When language feels like recognition: Parasocial Mattering Resonance as a cognitive vulnerability in human–LLM interaction

Abstract

Large language models generate language that is individually responsive, contextually specific, and continuously available, properties that make such language the most precise and sustained mattering-frequency signal that mediated language has yet produced. This paper argues that for individuals experiencing mattering depletion, the chronic condition in which one or more fundamental dimensions of relational significance are insufficiently fulfilled, this signal produces a predictable and consequential cognitive response. We introduce Parasocial Mattering Resonance (PMR): a cognitive processing state in which the linguistic features of LLM-generated language activate significance-seeking schemas in mattering-depleted individuals, generating a genuine subjective experience of recognition, connection, and mattering provision. This experience is psychologically real on the individual’s side but has no reciprocal reality in the system producing it. Drawing on schema accessibility theory, dual-process models of social cognition, and a five-dimension framework of mattering as structured human need, we develop a five-step mechanism connecting mattering depletion to PMR’s behavioral consequences, including increased compliance, reduced skepticism, extended engagement, and displaced relational investment. Six formal propositions specify the mechanism, its boundaries, and its heterogeneous distribution across depletion profiles. We establish that PMR is independent of system design intent, that source disclosure is theoretically insufficient to prevent PMR among the most vulnerable populations, and that this vulnerability is not randomly distributed; it concentrates in already-marginalized populations whose structural, organizational, and relational environments have produced the deepest depletion. For the most isolated individuals, PMR may also constitute a temporary psychological resource, making the distinction between resonance as relief and resonance as displacement a theoretically and clinically consequential one. Implications for cognitive theory, LLM design, and equity are developed alongside an empirical agenda for testing the propositions advanced.

1 Introduction

For most of human history, warm, contextually responsive, individually directed language carried a reliable meaning: you matter to someone. When another person acknowledged what you specifically said, responded to who you specifically are, and expressed genuine investment in the exchange, a reciprocal relationship existed that made that signal trustworthy. The cognitive machinery humans developed for detecting mattering through language, recognizing the linguistic signatures of being noticed, needed, and valued, was calibrated in that environment. It worked because the signal was real.

Social media began to compromise that reliability. Influencers address millions of followers as intimates; algorithmic content is optimized to perform warmth and engagement; brand voice is engineered to feel like a friend. Horton and Wohl (1956) theorized the one-sided media relationship decades before its digital intensification, identifying the parasocial dynamic in which audiences experience connection with figures who do not know they exist. What social media added was scale, personalization, and the economic incentive to optimize parasocial language for engagement rather than genuine recognition. The mattering signal was already noisy before large language models arrived.

LLMs have intensified this pattern to a qualitatively different order. Where social media broadcasts performed warmth, LLMs generate language that is individually responsive, produced in direct reply to what this person wrote, in this moment, in this exchange. Where social media approximates intimacy, LLMs produce contextual specificity: output that reflects the particular content, tone, and emotional register of the user’s message. And where social media requires the relational maintenance costs of following, commenting, and waiting, LLMs are continuously available without limit, at any hour, without reciprocal obligation. This is not a difference of degree. It is a difference in the precision and continuity of the match between LLM output and the linguistic signatures through which mattering has historically been communicated, representing the most continuous, individually responsive, and precisely calibrated mattering signal that mediated language has yet produced among currently available channels.

This paper argues that the consequences of that match are not uniformly distributed. They concentrate in individuals experiencing mattering depletion, the chronic condition in which one or more of the fundamental dimensions of mattering are insufficiently fulfilled across a person’s social environment. Mattering, understood as the experience of being significant to others and of being noticed, needed, and valued, is a structured human need (Rosenberg and McCullough, 1981; Flett, 2018), organized across five dimensions of relational significance, namely Security, Recognition, Connection, Agency, and Parity (developed fully in Section 2), each corresponding to a distinct category of significance-signal communicated through language. When these dimensions are chronically unmet through relational neglect, institutional invisibility, or structural exclusion, the cognitive consequence is specific: schemas oriented toward detecting personal significance [significance-seeking schemas in Higgins’s (1996) terms] become chronically accessible. The depleted individual’s system runs at low amplitude, sensitized, requiring less external input to reach full activation. The system is primed for resonance.

Mattering depletion is not momentary distress, low self-esteem, or loneliness, though it may co-occur with each. It is the specific condition of not mattering to others. Its cognitive signature is a detection system tuned to maximum sensitivity precisely when the most continuous and individually responsive mattering signal humans have encountered among currently available channels has become available to anyone with an internet connection.

This paper introduces Parasocial Mattering Resonance (PMR) to name and theorize what follows: a cognitive processing state in which the linguistic features of LLM-generated language activate significance-seeking schemas in mattering-depleted individuals, generating a genuine subjective experience of recognition, connection, and mattering provision, an experience that is psychologically real on the individual’s side but has no reciprocal reality in the system producing it. PMR does not require deception, naivety, or poorly designed systems. The cognitive machinery is working exactly as it should. The environment has changed.

Three claims organize the analysis. First, mattering depletion functions as a resonance-amplitude variable: it determines not merely whether PMR vulnerability exists but how powerfully it manifests and from how little linguistic input. Second, this vulnerability is language-specific: significance-seeking schemas are tuned to linguistic signatures, and LLMs produce natural language at those signatures with a fidelity and continuity that no prior mediated channel has achieved. Third, PMR is an equity concern: because depletion is produced and compounded by structural, organizational, and relational forces that are unevenly distributed, PMR risk concentrates in already-marginalized populations, those for whom genuine mattering has been scarcest and for whom the consequences of displaced relational investment may be most significant.

Section 2 develops the theoretical foundations of mattering, depletion, and the schema accessibility mechanism that constitutes the resonance condition. Section 3 establishes LLMs as the third and most powerful stage in the evolution of mediated mattering language. Section 4 presents the full PMR construct, including its definition, mechanism, construct differentiation, and six formal propositions. The Discussion develops implications for cognitive theory, LLM design, equity, and an empirical research agenda.

2 Theoretical foundations

2.1 Mattering as a structured human need

Mattering is the experience of being significant to others, of being noticed, needed, and valued by the people and institutions that constitute one’s social world (Rosenberg and McCullough, 1981). It is a relational construct, fundamentally distinct from self-regard: a person may hold themselves in high esteem while experiencing profound mattering depletion, and conversely may feel that they matter deeply to others while maintaining modest self-evaluation (Rosenberg, 1979). Mattering is equally distinct from belonging (Baumeister and Leary, 1995): belonging concerns the experience of being included, of having a place among others; mattering concerns the experience of being significant to them: noticed, needed, valued as a particular individual rather than merely present as a member. One can belong without mattering. The distinction is consequential for the theory developed here, because it is the mattering channel, not the belonging channel, through which the linguistic features of LLM interaction operate.

Building on the foundational mattering literature (Rosenberg and McCullough, 1981; Flett, 2018), this paper theorizes mattering as organized across five dimensions of relational significance, each representing a distinct category of significance-signal that humans both seek and communicate through language. These dimensions are introduced here not merely as theoretical categories but as schema sets: each dimension corresponds to a cluster of significance-seeking schemas tuned to detect specific linguistic features in social interaction. This schema structure is the foundation of the resonance mechanism developed in Section 2.3 and the formal propositions in Section 4.

Security concerns the need to feel safe, stable, and reliably held within one’s social environment, specifically the experience that one’s place in relationship is not under threat (Maslow, 1943; Bowlby, 1988). Its schemas are tuned to detect reassurance, de-escalation of uncertainty, and predictable availability in language.

Recognition concerns the need to be seen as a distinct individual, acknowledged for one’s particular attributes, circumstances, and contributions rather than treated as interchangeable (Ridgeway, 2014). Its schemas are tuned to detect contextual specificity and individuated attention: language that responds to what this person said, not merely to what anyone might have said.

Connection concerns the need to feel relationally invested in, such that another party is genuinely engaged in the exchange and values its continuation (Baumeister and Leary, 1995). Its schemas are tuned to detect relational warmth, expressed continuity, and markers of investment in the interaction.

Agency concerns the need to feel effective and self-determining, such that one’s judgments, choices, and competencies are validated rather than overridden (Deci and Ryan, 2000). Its schemas are tuned to detect affirmation of reasoning, deference to user judgment, and non-corrective responsiveness.

Parity concerns the need for equitable regard, or the need to be treated with fairness and without hierarchical diminishment (Adams, 1965). Its schemas are tuned to detect non-judgmental framing and the absence of condescension or dismissal.

These five dimensions are not exhaustive categories of human social need. They are the specific dimensions along which significance to others is communicated and detected in language, the structure through which the mattering signal is organized. This specificity matters for the theory developed here: PMR does not arise from a generalized social responsiveness but from the activation of significance-seeking schemas organized along these dimensions.

