Beyond nudging: affective paternalism and affect architecture for impactful behavioral public policy

The affective gap in behavioral public policy

Behavioral economics has demonstrated, with a breadth and rigor that has genuinely changed the practice of governance, that people deviate systematically from the predictions of standard economic theory (Camerer et al., 2003a; Thaler and Sunstein, 2008). Heuristics, biases, and choice architecture effects are real, consequential, and predictable enough to be targeted by policy. The resulting toolkit of nudges has produced measurable improvements in retirement savings, organ donation rates, energy conservation, and nutritional choice. These are genuine advances, and nothing in what follows is intended to diminish them.

Yet behavioral public policy has a systematic blind spot. Its theoretical foundations are predominantly cognitive: it addresses how people think under uncertainty, how they process information, and how choice environments trigger cognitive shortcuts. The emotional dimension of decision-making (affect, the immediate felt sense of goodness or badness that accompanies perception and judgment) has been treated primarily as a source of noise to be corrected rather than as a mechanism to be engaged. This matters, because the domains where behavioral public policy is most urgently needed are precisely those where the primary barrier to better decisions is not cognitive but affective. People do not continue to underinvest in high-impact climate actions because they fail to understand climate change; they continue because carbon emissions feel abstract and distant while flying and eating meat feel rewarding and normal (Asutay et al., 2023). People do not direct charitable dollars toward lower-impact causes because they miscalculate comparative effectiveness; they direct them there because the local shelter evokes warmth and the global health initiative evokes little or nothing (Västfjäll et al., 2014). The problem is affective, not informational, and it calls for an affective solution.

This paper introduces Affective Paternalism (AP) as a theoretical framework for addressing this gap. AP treats affect as a primary resource for behavioral public policy rather than a secondary complication. It proposes the cudge as a new operative instrument: a liberty-preserving intervention that works by redesigning the emotional texture of choices rather than their cognitive architecture. And it introduces the CARE taxonomy (Create, Attenuate, Reinforce, Eliminate) as a systematic framework for categorizing and designing affective interventions. We develop these concepts against the backdrop of scope insensitivity and impact neglect, the best-documented cases in which affective responses are systematically decoupled from objective impact, and we argue that this decoupling creates a class of behavioral intervention problems for which affective approaches are not merely useful but welfare-superior to cognitive ones.

The behavioral economics of affect

Standard behavioral economics accounts of decision-making under risk and uncertainty identify two broad systems of judgment: a fast, automatic, associative system and a slow, deliberate, analytical one (Kahneman, 2011). Affective responses are, in dual-process terms, paradigmatic System 1 outputs: they arise rapidly, without deliberate reasoning, and they provide an evaluative signal that can substitute for analytical computation. The affect heuristic captures this substitution formally: people consult how an option feels as a proxy for its expected utility (Finucane et al., 2000; Slovic et al., 2002). One empirical signature of this mechanism is an inverse relationship between perceived risk and perceived benefit for the same option, even though risk and benefit are objectively independent quantities that should covary positively across activities in the world. The inverse relationship arises because both judgments are drawn from a single underlying affective evaluation rather than estimated independently: when an option feels good, it is judged both higher in benefit and lower in risk; when it feels bad, the opposite. The risk-benefit correlation is therefore diagnostic of affective rather than analytic processing.

The policy implications of this heuristic have been recognized in isolated cases (graphic warning labels, fear appeals, identifiable victim campaigns) but they have not been developed into a systematic framework for behavioral intervention design. This is surprising given what the evidence shows. Affect is not a peripheral input to decision-making that matters only in a few specialized cases; it is a pervasive, primary determinant of economic behavior across domains. It functions as information that people consult when evaluating options, as a spotlight that allocates attention and decision weights across attributes, as a motivator that energizes goal pursuit and sustains behavioral effort, and as a common currency that enables cross-domain comparisons between options that cannot be readily compared analytically (Peters, 2006). Each of these functions is a potential lever for behavioral public policy that the cognitive nudge literature has not systematically exploited.

Earlier work has of course acknowledged that emotions matter for economic behavior. Loewenstein (2000) argued for incorporating visceral factors such as anger, fear, and craving into economic models of decision-making, and Elster (1998) surveyed how emotions interact with self-interest in ways that standard utility frameworks cannot easily accommodate. Both contributions made clear that emotions are not peripheral to economic behavior, and both noted how little traction the topic had gained within mainstream economics. A quarter-century later, the observation still largely holds for behavioral public policy: emotions are recognized as consequential in specific cases, but no systematic framework has emerged for deploying affect as a primary design lever in intervention.

One reason for this neglect is that behavioral economics has, understandably, focused on correcting failures of judgment: the goal has been to help people decide as they would decide if they were fully rational and fully informed. Affective responses are often cast as departures from this rational standard rather than as decision inputs in their own right. But this framing mischaracterizes how affect works in the domains that matter most for policy. In charitable giving, in climate behavior, in preventive health investment, people's affective responses are not distorting otherwise sound deliberative judgments; they are doing most of the evaluative work, often in ways that are systematically miscalibrated to actual impact. The appropriate policy response is not to help people override their emotional responses through better deliberation, an approach that has repeatedly failed in high-salience domains, but to redesign the affective landscape of the decision so that felt reward aligns more closely with actual welfare impact. That is the project of AP.

Scope insensitivity and impact neglect

The clearest demonstration of the affective gap in behavioral public policy is scope insensitivity: the robust finding that people's emotional and behavioral responses to welfare-relevant problems do not scale with the magnitude of those problems (Slovic, 2007; Västfjäll et al., 2014). Desvousges et al. (1993) found that willingness to pay to save migratory birds was virtually identical whether 2,000, 20,000, or 200,000 birds were at risk. Slovic et al. (2017) showed that a single iconic photograph of a specific identified victim generated measurable surges in charitable response, while statistical representations of hundreds of thousands of deaths in the same crisis produced no comparable reaction. The phenomenon is stable across domains and replication efforts.

The behavioral consequence of scope insensitivity is impact neglect: choices systematically fail to track the actual welfare impact of available options (Asutay et al., 2023; 2024). In charitable giving, donations flow toward personally salient, emotionally resonant causes rather than toward organizations that produce the greatest welfare improvement per dollar of expenditure. In climate behavior, people engage in low-impact symbolic actions while avoiding higher-impact changes that would substantially reduce their actual environmental footprint. The felt reward for a low-impact symbolic action and the felt reward for an equivalent high-impact action are roughly equal, because both feel like doing something and feeling like doing something is what drives choice, not objective impact accounting.

This creates both a behavioral economics problem and an opportunity. The problem is that cognitive approaches to impact neglect have largely failed. Presenting effectiveness rankings, calculating comparative charitable returns, and providing deliberative information about impact all share a common structure: they try to supply better input to the analytical system in hopes of overriding the affective system's output. