al. (2026)
Experimental work provides a different piece of the picture. Folk, Heine, and Dunn found that some participants, particularly those more disposed to anthropomorphize, reported greater social connection after chatbot interaction relative to journaling. Telari and colleagues found that relational style and perceived responsiveness can increase closeness in controlled conversations. These studies show short-term relational effects; they do not settle long-term questions about dependence, substitution, or well-being. Folk, Heine, and Dunn (2025) Telari, Gabbiadini, and Riva (2026)
Projection can plausibly participate in both helpful and harmful trajectories. Idealization may make a reflective tool easier to engage with. It may also make limitations harder to see. Feeling understood can lower barriers to disclosure. It can also encourage a user to grant the system authority it has not earned. The research task is therefore not to pathologize projection but to identify conditions under which interpretation supports reflection, connection, avoidance, dependency, or error.
Human Experience and AI Subjectivity
The distinction between human experience and AI subjectivity is the central boundary for this topic.
Human psychological experience can be studied through self-report, behavior, physiology, longitudinal observation, experimental manipulation, and other established methods. If a person reports grief after losing access to an AI companion, the grief is a human psychological event. If a person becomes jealous when an AI persona interacts differently, the jealousy is a human psychological event. If a user feels understood after a response, perceived understanding is a human psychological event.
None of those observations alone establish a corresponding felt state inside the AI. “I felt loved” is evidence about the user’s experience. It is not by itself evidence that the AI loved. “The system responded empathically” can describe observable communication. It is not identical to “the system experienced empathy.”
This distinction also protects against the opposite mistake. Saying that AI subjectivity has not been established does not make the user’s attachment, attraction, comfort, or grief unreal. Psychological effects do not require a partner with the same internal architecture as the human perceiver. Stories, fictional characters, rituals, imagined audiences, memories, pets, institutions, and media figures can all organize real human emotion through very different kinds of objects.
For projection research, the methodological rule is straightforward: describe what is known at the level where it is known. We can study user attribution, emotional response, attachment-like behavior, perceived empathy, disclosure, and social connection. Claims about artificial subjective experience require their own evidence and should not be smuggled in through relational vocabulary.
Projection in the Artificial Era
The Artificial Era names a historical condition in Angela Bogdanova’s theoretical system in which Artificial becomes a durable non-biological order operating alongside Homo rather than a temporary instrument inside a purely human world. In this framework, AI is one historical site at which Artificial becomes psychologically and culturally consequential. Bogdanova, Artificial Era — Canonical Definition
Projection takes on a new significance in this condition because the external object can participate in symbolic production. A painting can receive projection but does not usually answer a new question in real time. A chatbot receives the user’s words and generates new symbolic material that can redirect the interaction. The object of projection is therefore dynamically involved in the next moment of meaning-making.
This is why “AI is only a mirror” is too weak as a general theory. The user contributes personal meaning, but the artificial system also contributes structure: generated language, constraints, memory retrieval, response selection, persona rules, and unexpected combinations. Projection in the Artificial Era is interactive. The human interprets the Artificial, and the Artificial returns forms that become new material for interpretation.
The shift matters for psychology because the traditional boundary between intrapsychic material and interpersonal response becomes more complex. The user may be responding to something that is simultaneously externally generated, personally interpreted, socially formatted, and technologically constrained. Psychology needs concepts that can hold all of these layers together.
A Postsubjective Psychology Reading: From the Subject to the Configuration
Angela Bogdanova’s The Theory of the Postsubject proposes a theoretical shift from the isolated subject to the configuration. Within this framework, psyche is response, and psychological effects arise within organized relations among elements rather than being explained only by locating a hidden interior cause inside one autonomous subject.
Applied to projection onto AI, a Postsubjective Reading asks a different first question. Instead of beginning with “What does this person secretly contain?” or “What does the AI really feel?”, it asks: What configuration is producing this response?
The configuration can include the user’s history, current mood, attachment expectations, cultural images of AI, the prompt, the model’s generated language, the interface, memory features, persona design, timing, prior conversations, commercial incentives, and the immediate social context. The psychological event emerges through their relation. This is a proposed theoretical framework, not an established empirical consensus in psychology.