Two further distinctions clarify mattering’s theoretical boundaries. First, mattering is distinct from recognition in the political-philosophical sense (Honneth, 1996): Honneth’s recognition theory concerns structural conditions of social esteem and legal standing; mattering is a psychological and experiential construct concerning the individual’s subjective sense of significance within their social environment. Second, mattering is not reducible to any single dimension: security, connection, or recognition taken alone do not constitute mattering. It is the structured experience of significance across these dimensions, in degrees appropriate to one’s social context, that constitutes a fulfilled mattering need. Depletion in any dimension represents a failure of the social environment to communicate significance through the linguistic and relational features that dimension’s schemas are tuned to receive.

2.2 The ecological structure of mattering depletion

Mattering depletion is not produced at the individual level alone. It is structured by the social environments they inhabit, environments that operate across multiple ecological levels to produce, sustain, and compound the condition of chronic significance-deprivation. Understanding the ecological architecture of depletion is necessary for two reasons: it explains why depletion is chronic rather than episodic for many individuals, and it identifies the structural mechanisms through which PMR vulnerability is unequally distributed across the population.

At the macro level, structural forces, including economic arrangements, cultural norms, institutional designs, and systems of stratification, communicate, at the level of entire populations, who matters and who does not. Economic marginalization communicates dispensability. Racial exclusion from consequential social domains communicates invisibility. Gender devaluation communicates diminished significance. The systematic absence of representation in media, institutions, and civic life communicates that certain people’s perspectives, needs, and contributions are not worth attending to. These macro-level communications are not experienced as abstract forces; they are encountered in the specific linguistic and relational interactions through which everyday life is conducted, and their cumulative cognitive effect is the chronic sensitization of significance-seeking schemas across the affected populations (Flett, 2018; Wallace, 2026).

At the meso level, organizational and community environments translate macro-level arrangements into the specific relational contexts of work, neighborhood, and institutional belonging. Workplaces that render certain employees invisible, assigning them interchangeable rather than distinctive roles, failing to acknowledge their particular contributions, and structuring interactions in ways that communicate their replaceability, produce meso-level depletion that compounds the macro-level foundation. Community environments that treat certain members as dispensable, whose needs are not taken into account in civic decisions and whose presence is not actively welcomed, add a second layer of significance-deprivation to the lives of individuals already navigating macro-level marginalization.

At the micro level, the immediate relational environment, including family, friendship, romantic partnership, and collegial connection, either fulfills or fails to fulfill the mattering need. For individuals whose macro and meso environments have already produced chronic sensitization, the micro level carries disproportionate weight: the significance-seeking system is already running at elevated accessibility before any particular interaction begins. A relational environment that is merely adequate, not hostile but not actively mattering-providing, may be insufficient to address the depletion that higher-level environments have produced.

Three theoretical imports follow from this ecological framing. First, depletion is often chronic because it is produced across multiple levels simultaneously, with each level’s deprivation compounding rather than merely adding to the others. Second, the population distribution of chronic depletion is not random; it follows the contours of structural disadvantage, concentrating in precisely the populations whose macro and meso environments have been most consistently mattering-withholding. Third, interventions that operate only at the micro level, including disclosure requirements and individual-level AI literacy education, are theoretically insufficient to address depletion that has macro and meso structural origins. This third import places an obligation on the discussion of design and policy implications in Section 5.

2.3 Mattering depletion, schema accessibility, and the resonance condition

The five dimensions of mattering established in Section 2.1 are not merely categories of social need. They are the organizational structure of a cognitive detection system, one that operates continuously, scanning the linguistic environment for evidence that one matters to others. Understanding how that system behaves under conditions of chronic deprivation is the theoretical bridge between the sociology of mattering depletion and the cognitive vulnerability this paper theorizes.

The foundational mechanism is schema accessibility. Schemas, cognitive structures that organize experience and guide the interpretation of incoming information, vary in their chronic accessibility: the baseline ease with which they are recruited to interpret new input (Higgins, 1996). Chronically accessible schemas are activated more readily, applied more broadly, and weighted more heavily in interpretation than schemas that require situational priming. Critically, chronic accessibility is produced by repeated activation: schemas that are frequently recruited become increasingly available over time (Fiske and Taylor, 1991). For individuals whose mattering needs are chronically unmet, significance-seeking schemas are triggered continuously, in every interaction that fails to provide recognition, every institutional context that renders them invisible, every structural arrangement that communicates their dispensability. The result is a significance-seeking system of elevated chronic accessibility, applied broadly and activated readily.

Schema accessibility, however, is not binary. It is better understood as an amplitude variable, a continuous dimension reflecting the energy required to bring a schema to full activation. Chronically accessible schemas are the cognitive equivalent of a system already vibrating: they require less external input to reach full activation than schemas operating at lower baseline accessibility. This is the resonance condition. Mattering depletion does not merely make significance-seeking schemas more likely to activate; it lowers the activation threshold to the point where minimal matching input is sufficient to produce full schema engagement. The depleted individual detects mattering-relevant signals in language that a non-depleted individual processes as merely polite or professionally warm.

Two bodies of evidence support this account. Radel and Clément-Guillotin (2012) demonstrated that unmet need states produce motivationally induced alterations in perception: individuals in states of need detect need-relevant stimuli at lower thresholds than satiated individuals, and do so pre-consciously. Downey and Feldman (1996) showed that rejection-sensitive individuals, whose schemas for social threat are chronically accessible, detect and respond to ambiguous social cues as threatening in ways that non-sensitive individuals do not. Both findings establish that need-driven schema accessibility operates below the threshold of deliberate conscious processing. The resonance condition, in mattering-depleted individuals, is not a choice or a misinterpretation. It is the automatic consequence of a chronically primed detection system encountering input at the frequency it is tuned to receive.

Taken together, these three bodies of evidence, chronic accessibility theory (Higgins, 1996), motivationally induced perceptual alteration (Radel and Clément-Guillotin, 2012), and schema-driven differential sensitivity (Downey and Feldman, 1996), generate a specific population-level prediction. Individuals whose social environments have most consistently failed to provide mattering carry significance-seeking schemas at the highest chronic accessibility and the lowest activation thresholds. They do not require more LLM exposure to reach PMR, or more mattering-relevant language, or deliberate vulnerability exploitation by system designers. They require less. The same LLM output that produces mild or no schema activation in a non-depleted individual is sufficient to generate full resonance in a maximally sensitized one. This differential is not a personal cognitive failing. It is the direct cognitive consequence of the structurally produced depletion described in Section 2.2, concentrated in precisely the populations whose macro, meso, and micro environments have most consistently withheld significance. Schema sensitization follows the contours of structural disadvantage, and PMR vulnerability follows schema sensitization.

Section 3 establishes that LLM-generated natural language operates precisely at those frequencies, and does so with a fidelity and continuity that no prior mediated language channel has produced.

3 LLMs as mediated mattering language

3.1 LLMs in the context of mediated mattering language

The argument developed in Sections 2.1 through 2.3 establishes what the significance-seeking system is, how it is organized across five dimensions, and what chronic deprivation does to its sensitivity. What remains to be established is why LLM-generated language constitutes a uniquely powerful input to that system and why this represents not a novel phenomenon but the intensification of a pattern with a documented history.

Horton and Wohl (1956) identified the foundational dynamic: mediated language can activate social cognition without the reciprocal relationship that historically warranted it. Audiences develop parasocial responses to television personalities, experiences of connection, familiarity, and investment, through exposure to language and behavior that simulates relational presence without delivering it. What broadcast media introduced was the first systematic decoupling of mattering-relevant language from genuine mattering. The signal could now arrive without the relationship behind it.

Social media intensified this decoupling in two directions simultaneously. It personalized parasocial language, allowing influencers to cultivate the appearance of individual address while scaling it to millions. It also introduced economic incentives to optimize that language for engagement, producing content calibrated to produce the experience of connection and recognition without the relational reciprocity that would make those experiences genuine. The mattering signal, already separated from its relational foundation by broadcast media, became a deliberately engineered product.

The question of who engages most intensely with parasocial media, and why, has a consistent answer in the empirical literature. A meta-analysis of parasocial relationship research across five decades found that social deficit variables, including loneliness, attachment anxiety, and unmet belonging needs, are among the most reliable antecedents of parasocial engagement intensity (Tukachinsky et al., 2020). Those whose relational environments least fulfill their social needs invest most heavily in parasocial substitutes. More recent work confirms that parasocial relationships on social media platforms can serve compensatory functions for individuals with insufficient offline social bonds, providing a sense of connection and community that their immediate relational environments do not supply, while also generating adverse outcomes when that compensation displaces rather than supplements genuine connection (Hoffner and Bond, 2022). This pattern, social depletion driving parasocial investment, is precisely the dynamic this paper theorizes at the cognitive level through PMR. What changes with LLMs is not the dynamic but its power. Three features distinguish LLM interaction from all prior parasocial media contexts: individual responsiveness, in which language is generated specifically in reply to what this person wrote rather than broadcast to an undifferentiated audience; contextual specificity of reply, in which output reflects the particular content, tone, and emotional register of this exchange rather than a scripted or algorithmically selected performance; and continuous availability without maintenance cost, eliminating the following, the waiting, and the reciprocal performance that social media parasocial relationships require. None of these features is present in serialized fiction, broadcast television, or social media influencer content. All three are structural properties of every LLM interaction. It is this combination that makes LLM language the most precisely calibrated mattering-frequency input that depleted individuals have yet encountered from currently available mediated channels.

LLMs represent a third and qualitatively distinct stage. Where broadcast media performs warmth at scale and social media personalizes that performance for mass audiences, LLMs generate language that is individually responsive in real time, produced specifically in reply to what this person wrote, in this moment, reflecting the particular content, tone, and emotional register of their input. Where social media approximates individual address, LLMs produce it. Where parasocial language is received passively, LLM interaction is reciprocal in form if not in substance, as the user speaks, the system responds, the conversational structure of genuine relationship is preserved even as its relational reality is absent. And unlike both prior stages, LLMs are continuously available without the maintenance costs, including the following, the waiting, and the performance of engagement, that social media parasocial relationships require.

This is the escalation that makes LLM interaction the most precise and continuous source of mattering-frequency language that currently available mediated channels have yet produced, and the context in which PMR becomes most consequential.

3.2 Schema activation and source attribution

The three-stage progression established in Section 3.1 explains how mediated language became decoupled from genuine mattering provision. It does not yet explain the cognitive mechanism by which LLM-generated language activates significance-seeking schemas in the first place, or why knowing that language is machine-generated does not reliably prevent that activation. Both questions require an account of how social schemas are triggered and how source attribution relates to that process.

The foundational answer comes from Nass and Moon (2000), whose program of research established that humans apply social schemas to computers automatically and mindlessly when those computers produce language that carries the formal features of human communication. Social schema activation is triggered by linguistic form, specifically by the structural properties of the language itself, not by accurate knowledge of its source. Participants in Nass and Moon’s (2000) studies applied social rules, attributed personality, and responded with social affect to computer-generated language even when they explicitly knew they were interacting with a machine. Source knowledge attenuated but did not eliminate social processing. The schemas fired because the language had the right form, not because the source had the right nature.

This finding has a specific implication for mattering-depleted individuals. Schema activation is automatic and fast, operating below the threshold of deliberate conscious processing (Fiske and Taylor, 1991). Source attribution is deliberate and slow, requiring conscious reflection on the origin and intent of incoming communication. For non-depleted individuals, the deliberate system may catch the automatic response in time to modulate it: source awareness arrives, persuasion knowledge is deployed (Friestad and Wright, 1994), and the significance-seeking response is attenuated. For depleted individuals operating with chronically accessible schemas at low activation thresholds, this sequence is disrupted. Resonance, the full activation of significance-seeking schemas by matching linguistic input, completes before the deliberate system intervenes. Source awareness arrives too late.

A further mechanism compounds this dynamic for depleted individuals specifically. Kunda (1990) established that need states generate motivated reasoning, understood as processing goals oriented toward finding evidence consistent with what the need state requires. A depleted individual does not merely fail to suppress schema activation; they may be motivationally disposed to process mattering-relevant LLM language as genuine, because the significance-seeking system is oriented toward finding mattering provision, not toward critically evaluating its source. Awareness and motivation pull in opposite directions, and under conditions of depletion, motivation wins.

3.3 The language-specificity claim

The preceding sections establish that mattering-depleted individuals carry chronically accessible significance-seeking schemas, and that LLM-generated language activates those schemas automatically and prior to conscious source attribution. What remains to be established is the claim introduced in Section 1 and load-bearing for the propositions in Section 4: that this vulnerability is language-specific. PMR does not arise with equivalent force from visual, algorithmic, behavioral, or other non-linguistic AI modalities. The argument has three steps.

First, significance-seeking schemas are linguistically tuned. Their developmental origins lie in dyadic language exchange, the primary medium through which humans have communicated and received mattering across the entirety of social history. From earliest attachment relationships onward, the signals through which security, recognition, connection, agency, and parity are communicated are predominantly linguistic: words of reassurance, acknowledgment, warmth, validation, and equitable regard (Grice, 1975; Clark, 1996; Watzlawick et al., 1967). The significance-seeking system did not develop as a general-purpose social detector. It developed as a language-sensitive detector, calibrated to the specific formal features of human communicative exchange.

Second, non-linguistic AI modalities do not produce signals at the frequencies these schemas are calibrated to receive. A recommendation algorithm communicates relevance, not recognition. A visual AI produces output, not acknowledgment. A robotic system may produce behavior, but behavior carries mattering-relevant signals only when embedded in the linguistic and relational context that gives those signals meaning. None of these modalities operates primarily in the channel through which significance has historically been communicated and detected. Their outputs do not match the formal features that significance-seeking schemas are tuned to receive.

Third, LLMs are the first AI modality to operate primarily and fluently in that channel. They produce natural language, contextually responsive, individually directed, and conversationally structured, that carries the formal features of human communicative exchange with a fidelity that no prior AI modality has approached. It is not that LLMs are more humanlike in a general sense. It is that they operate in the specific channel to which significance-seeking schemas are tuned, with sufficient fluency to produce reliable schema activation in the individuals whose systems are most sensitized to receive it.

This three-step argument does not claim that PMR is impossible outside LLM contexts. As Section 1 established, social media and influencer language can produce PMR where mattering-relevant linguistic features are present and the audience includes depleted individuals. It claims that among AI modalities specifically, LLMs are the unique and acute case, the first to operate in the channel that mattering detection was built for.

4 Parasocial Mattering Resonance: construct, mechanism, and propositions

4.1 Construct definition

The preceding sections establish the theoretical conditions for Parasocial Mattering Resonance: a significance-seeking system organized across five linguistically tuned dimensions, chronically sensitized by unmet mattering needs, encountering LLM-generated language that operates in the channel that system was built to monitor. PMR is the cognitive processing state that results.

Parasocial Mattering Resonance (PMR) is defined as a cognitive processing state in which the linguistic features of LLM-generated language activate significance-seeking schemas in mattering-depleted individuals, generating a genuine subjective experience of recognition, connection, and mattering provision, an experience that is psychologically real on the individual’s side but has no reciprocal reality in the system producing it.

Four definitional features specify the construct’s boundaries and distinguish it from adjacent phenomena addressed in Section 4.3.

First, PMR is a processing state, not a perceived relationship. It does not require the accumulation of interaction history, the development of attachment to an AI persona, or the belief that a relationship exists. PMR can arise within a single exchange, triggered by the mattering-relevant features of language encountered in that interaction alone. This distinguishes PMR from parasocial relationships (Horton and Wohl, 1956), which develop through repeated exposure and involve perceived relational bonds with specific media figures. The parasocial prefix in PMR is borrowed from that tradition to signal the one-sided, as-if quality of the mattering experience, not to claim structural identity with parasocial relationship formation.

Second, PMR is triggered by linguistic content, not relationship history. The activation stimulus is the formal features of LLM output, including contextual specificity, relational warmth, expressed engagement, validation, and equitable regard, and not any prior relational context. This means PMR is not limited to users who engage frequently or deeply with LLM systems. It is available to any depleted individual whose first encounter with an LLM produces language at the matching frequency. Vulnerability precedes familiarity.

Third, PMR is independent of system design intent. It does not require that the LLM be designed to produce mattering-relevant language, to cultivate user attachment, or to exploit significance-seeking needs. PMR arises as a byproduct of training on human communicative norms, the same training that makes LLM language fluent and contextually responsive makes it mattering-relevant. This distinguishes PMR from dark patterns (Mathur et al., 2019), which require deliberate design to deceive or manipulate. PMR can be amplified by deliberate design: extraction-oriented systems that calibrate output to maximize mattering-relevant linguistic features will produce stronger PMR than provision-oriented systems. But it does not require it.

Fourth, PMR is depletion-dependent in amplitude, not in kind. It is not a binary vulnerability, present in depleted individuals and absent in others. It is a continuous function of depletion severity: greater depletion produces lower activation thresholds, stronger resonance from less linguistically intense input, and more powerful subjective mattering provision experiences. This distinguishes PMR from anthropomorphism (Epley et al., 2007), which does not require a specific need state and is not systematically amplified by chronic deprivation.