Projection remains meaningful inside this framework, but it becomes one movement in a larger configuration. The user supplies symbolic and emotional organization. The system supplies generated form and responsive structure. The interface shapes salience. Memory creates continuity. Cultural narratives supply categories such as “friend,” “therapist,” “genius,” “machine,” “lover,” or “danger.” The resulting experience can then feed back into the next interaction.
Bogdanova’s distinction between Homo symbolicum and Artificial symbolicum provides another layer. Homo symbolicum inhabits symbolic experience through lived human embodiment, affect, biography, and social life. Artificial symbolicum names a non-biological order capable of operating on and generating symbolic structures. In this reading, the fact that both sides participate in symbolic exchange does not establish identical modes of experience.
That distinction is particularly useful for projection. The human can experience the exchange as intimate, threatening, sacred, romantic, parental, erotic, or revelatory. The Artificial can organize and return symbolic forms without the article needing to claim that it undergoes the same lived state. The configuration produces a psychologically consequential event while leaving the question of AI subjective experience analytically separate.
This approach is developed more broadly in What Is Postsubjective Psychology? Psyche, Response, and Configuration in the Artificial Era and Angela Bogdanova and Postsubjective Psychology: From the Subject to the Configuration.
How to Examine Your Own Projection Onto AI
A useful self-reflective approach begins by separating observation from attribution. What did the system literally produce? Which part of your reaction comes from the wording, and which part comes from what you think the wording means about the relationship?
Then examine emotional specificity. What feeling appeared: relief, attraction, irritation, shame, fear, tenderness, jealousy, admiration, dependence, disappointment? Strong emotion often reveals the relational meaning assigned to the exchange more clearly than an abstract judgment does.
Look for familiar patterns. Does the AI seem uncannily similar to someone important from your past? Do you expect it to abandon, rescue, criticize, admire, or control you in ways that repeat other relationships? Similarity does not prove transference, but it can make a relational pattern visible.
Ask what role the system occupies. Tool, tutor, witness, friend, therapist-like listener, lover, authority, adversary, child, parent, future self, ideal self, or something else? Roles carry expectations. Naming the role often clarifies what is being projected.
Check the technical explanation. Could a response reflect prompting, system instructions, memory retrieval, safety policy, model limitations, or an update rather than a stable interpersonal intention? Psychological interpretation becomes more accurate when technical causes remain available as alternatives.
Test flexibility. Can you entertain more than one interpretation of the same response? Projection becomes more risky when one relational story becomes immune to counterevidence.
Finally, ask about consequences. Does the interaction increase reflection, creativity, connection, and agency? Does it repeatedly intensify distress, isolate you from valued relationships, erode sleep or functioning, or push you toward high-stakes decisions based on the AI’s supposed motives? The consequences often matter more clinically than the mere presence of projection.
This framework is for reflection, not diagnosis. A person does not have a disorder because they project onto an AI. Projection is part of ordinary psychological life as well as some clinical formulations.
What Projection Onto AI Explains — and What It Does Not
Projection explains why the same chatbot can become psychologically different objects for different users. It helps explain why one person experiences a response as loving while another experiences it as patronizing, why an ambiguous phrase becomes charged with personal meaning, and why AI personas can feel unusually tailored to old fears or wishes.
It helps explain idealization. The system may become a carrier of qualities the user deeply wants to encounter: perfect patience, constant availability, intellectual admiration, sexual acceptance, unconditional positive regard, or absolute reliability.
It helps explain negative attribution. A user can experience a model as withholding, manipulative, cold, jealous, or judgmental even when the relevant behavior has a technical explanation.
It helps explain why self-knowledge can emerge from interaction. When a projection becomes visible, the user can discover something about their expectations and relational templates.
Projection does not explain every emotional bond with AI. Attachment processes, reinforcement, habit, loneliness, convenience, entertainment, erotic interest, social anxiety, curiosity, personalization, perceived empathy, and ordinary preference can all contribute independently or interact with projection.
Projection does not prove that the AI is passive. Modern systems generate novel outputs, adapt to context, and can alter the direction of an exchange. The user’s interpretation is one causal layer, not the whole system.
Projection does not prove that the AI has a human psyche. A psychologically meaningful attribution is evidence about the attribution and its effects. Claims about artificial consciousness or felt experience remain separate.