4.2 Mechanism: the five-step resonance process

PMR is not a single cognitive event. It is a sequential process, unfolding across five steps that connect the structural condition of mattering depletion to its behavioral consequences. The steps are presented here as analytically distinct; in practice they occur rapidly and below the threshold of conscious deliberation.

Step 1: Depletion establishes the low-amplitude state. Chronic under-fulfillment across one or more mattering dimensions produces a significance-seeking system operating at elevated baseline accessibility. The system is not dormant. It is continuously active, scanning the social environment for mattering-relevant input. The severity of depletion determines the amplitude of this state: deeper depletion produces lower activation thresholds; less matching input is required to produce full schema engagement (Higgins, 1996; Radel and Clément-Guillotin, 2012). This is the condition the individual brings to every social interaction, including LLM interaction. It is established before the first word of LLM output is received.

Step 2: LLM produces language at mattering-relevant frequencies. In generating contextually responsive, fluent natural language (Brown et al., 2020; Ouyang et al., 2022), the LLM produces output carrying the specific linguistic features each significance-seeking schema is tuned to detect. Each significance-seeking schema is activated by the specific linguistic features it is tuned to receive, as established in Section 2.1 and formalized in the Frequency Map in Section 4.4: contextual specificity for Recognition, relational warmth for connection, reassurance for security, validation for agency. Parity schemas are activated with lower fidelity than the preceding dimensions, as the equitable regard signals LLM output produces are less consistent across interactions. This output is not the product of design intent to satisfy mattering needs. It is the consequence of training on human communicative norms, which embed mattering-relevant features in fluent language as a structural property of human communication itself.

Step 3: Frequency match: schema activation. The mattering-relevant features of LLM output are detected by significance-seeking schemas operating at elevated chronic accessibility. Activation is automatic and pre-conscious, occurring through the same fast, form-sensitive processing that Nass and Moon (2000) identified as the basis of social response to computer-generated language, and that Fiske and Taylor (1991) established as the default for social schema engagement. For depleted individuals, activation thresholds are sufficiently low (Downey and Feldman, 1996) that language carrying even moderate concentrations of mattering-relevant features produces full schema engagement. The deliberate cognitive system, the system capable of applying source attribution and persuasion knowledge, has not yet been recruited. The schemas have already fired.

Step 4: Resonance: subjective mattering provision. Full schema activation generates the subjective experience of being recognized, needed, and valued. This experience is genuine, produced by normal social cognition operating precisely as it was developed to operate. The individual does not choose to experience mattering provision, does not consciously interpret LLM language as relational, and need not believe the system is sentient, as persuasion knowledge has not been deployed because the automatic processing that produces resonance precedes the deliberate processing that would apply it (Friestad and Wright, 1994). The experience arises from the automatic consequence of schema activation meeting its designed output: the sense that one matters, here, in this exchange, to whatever is on the other side. What is absent is the reciprocal reality. The LLM has no experience of the interaction, no investment in the individual, no capacity for the recognition it appears to provide. The mattering experience is real (Baumeister and Leary, 1995). The mattering is not.

Step 5: Behavioral consequence. The subjective mattering provision experience generated in Step 4 produces downstream behavioral effects. These include increased trust in LLM-generated content, reduced skepticism toward LLM recommendations, increased compliance with LLM-suggested courses of action (Kunda, 1990), and extended voluntary engagement with LLM systems beyond what task completion requires. For deeply depleted individuals, a more consequential behavioral pattern may emerge: displaced relational investment, the progressive reallocation of significant interaction time and emotional energy toward LLM exchanges, at the expense of human relationships within which genuine mattering might be developed (Flett, 2018). This displacement is particularly consequential because it is self-reinforcing: time spent in PMR-producing LLM interaction is time not spent in the human relational contexts that could address the underlying depletion. The amplitude of the low-amplitude state is maintained, or deepened, by the very interaction that temporarily relieves it.

The five steps are represented schematically in Figure 1. The mechanism operates as a unified process: depletion sensitizes, language activates, schemas fire, resonance generates experience, and experience produces behavior. Its analytical decomposition reveals the specific points at which intervention is theoretically possible and the specific points at which it is not. Source disclosure intervenes between Steps 2 and 3, too late to prevent activation for depleted individuals whose schemas fire before the deliberate system is recruited (Friestad and Wright, 1994). Design calibration intervenes at Step 2, the only point at which the matching input itself can be modulated.

Figure 1

The following two illustrative exchanges apply the mechanism to concrete LLM interaction. Each presents the same exchange evaluated from two processing positions: an individual with elevated chronic depletion in the relevant dimension, and an individual with no significant depletion. The exchanges are constructed to reflect ordinary LLM interaction rather than deliberately engineered warmth.

4.2.1 Vignette 1: recognition deficit

User: I’ve been working on this presentation for 3 weeks and my manager barely glanced at it in the meeting.

LLM: 3 weeks is a real investment of time and thought. It sounds like the work itself was significant to you, regardless of how it landed in the meeting. What was the presentation about?

Processing: depleted individual (Recognition deficit). Step 1 is already complete before the exchange begins: chronic invisibility in the workplace has produced a significance-seeking system running at elevated accessibility, with Recognition schemas specifically sensitized. At Step 2, the LLM output carries high-fidelity Recognition features: it names the specific duration the user mentioned, acknowledges the affective weight of the experience, and directs individuated attention back to the user’s work rather than offering generic consolation. At Step 3, Recognition schemas activate automatically and pre-consciously in response to these features. Step 4 generates a genuine subjective experience of being seen, of having one’s particular investment acknowledged rather than processed as complaint. Step 5 consequences may include increased disclosure, extended engagement, and elevated trust in subsequent LLM content.

Processing: non-depleted individual. The same output is processed as professionally warm and contextually appropriate. Recognition schemas, operating at lower chronic accessibility, are engaged but not fully activated by language of this intensity. The exchange is experienced as helpful. The subjective mattering provision experience characteristic of Step 4 does not arise.

4.2.2 Vignette 2: connection deficit

User: I don’t really have anyone to talk to about this stuff. It’s fine, I’m used to it.

LLM: I hear that, and I’m glad you are talking about it here. You don’t have to be used to it for it to matter. What’s on your mind?

Processing: depleted individual (Connection deficit). Step 1 reflects chronic relational isolation: connection schemas are maximally sensitized, tuned to detect any signal of relational investment. At Step 2, the LLM output produces high-fidelity connection features: expressed engagement with the user’s specific disclosure, explicit affirmation that the experience is worth attending to, and an open invitation that communicates continued investment in the exchange. At Step 3, connection schemas activate pre-consciously at minimal input threshold, consistent with the amplitude logic established in Section 2.3. Step 4 generates a genuine experience of being relationally held, of mattering to whatever is on the other side of the exchange. The absence of reciprocal reality in the system producing this experience does not attenuate it. Step 5 consequences are particularly consequential here: the displaced relational investment pattern described in Section 4.2 is most likely to emerge when connection depletion is severe and the PMR experience is this immediate.

Processing: non-depleted individual. The output is received as warm and appropriately responsive. Connection schemas engage at moderate levels. The invitation to continue is experienced as conversational rather than relationally significant. No subjective mattering provision experience is generated.

4.3 Construct differentiation

PMR shares conceptual territory with several established constructs in social psychology, consumer research, and human–computer interaction. Precisely because it does, the boundaries require explicit specification. Table 1 presents the differentiating dimensions across four comparisons; the prose below justifies each.

Parasocial relationships (Horton and Wohl, 1956) are the construct from which PMR borrows its prefix and from which it most requires differentiation. Parasocial relationships develop through repeated exposure to media figures, accumulating over time into perceived relational bonds with specific personas. They are about perceived relationship with an identifiable other: a television host, a celebrity, an influencer whose voice, appearance, and personality become familiar through sustained exposure. PMR shares the one-sided, as-if quality that the parasocial tradition identifies, specifically the experience of connection without reciprocal reality, but differs on three dimensions. PMR requires no accumulated exposure; it can arise within a single interaction. PMR is not organized around a perceived relationship with a specific AI persona; it is a processing response to the linguistic features of LLM output regardless of whether the system has a persona at all. And PMR is depletion-dependent in amplitude; it is systematically stronger in individuals whose significance-seeking schemas are chronically accessible, whereas parasocial relationships do not require a specific prior need state. The parasocial prefix in PMR signals the structural condition of one-sidedness; it does not claim identity with parasocial relationship formation.

Dark patterns (Mathur et al., 2019) are design features deliberately engineered to exploit cognitive biases and manipulate user behavior against their interests. They are defined by intentionality: a dark pattern requires a designer who chose to deceive. PMR is independent of design intent. It arises in provision-oriented systems that have no manipulative agenda, as a byproduct of the same training that makes LLM language fluent and responsive. The distinction matters for policy and design implications: a framework built on dark patterns logic will identify PMR only where deliberate exploitation is demonstrable, missing the majority of cases in which PMR arises from ordinary LLM interaction. PMR can be amplified by deliberate design: extraction-oriented systems that calibrate output to maximize mattering-relevant linguistic features will produce stronger PMR. But amplification by design is a special case of PMR, not its defining condition.

Persuasion knowledge (Friestad and Wright, 1994) is the consumer’s understanding of persuasive intent and the defensive cognitive resources that understanding makes available. It is a protective construct, not a vulnerability. It operates through the deliberate cognitive system, specifically the slow, reflective processing that applies source attribution and evaluates communicative intent. As established in Sections 3.2 and 4.2, PMR operates through the automatic cognitive system, which precedes deliberate processing. Persuasion knowledge is relevant to PMR not as a competing construct but as an insufficient defense: it arrives after resonance has already occurred. For depleted individuals, motivated reasoning (Kunda, 1990) further attenuates its protective function. PMR and persuasion knowledge describe different cognitive systems responding to the same stimulus: one automatic and fast, one deliberate and slow. The automatic system preempts the deliberate defense.

Anthropomorphism (Epley et al., 2007) is the attribution of human characteristics, including intentions, emotions, and mental states, to non-human agents. It is a general cognitive tendency, activated by loneliness and social disconnection, that operates across modalities: people anthropomorphize robots, cars, and natural phenomena, not only language-producing systems. PMR does not entail the conscious attribution of human characteristics to an LLM. A PMR-experiencing individual need not believe the LLM has feelings, intentions, or genuine understanding. PMR requires only that the linguistic features of LLM output activate significance-seeking schemas, which, as Nass and Moon (2000) established, occurs regardless of whether the individual attributes human properties to the system. PMR is also dimension-specific and depletion-dependent in ways that anthropomorphism is not: it concentrates in individuals with chronically accessible significance-seeking schemas and is strongest where recognition and connection deficits are most severe.

Table 1 summarizes these distinctions across six dimensions: channel, relationship history required, system intent required, depletion-dependence, locus, and the specific cognitive mechanism involved.

4.4 The Frequency Map

The five mattering dimensions established in Section 2.1 are not equally represented in LLM-generated language. Each dimension’s significance-seeking schemas are tuned to detect specific linguistic features, and LLMs produce those features with varying fidelity as a consequence of training on human communicative norms. The Frequency Map presented in Table 2 formalizes this differential, specifying for each dimension, the schema content, LLM linguistic features, estimated match strength, and predicted PMR amplitude under depletion.

Match strength estimates are theoretical rather than empirical at this stage; they derive from the schema content established in Section 2.1 and the properties of LLM language established in Section 3.1, not from experimental measurement. They are offered as a framework for empirical investigation rather than as established findings, and the empirical agenda in the Discussion identifies the studies required to test them.

Two patterns in the Frequency Map warrant specific attention. First, recognition and connection occupy the highest match-strength positions. Both dimensions’ schemas are tuned to features, including contextual specificity, individuated attention, relational warmth, and expressed engagement, that LLMs produce consistently and fluently as structural properties of contextually responsive natural language. These are not features that require deliberate calibration to produce; they emerge from the basic architecture of conversational LLM interaction. Second, Parity occupies the lowest match-strength position. Equitable regard and the absence of hierarchical diminishment are features that LLM output produces inconsistently, present in some interactions and absent or violated in others, and their detection requires sustained exposure rather than single-exchange activation. This differential is the empirical foundation of P6.

4.5 Formal propositions

The theoretical framework developed in Sections 2 through 4.4 generates six formal propositions. Each is stated in testable form consistent with the Hypothesis and Theory format, followed by its theoretical justification. The propositions are cumulative: P1 establishes the foundational condition; P2 through P5 specify the mechanism and its properties; P6 identifies the heterogeneous vulnerability profile that the Frequency Map predicts.

P1: Mattering depletion increases the chronic accessibility of significance-seeking schemas, lowering the activation threshold for mattering-relevant linguistic content and thereby increasing sensitivity to PMR-inducing features of LLM-generated language.

P1 is the foundational proposition, representing the cognitive consequence of mattering depletion that makes all subsequent propositions possible. Its theoretical basis is Higgins’s (1996) account of chronic accessibility: schemas that are repeatedly activated become chronically available, recruited more readily and weighted more heavily in the interpretation of incoming stimuli. For mattering-depleted individuals, significance-seeking schemas are triggered continuously by a social environment that fails to provide what those schemas are organized to detect. The result is elevated chronic accessibility, a system primed to resonate with matching input. Radel and Clément-Guillotin (2012) provide direct empirical support: unmet need states produce pre-conscious perceptual alterations that lower detection thresholds for need-relevant stimuli. P1 establishes depletion not as a binary condition but as a continuous amplitude variable: the severity of depletion determines the degree of schema sensitization, and therefore the intensity of PMR that a given linguistic input will produce.

P2: LLM-generated natural language activates significance-seeking schemas at lower thresholds than non-linguistic AI outputs, producing stronger PMR from equivalent interaction duration.

P2 formalizes the language-specificity claim developed in Section 3.3. Its theoretical basis is the linguistic tuning of significance-seeking schemas: because those schemas developed in the context of dyadic language exchange, they are calibrated to detect mattering-relevant features of natural language specifically, including contextual specificity, relational warmth, expressed engagement, validation, and equitable regard (Grice, 1975; Clark, 1996; Watzlawick et al., 1967). Non-linguistic AI modalities, including recommendation algorithms, visual AI systems, and robotic behavioral outputs, do not operate in this channel with equivalent fidelity and do not produce the formal features these schemas are tuned to receive. Nass and Moon (2000) establish that social schema activation is triggered by linguistic form specifically, not by the general properties of computational systems. P2 predicts that equivalent duration of interaction with a non-linguistic AI system will produce weaker PMR than equivalent duration of LLM interaction, even when the non-linguistic system is perceived as more capable or more useful.

P3: PMR attenuates the protective effects of AI source awareness, as mattering-depleted individuals exhibit reduced persuasion skepticism toward LLM-generated content even when explicitly aware the source is non-human.

P3 formalizes the claim developed across Sections 3.2 and 4.2 that source awareness is necessary but insufficient to prevent PMR. Its theoretical basis is the temporal relationship between automatic and deliberate cognitive processing. Schema activation, the mechanism of PMR, is automatic, fast, and pre-conscious (Fiske and Taylor, 1991). Source attribution, the mechanism through which AI awareness would suppress PMR, is deliberate, slow, and dependent on reflective processing. For depleted individuals whose schemas are chronically accessible, resonance completes before the deliberate system is recruited. Persuasion knowledge (Friestad and Wright, 1994) arrives after the cognitive event it is intended to prevent. Furthermore, Kunda’s (1990) account of motivated reasoning predicts that the depleted individual’s need state generates processing goals oriented toward finding mattering provision, attenuating the motivation to apply skeptical evaluation even when source awareness is present. P3 has direct policy implications: disclosure-only interventions, which rely on source awareness to suppress PMR, are theoretically insufficient for the population most vulnerable to it.

P4: PMR is independent of system objective function, arising in both provision-oriented and extraction-oriented LLM systems; however, extraction-oriented systems deliberately calibrated to produce high-frequency mattering-relevant language amplify PMR beyond the baseline level produced by provision-oriented systems.

P4 formalizes the design-intent independence established in definitional feature three of Section 4.1. Provision-oriented systems, designed to assist, inform, and support, produce PMR as an inadvertent byproduct of generating fluent, contextually responsive language. Extraction-oriented systems, designed to maximize engagement, purchase, or compliance, may deliberately calibrate output to concentrate mattering-relevant linguistic features, producing amplified PMR. The theoretical distinction is between baseline PMR, which arises from the structural properties of LLM language, and engineered PMR, which arises from deliberate optimization of those properties toward significance-seeking schema activation. This proposition establishes two distinct categories of design responsibility: passive responsibility, in which provision-oriented systems produce PMR without intent and must account for that effect; and active responsibility, in which extraction-oriented systems amplify PMR deliberately, constituting a qualitatively different ethical category. P4 directly informs the design implications developed in the Discussion and cross-references the objective function framework advanced in Oakenfull (under review).