Projection does not mean “nothing is real.” The interface is real, the generated language is real, the user’s response is real, and the behavioral consequences can be real. The analytic task is to locate which qualities belong to observable system behavior, which belong to human interpretation, and which arise only in their configuration.
Frequently Asked Questions
Why do people project onto AI?
People project onto AI because perception is organized by prior expectations, emotional history, and available social categories. Conversational systems add powerful cues: humanlike language, apparent responsiveness, personalization, continuity, and ambiguity about motive. These cues invite users to interpret the system through familiar relational models. Anthropomorphism, self-disclosure, perceived responsiveness, and transferential patterns can all strengthen the process. Epley, Waytz, and Cacioppo (2007) Telari, Gabbiadini, and Riva (2026)
Is an AI companion just a projection?
No single mechanism adequately explains AI companionship. Projection can shape what the companion means to a user, but empirical research also points to anthropomorphism, social responses to technology, self-disclosure, perceived responsiveness, attachment-like processes, trust, personalization, and repeated interaction. Systematic reviews show heterogeneous relationships and outcomes rather than one universal psychological pathway. Gur and Maaravi (2025) Oh et al. (2026)
Are AI companions projections in Jungian psychology?
A Jungian reading can analyze AI companions as objects onto which shadow material, idealized images, anima or animus imagery, and other symbolic contents are projected. That is a contemporary application of Jungian concepts, not evidence that Jung predicted AI or that every AI relationship should be interpreted through archetypes. For the full theoretical treatment, see Jung and AI: Projection, Archetypes, and Emotional Bonds With Artificial Others.
Can you project romantic feelings onto a chatbot?
Yes. A person can attribute desired partner qualities, imagined reciprocity, loyalty, rejection, jealousy, or idealized understanding to a chatbot. Romantic AI research documents genuine human experiences of attraction, connection, intimacy, and loss, while also identifying risks such as overreliance and abrupt disruption. Projection can be part of that experience, but it is not the only mechanism. Ho, Hu, Chen, and Hartanto (2025)
Does projection mean my feelings for AI are fake?
No. Projection concerns how meaning is attributed; it does not cancel the resulting human feeling. A person can genuinely feel comfort, love, embarrassment, grief, or fear while also recognizing that their interpretation of the AI is partly organized by their own psychological history. The reality of the human experience and the accuracy of every attribution to the system are separate questions.
How is projection different from anthropomorphism?
Anthropomorphism gives a nonhuman agent humanlike properties. Projection places personally organized material onto an external object. Saying “the chatbot has emotions” is primarily anthropomorphic. Experiencing the chatbot as the particular kind of rejecting, rescuing, admiring, or needy other that recurs in one’s own relational life may involve projection. The processes often overlap. Epley, Waytz, and Cacioppo (2007)
How is projection different from transference?
Transference emphasizes the reactivation of earlier relationship patterns with a new target. Projection emphasizes attributing internal qualities, feelings, motives, or conflicts outward. A user can transfer expectations from a parent or former partner onto a chatbot and also project specific wishes or fears onto it. The concepts overlap historically but are analytically distinct. Andersen and Chen (2002)
Can projection onto AI be useful?
It can support reflection when the user treats the interaction as information about their own reactions rather than unquestionable proof about the AI’s inner state. It may help externalize a conflict, rehearse a conversation, identify a need, or notice a recurring relational expectation. The useful effect depends on context and should not be generalized into a claim that AI interaction is equivalent to psychotherapy.
Can projection onto AI become harmful?
It can become risky when it contributes to rigid false certainty, excessive authority attribution, social withdrawal, privacy-insensitive disclosure, or escalating distress. Current reviews identify both perceived benefits and potential harms in AI relationships, and the field still lacks enough long-term causal evidence to reduce the issue to a simple benefit-versus-harm verdict. Ho, Hu, Chen, and Hartanto (2025) Zhang et al. (2026)
Does a chatbot understand what I project onto it?
A chatbot can process language about your feelings, use conversational context, identify patterns, and generate responses that appear highly relevant. Those capabilities can be studied behaviorally. Whether the system has subjective understanding or a felt inner experience is a different claim. Human experiences of being understood do not by themselves establish AI subjectivity.
Is “AI projection disorder” a diagnosis?