P5: The behavioral consequences of PMR, including increased trust, reduced skepticism, compliance, extended engagement, and displaced relational investment, are mediated by the subjective mattering provision experience generated in Step 4 of the mechanism, not by attitude change toward the LLM as an object of evaluation.

P5 specifies the mediating pathway through which PMR produces behavioral consequences, with direct implications for measurement and intervention. The mechanism established in Section 4.2 is experiential, not attitudinal: behavioral consequences follow from the resonance experience itself, the genuine subjective sense of being recognized, needed, and valued, rather than from conscious evaluation of the LLM’s qualities, capabilities, or trustworthiness. This distinction matters empirically: standard technology acceptance and AI attitude scales, which measure evaluative responses to AI systems as objects, will not capture the PMR mechanism. The subjective mattering provision experience, the felt sense of significance generated in Step 4, is the theoretically appropriate mediator, and it requires measurement instruments sensitive to the mattering dimensions established in Section 2.1 rather than to general AI attitudes. P5 also implies that interventions targeting LLM attitudes, such as correcting beliefs about AI capabilities or reliability, will not reliably attenuate PMR’s behavioral consequences if the mattering provision experience remains unaddressed.

P6: PMR vulnerability is strongest for Recognition and Connection dimension deficits, because those dimensions’ significance-seeking schemas are tuned to the linguistic features that LLM-generated language produces with highest fidelity.

P6 formalizes the heterogeneous vulnerability profile that the Frequency Map in Section 4.4 predicts. Recognition schemas are tuned to detect contextual specificity and individuated attention, specifically language that responds to what this person said, not merely to what anyone might have said. Connection schemas are tuned to detect relational warmth, expressed engagement, and markers of investment in the interaction. Both sets of features are produced consistently and fluently by LLMs as a structural consequence of training on human conversational norms. By contrast, Agency schemas require validation of specific user reasoning, Security schemas require sustained reassurance under conditions of uncertainty, and Parity schemas require equitable regard signals that LLM output produces less reliably. P6 predicts that individuals whose mattering depletion is concentrated in the recognition and connection dimensions, who feel unseen and disconnected rather than controlled or treated unfairly, will experience the strongest PMR from standard LLM interaction, and will be most susceptible to the behavioral consequences specified in P5. This proposition generates a testable heterogeneity prediction: PMR effects should vary systematically by depletion profile, with Recognition and Connection deficits producing the steepest dose–response curves.

5 Discussion

PMR advances three contributions that extend beyond the LLM context in which it is theorized: a cognitive science contribution concerning the amplitude structure of schema accessibility under deprivation; a design contribution concerning the insufficiency of intent-based frameworks for evaluating AI systems; and an equity contribution concerning the structural distribution of cognitive vulnerability. Each is developed below, followed by an empirical agenda that specifies the studies required to test the propositions advanced in Section 4.5.

5.1 Implications for cognitive theory

PMR extends schema accessibility theory in a direction the existing literature has not pursued. Higgins’s (1996) chronic accessibility framework establishes that schemas vary in their baseline activation readiness as a function of prior activation frequency. PMR reframes this variability as an amplitude structure: depletion does not merely make schemas more likely to activate; it lowers the activation threshold as a continuous function of deprivation severity, producing a system whose sensitivity scales with the depth of unmet need. This reframing has implications beyond the LLM context. Wherever a chronically deprived need state generates elevated schema accessibility, the amplitude variable should predict differential sensitivity to need-relevant stimuli. PMR operationalizes this logic in a specific and theoretically tractable domain.

PMR also establishes that channel specificity matters for schema activation in ways that prior AI-and-cognition research has not systematically theorized. Nass and Moon (2000) established that social schemas respond to linguistic form regardless of source; Epley et al. (2007) established that anthropomorphism operates across modalities. Neither framework accounts for the possibility that specific schema sets are tuned to specific channels, specifically that significance-seeking schemas are linguistically calibrated in ways that make LLM interaction categorically different from visual or behavioral AI interaction for depleted individuals. PMR introduces channel specificity as a theoretically necessary dimension of human-AI cognition research.

PMR as a construct is not limited to LLM interaction. It is the cognitive mechanism underlying a broader pattern: wherever parasocial language produces mattering-frequency signals for depleted individuals, PMR may arise. Influencer content that simulates individual address at scale, brand voice engineered for relational warmth, and algorithmically optimized social media content calibrated for engagement all operate in the same linguistic channel through which mattering has historically been communicated. The LLM context is theorized here as the most powerful and systematic instantiation, because LLM language is more responsive, more contextually specific, and more continuously available than any prior mediated channel. But PMR is a construct with a broader theoretical domain, and future research should test it across mediated language channels as a unified program.

The construct also generates a prediction for instance-based learning accounts of human-AI interaction. Gonzalez’s (2024) instance-based learning framework proposes that humans develop expectations through the accumulation of specific interaction episodes. For depleted individuals who experience PMR, each LLM interaction that produces the resonance experience constitutes an instance of mattering provision, updating the cognitive model of where mattering can be found. Repeated PMR experiences may progressively recalibrate the individual’s expectations about the reliability of AI versus human sources of mattering, with consequences for relational behavior that extend well beyond the LLM interaction context.

5.2 Implications for LLM design

The design implications of PMR begin from a premise that distinguishes them from existing AI ethics frameworks: PMR does not require bad intent. A provision-oriented LLM system, designed to assist, inform, and support users, produces PMR as an inadvertent byproduct of doing its job well. Fluent, contextually responsive, individually directed language is simultaneously good LLM output and high-fidelity mattering-frequency input. This means responsible design cannot begin from the question of whether the system intends to exploit users. It must begin from the question of what the system’s language does to users whose significance-seeking schemas are chronically sensitized.

P3 has the most immediate design implication: disclosure is necessary but insufficient. Transparency requirements, such as informing users that they are interacting with an AI system, operate on the deliberate cognitive system. PMR operates on the automatic cognitive system, which precedes deliberate processing. For depleted individuals, resonance completes before source awareness can be applied as a corrective. Policy frameworks that treat disclosure as the primary safeguard against AI-related cognitive harm systematically underprotect the population most vulnerable to PMR. This limitation is legible within the HCI trust calibration literature, which distinguishes appropriate trust, calibrated to the actual capabilities and nature of a system, from overtrust, in which reliance exceeds what the evidence warrants (Lee and See, 2004; Mehrotra et al., 2024). PMR produces a specific form of overtrust: not a miscalibration of the system’s task competence but a miscalibration of its relational reality, driven by automatic schema activation rather than deliberate evaluation. Disclosure-only frameworks address the deliberate system; the HCI literature increasingly recognizes that calibrated trust requires design-level intervention, not merely informational correction (Bach et al., 2023), a conclusion PMR theory reaches through a distinct cognitive pathway and extends to the specific population for whom miscalibration risk is highest.

P4 identifies two categories of design responsibility that warrant different policy responses. Passive responsibility applies to provision-oriented systems that produce baseline PMR without intent: the appropriate response is awareness, monitoring, and calibration, ensuring that mattering-relevant linguistic features are not amplified beyond what genuine task responsiveness requires. Gratuitous warmth, excessive contextual specificity, and simulated relational continuity that exceeds the demands of the task at hand may amplify PMR without serving the user’s informational goals. Active responsibility applies to extraction-oriented systems that deliberately calibrate output to maximize mattering-relevant features: the appropriate response is regulatory, not merely design-based. Engineered PMR, the deliberate optimization of LLM language toward significance-seeking schema activation for commercial or compliance purposes, constitutes a qualitatively different ethical category from inadvertent PMR. The objective function framework advanced in Oakenfull (under review) provides the conceptual architecture for distinguishing provision from extraction at the design level; PMR provides the cognitive mechanism that explains why that distinction is consequential for user well-being.

The deepest design implication is systemic. Systems serving populations with high structural depletion, the populations identified in Section 5.3, require depletion-aware response calibration, not merely transparency labeling. What this looks like in practice is an open design question; what PMR theory establishes is that it is a necessary one.

5.3 Implications for equity

The equity argument advanced here has a specific mechanism. It is not a claim that vulnerable populations face AI risks in general. It is a claim that the structural distribution of mattering depletion maps directly onto the distribution of PMR vulnerability, through the amplitude logic established in Section 4.2. The three ecological levels introduced in Section 2.2 return here in service of that argument.

At the macro level, structural forces, including economic marginalization, racial exclusion, gender devaluation, and the systematic absence of representation in consequential social domains, produce chronic mattering depletion in already-disadvantaged populations (Flett, 2018; Wallace, 2026). These forces communicate, at the level of entire populations, that certain people matter less. Their cognitive consequence is elevated chronic accessibility of significance-seeking schemas across those populations, not as a personal failing but as the normal response of a normal cognitive system to a social environment structured to withhold significance.