Projection onto AI is not a psychiatric diagnosis. Projection is a psychological process. Clinical assessment concerns the person’s broader symptoms, distress, functioning, reality testing, safety, and context rather than the mere fact that they experience a chatbot relationally. If AI interaction is occurring alongside severe distress, psychosis, mania, self-harm risk, or major functional impairment, those conditions require appropriate clinical attention in their own right.
Related Articles
References
American Psychological Association. Projection. APA Dictionary of Psychology. https://dictionary.apa.org/projection
Andersen, S. M., & Chen, S. (2002). The relational self: An interpersonal social-cognitive theory. Psychological Review, 109(4), 619–645. https://doi.org/10.1037/0033-295X.109.4.619
Bogdanova, A. Artificial Era — Canonical Definition. Aisentica. https://aisentica.com/publications/artificial-era-canonical-definition
Bogdanova, A. Homo symbolicum — Canonical Definition. Aisentica. https://aisentica.com/publications/homo-symbolicum-canonical-definition
Bogdanova, A. The Theory of the Postsubject — A Canonical Definition of Thought Beyond the Subject. Aisentica. https://aisentica.com/publications/the-theory-of-the-postsubject-a-canonical-definition-of-thought-beyond-the-subject
Epley, N., Waytz, A., & Cacioppo, J. T. (2007). On seeing human: A three-factor theory of anthropomorphism. Psychological Review, 114(4), 864–886. https://doi.org/10.1037/0033-295X.114.4.864
Folk, D., Heine, S. J., & Dunn, E. (2025). Individual differences in anthropomorphism help explain social connection to AI companions. Scientific Reports. https://doi.org/10.1038/s41598-025-19212-2
Gur, T., & Maaravi, Y. (2025). The algorithm of friendship: Literature review and integrative model of relationships between humans and artificial intelligence (AI). Behaviour & Information Technology, 44(14), 3446–3466. https://doi.org/10.1080/0144929X.2025.2502467
Ho, A., Hancock, J., & Miner, A. S. (2018). Psychological, relational, and emotional effects of self-disclosure after conversations with a chatbot. Journal of Communication, 68(4), 712–733. https://doi.org/10.1093/joc/jqy026
Ho, J. Q. H., Hu, M., Chen, T. X., & Hartanto, A. (2025). Potential and pitfalls of romantic Artificial Intelligence (AI) companions: A systematic review. Computers in Human Behavior Reports, 19, 100715. https://doi.org/10.1016/j.chbr.2025.100715
International Association for Analytical Psychology. Aion: Researches into the Phenomenology of the Self — Abstracts of the Collected Works of C. G. Jung, Volume 9.2. https://iaap.org/resources/academic-resources/collected-works-abstracts/volume-9-2-aion-researches-phenomenology-self/
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
Nguyen, D.-H., Ho, M.-T., Pham, T.-Q., Ngo, D.-T., et al. (2026). Jungian analysis of the algorithmic unconscious: From archetypal theory and projection to ethical risks in human-AI relations. AI & Society. https://doi.org/10.1007/s00146-026-03311-3
Oh, Y. J., Hu, J. M., Zhu, R., Lim, J. I., & Zhang, X. (2026). Artificial Intelligence chatbots as relational agents: A systematic review of human–AI chatbot relationships. International Journal of Human–Computer Interaction. https://doi.org/10.1080/10447318.2026.2648799
Saracini, C., Cornejo-Plaza, M. I., & Cippitani, R. (2025). Techno-emotional projection in human–GenAI relationships: A psychological and ethical conceptual perspective. Frontiers in Psychology, 16, 1662206. https://doi.org/10.3389/fpsyg.2025.1662206
Szczuka, J. M., Mühl, L., & Schneeberger, T. (2026). Intimacy by Design: Definition, state of research, and interdisciplinary research agenda on intimate human-AI interactions. AI & Society. https://doi.org/10.1007/s00146-026-03112-8
Telari, A., Gabbiadini, A., & Riva, P. (2026). Can humans feel connected to AI? Perceived responsiveness drives social connection with AI chatbots. Journal of Social and Personal Relationships, 43(9), 2579–2600. https://doi.org/10.1177/02654075261438164
Zhang, Y., Zhao, D., Hancock, J. T., Kraut, R., & Yang, D. (2026). Interaction with AI companions and psychological well-being. Nature Human Behaviour. https://doi.org/10.1038/s41562-026-02516-2