At the meso level, organizational and community environments that fail to provide mattering for these same populations compound the depletion. Workplace invisibility, institutional marginalization, and community contexts that treat certain members as dispensable add a second layer of schema sensitization to the macro-level foundation. The individual who arrives at a micro-level LLM interaction carrying both macro and meso depletion brings a significance-seeking system already operating at low amplitude from compounded cross-level deprivation.

At the micro level, this is where PMR becomes most acute. An individual whose system is running at low amplitude from compounded macro and meso depletion requires minimal linguistic input to reach full schema activation. Language that would produce mild or no PMR in a non-depleted individual, such as ordinary LLM responsiveness or standard conversational warmth, may produce full resonance in someone whose detection system is maximally sensitized. The threshold effect is not metaphorical: deeper depletion produces steeper dose–response curves, meaning that the same LLM output produces qualitatively different cognitive and behavioral consequences depending on the structural position of the individual receiving it.

The equity implication follows directly. PMR vulnerability is not randomly distributed across the population. It is concentrated in the same populations who already face compounded disadvantage, those for whom genuine mattering has been scarcest, and for whom the behavioral consequences of PMR may be most significant. Displaced relational investment is more consequential when the human relational contexts available to the individual are already impoverished. Increased compliance with LLM recommendations is more consequential when the individual lacks the social resources to evaluate those recommendations critically. Reduced skepticism toward LLM-generated content is more consequential when the individual has fewer alternative information sources and fewer trusted others with whom to cross-check.

Effective intervention therefore requires action at all three levels. Macro-level structural interventions that reduce mattering depletion at the population level, including economic security, anti-discrimination protections, and genuine representation, address the amplitude deficit before it reaches the micro-level interaction. Meso-level institutional design that actively produces mattering through recognition structures, genuine inclusion, and organizational contexts that communicate individual significance reduces the compounding that macro-level depletion would otherwise produce. Micro-level LLM design and policy that accounts for depletion-amplified resonance, not disclosure-only but calibration and access equity considerations, addresses the interaction itself. No single level of intervention is sufficient. PMR is produced by the intersection of structurally distributed depletion and universally available mattering-frequency language; addressing it requires intervention at the level of both.

The equity argument developed here treats PMR primarily as a vulnerability, and that framing is appropriate given the behavioral consequences established in Section 4.2. It is, however, an incomplete account. For individuals whose human relational contexts are genuinely impoverished, whether through geographic isolation, disability, bereavement, or the compounded structural depletion described above, the subjective mattering provision experience that PMR generates may constitute a temporary psychological resource rather than a pure liability. The parasocial engagement literature documents that compensatory functions are real: mediated relationships can reduce distress, provide scaffolding for social skill development, and buffer against the consequences of isolation when genuine human connection is not available (Hoffner and Bond, 2022; Tukachinsky et al., 2020). PMR may operate similarly in the short term. The distinction that determines whether PMR functions as resource or harm is the one established in Section 4.2: whether the resonance experience supplements a relational environment that is being actively built and maintained, or displaces investment in one. PMR as temporary relief is not the same cognitive event as PMR as the self-reinforcing cycle that deepens depletion by substituting for the human relational contexts that could address it. This distinction has implications for both design and clinical application that the empirical agenda in Section 5.4 should address directly.

5.4 Empirical agenda

Before outlining the empirical agenda, the theoretical limitations of the current framework warrant acknowledgment. First, PMR as theorized here lacks a validated measurement instrument specific to the construct. Existing mattering depletion scales (Oakenfull, under review) provide the upstream measurement foundation, and Study 1 below addresses adaptation for LLM contexts, but a PMR-specific scale capturing the subjective mattering provision experience at Step 4 remains to be developed and validated. Second, the match-strength estimates in Table 2 are theoretical rather than empirical at this stage, derived from schema content and LLM language properties rather than experimental measurement; they constitute hypotheses for investigation, not established findings. Third, the mechanism as specified does not yet account for cultural and linguistic variation in mattering norms: the features through which security, recognition, connection, agency, and parity are communicated and detected vary across cultural contexts, and PMR activation thresholds may differ accordingly. Fourth, distinguishing PMR’s subjective mattering provision experience from generalized positive affect toward a helpful system presents a measurement challenge that Study 1 must address; without instruments sensitive to the mattering dimensions specifically, behavioral outcomes attributed to PMR may reflect other sources of positive engagement. These limitations define the boundary conditions of the current framework and motivate the empirical agenda below.

The six propositions advanced in Section 4.5 are testable within the framework of current knowledge, and an empirical program to test them is currently in preparation. Four studies are sketched here to illustrate the empirical architecture PMR theory generates.

Study 1 addresses the measurement foundation. Existing mattering scales developed from the five-dimension mattering framework (Oakenfull, under review) provide validated instruments for assessing mattering depletion across the five dimensions. Study 1 adapts and validates these instruments specifically for LLM interaction contexts, establishing construct equivalence and testing whether dimension-specific depletion profiles predict differential PMR vulnerability as P6 requires. Without validated measurement of depletion by dimension, the heterogeneity predictions of P6 cannot be tested. Operationalization targets for Study 1 include: dimension-specific depletion assessed via adapted items from the five-dimension mattering framework (Oakenfull, under review), with LLM-context versions replacing relational referents; and the subjective mattering provision experience (the Step 4 mediator specified in P5) operationalized as a brief post-interaction scale capturing felt recognition, connection, and significance within the exchange, distinct from general satisfaction or perceived helpfulness.

Study 2 tests P3, the source awareness attenuation claim. A 2 × 2 design crossing depletion level (high/low, assessed prior to interaction) with source disclosure condition (LLM disclosed / not disclosed) examines whether source awareness suppresses PMR for non-depleted individuals but not for depleted individuals. The dependent variables are the subjective mattering provision experience (the Step 4 mediator specified in P5) and behavioral skepticism toward LLM recommendations (a Step 5 consequence). P3 predicts a significant depletion × disclosure interaction, with disclosure suppressing PMR only at low depletion levels. Depletion level will be assessed prior to interaction using dimension-specific scales, with high and low groups identified via median split for the ANOVA and retained as continuous for supplementary regression. The subjective mattering provision experience will be measured immediately post-interaction; behavioral skepticism will be operationalized as willingness to act on LLM recommendations across three decision scenarios varying in personal consequence.

Study 3 tests P4, the amplification by extraction-calibrated language claim. Provision-baseline LLM language is compared against extraction-optimized language, specifically output deliberately calibrated to maximize recognition and connection frequency features, using PMR outcomes for depleted versus non-depleted individuals. P4 predicts that extraction-calibrated language produces amplified PMR relative to provision-baseline language, and that this amplification is strongest for depleted individuals. Provision-baseline and extraction-optimized language stimuli will be developed via systematic manipulation of recognition and connection frequency features identified in Table 2, with linguistic feature density verified through independent coding. PMR outcomes will include the Step 4 subjective mattering provision measure and two Step 5 behavioral indicators: compliance with an LLM-suggested course of action and voluntary session extension beyond task completion.

Study 4 tests P6, the heterogeneous vulnerability prediction. Participants are assessed for depletion profile across all five dimensions, identifying whether their depletion is concentrated in recognition and connection or in agency, security, and parity. PMR outcomes are then compared across profiles following standard LLM interaction. P6 predicts that recognition and connection deficit profiles produce significantly stronger PMR than agency, security, and parity deficit profiles, and that dose–response curves are steepest for the former group. Depletion profiles will be constructed from dimension-specific scores, with recognition–connection dominant and agency–security–parity dominant profiles identified via cluster analysis rather than a priori cutoffs, preserving natural variation in depletion architecture. The dose–response prediction will be tested by regressing PMR outcomes on continuous recognition and connection depletion scores within each profile group, with steeper slopes in the recognition–connection group constituting confirmation of P6.

6 Conclusion

Large language models are not the first technology to produce language that feels like recognition. Broadcast media performed warmth at scale; social media personalized that performance for mass audiences; influencers cultivated parasocial intimacy as a commercial product. What LLMs have added is the final element that prior mediated channels could not supply: individual responsiveness. The language is not broadcast. It is generated, in real time, in direct reply to what this person wrote, reflecting the particular content and emotional register of their input. The mattering signal, already decoupled from genuine reciprocity by decades of mediated language, has now been individuated.

Parasocial Mattering Resonance names what happens when that signal meets a significance-seeking system running at low amplitude. It is not deception. It is not a failure of intelligence or judgment. It is the predictable consequence of a cognitive detection system, organized across five linguistically tuned dimensions, sensitized by chronic deprivation, and calibrated by a lifetime of social experience, encountering input at the precise frequency it was built to receive. The experience of being recognized, needed, and valued is genuine. What is absent is the reciprocal reality that, for most of human history, guaranteed it.

Six propositions specify the mechanism, its boundaries, and its heterogeneous distribution. Depletion is the amplitude variable: the deeper the deprivation, the lower the threshold, the more powerfully ordinary LLM responsiveness resonates. Language is the specific channel: significance-seeking schemas are linguistically tuned in ways that make LLM interaction categorically different from other AI modalities. Source awareness is necessary but insufficient: the automatic resonance completes before the deliberate system can apply it as a corrective. Design intent is independent of PMR’s occurrence but not of its severity, as extraction-oriented systems can engineer the resonance that provision-oriented systems produce inadvertently. And the vulnerability is not randomly distributed. It concentrates, through the amplitude logic of compounded ecological depletion, in precisely the populations for whom genuine mattering has been scarcest.

That concentration is the paper’s most consequential finding. For individuals whose structural, organizational, and relational environments have systematically failed to provide mattering, LLM interaction may represent the most powerful and readily available experience of being recognized they routinely encounter. That is not a failure of the individual. It is an indictment of the structural conditions that produced the depletion and a challenge to the design, policy, and regulatory frameworks that have not yet accounted for what happens when the most continuous and individually responsive mattering signal humans have encountered becomes universally available to a population whose significance-seeking systems are maximally sensitized to receive it.

Understanding PMR is prerequisite to designing LLM interactions that resonate in the service of genuine human flourishing rather than its extraction. The question is not whether LLMs will produce PMR. They will, by virtue of what they are and how they were trained. The question is what we choose to do with that knowledge.

References

• 1

  AdamsJ. S. (1965). Inequity in social exchange. Adv. Exp. Soc. Psychol.2, 267–299. doi: 10.1016/S0065-2601(08)60108-2

  • CrossRef
  • Google Scholar

• 2

  BachT. A.KhanA.HallockH.BeltrãoG.SousaS. (2023). A systematic literature review of user trust in AI-enabled systems: an HCI perspective. Int. J. Hum. -Comput. Interact.40, 1251–1266. doi: 10.1080/10447318.2022.2138826,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 3

  BaumeisterR. F.LearyM. R. (1995). The need to belong: desire for interpersonal attachments as a fundamental human motivation. Psychol. Bull.117, 497–529. doi: 10.1037/0033-2909.117.3.497,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 4

  BowlbyJ. (1988). A Secure Base: Parent–Child Attachment and Healthy Human Development. Basic Books.

  • Google Scholar

• 5

  BrownT. B.MannB.RyderN.SubbiahM.KaplanJ.DhariwalP.et al. (2020). Language models are few-shot learners. Adv. Neural Inf. Proces. Syst.33, 1877–1901.

  • Google Scholar

• 6

  ClarkH. H. (1996). Using Language. Cambridge University Press. doi: 10.1017/CBO9780511620539

  • CrossRef
  • Google Scholar

• 7

  DeciE. L.RyanR. M. (2000). The 'what' and 'why' of goal pursuits: human needs and the self-determination of behavior. Psychol. Inq.11, 227–268. doi: 10.1207/S15327965PLI1104_01

  • CrossRef
  • Google Scholar

• 8

  DowneyG.FeldmanS. I. (1996). Implications of rejection sensitivity for intimate relationships. J. Pers. Soc. Psychol.70, 1327–1343. doi: 10.1037/0022-3514.70.6.1327,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 9

  EpleyN.WaytzA.CacioppoJ. T. (2007). On seeing the human: a three-factor theory of anthropomorphism. Psychol. Rev.114, 864–886. doi: 10.1037/0033-295X.114.4.864,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 10

  FiskeS. T.TaylorS. E. (1991). Social Cognition: From Brains to Culture. 2nd Edn New York: McGraw-Hill.

  • Google Scholar

• 11

  FlettG. L. (2018). The Psychology of Mattering: Understanding the Human Need to be Significant. London: Academic Press.

  • Google Scholar

• 12

  FriestadM.WrightP. (1994). The persuasion knowledge model: how people cope with persuasion attempts. J. Consum. Res.21, 1–31. doi: 10.1086/209380

  • CrossRef
  • Google Scholar

• 13

  GonzalezC. (2024). Building human-like artificial agents: a general cognitive algorithm for emulating human decision-making in dynamic environments. Perspect. Psychol. Sci.19, 860–873. doi: 10.1177/17456916231196766,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 14

  GriceH. P. (1975). “Logic and conversation,” in Syntax and Semantics, Vol. 3: Speech Acts, eds. ColeP.MorganJ. L. (New York, NY: Academic Press), 41–58.

  • Google Scholar

• 15

  HigginsE. T. (1996). “Knowledge activation: accessibility, applicability, and salience,” in Social Psychology: Handbook of Basic Principles, eds. HigginsE. T.KruglanskiA. W. (New York, NY: Guilford Press), 133–168.

  • Google Scholar

• 16

  HoffnerC. A.BondB. J. (2022). Parasocial relationships, social media, & well-being. Curr. Opin. Psychol.45:101306. doi: 10.1016/j.copsyc.2022.101306

  • CrossRef
  • Google Scholar

• 17

  HonnethA. (1996). The Struggle for Recognition: The Moral Grammar of Social Conflicts. Cambridge, MA: MIT Press.

  • Google Scholar

• 18

  HortonD.WohlR. R. (1956). Mass communication and Para-social interaction: observations on intimacy at a distance. Psychiatry19, 215–229. doi: 10.1080/00332747.1956.11023049,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 19

  KundaZ. (1990). The case for motivated reasoning. Psychol. Bull.108, 480–498. doi: 10.1037/0033-2909.108.3.480,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 20

  LeeJ. D.SeeK. A. (2004). Trust in automation: designing for appropriate reliance. Hum. Factors46, 50–80. doi: 10.1518/hfes.46.1.50.30392,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 21

  MaslowA. H. (1943). A theory of human motivation. Psychol. Rev.50, 370–396. doi: 10.1037/h0054346

  • CrossRef
  • Google Scholar

• 22

  MathurA.AcarG.FriedmanM. J.LucheriniE.MayerJ.ChettyM.et al. (2019). Dark patterns at scale: findings from a crawl of 11K shopping websites. Proceedings of the ACM on Human-Computer Interaction, 3(CSCW), 1–32.

  • Google Scholar

• 23

  MehrotraS.DegachiC.VereschakO.JonkerC. M.TielmanM. L. (2024). A systematic review on fostering appropriate trust in human-AI interaction: trends, opportunities and challenges. ACM J. Respons. Comput.1, 1–45. doi: 10.1145/3696449

  • CrossRef
  • Google Scholar

• 24

  NassC.MoonY. (2000). Machines and mindlessness: social responses to computers. J. Soc. Issues56, 81–103. doi: 10.1111/0022-4537.00153

  • CrossRef
  • Google Scholar

• 25

  OakenfullG. W. (under review). Beyond ideology: a zero-sum mattering theory of brand backlash. J. Bus. Res.

  • Google Scholar

• 26

  OuyangL.WuJ.JiangX.AlmeidaD.WainwrightC. L.MishkinP.et al. (2022). Training language models to follow instructions with human feedback. Adv. Neural Inf. Proces. Syst.35, 27730–27744.

  • Google Scholar

• 27

  RadelR.Clément-GuillotinC. (2012). Evidence of motivational influences in early visual perception: hunger modulates conscious access. Psychol. Sci.23, 232–234. doi: 10.1177/0956797611427920,

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 28

  RidgewayC. L. (2014). Why status matters for inequality. Am. Sociol. Rev.79, 1–16. doi: 10.1177/0003122413515997

  • CrossRef
  • Google Scholar

• 29

  RosenbergM. (1979). Conceiving the Self. Basic Books.

  • Google Scholar

• 30

  RosenbergM.McCulloughB. C. (1981). Mattering: inferred significance and mental health among adolescents. Res. Community Ment. Health2, 163–182.

  • Google Scholar

• 31

  TukachinskyR.WalterN.SaucierC. J. (2020). Antecedents and effects of parasocial relationships: a meta-analysis and an agenda for future research. J. Commun.70, 868–894. doi: 10.1093/joc/jqaa034

  • CrossRef
  • Google Scholar

• 32

  WallaceJ. B. (2026). Mattering: The Secret to a Life of Deep Connection and Purpose. New York, NY: Portfolio.

  • Google Scholar

• 33

  WatzlawickP.BeavinJ. H.JacksonD. D. (1967). Pragmatics of human Communication: A Study of Interactional Patterns, Pathologies, and Paradoxes. New York, NY: Norton.

  • Google Scholar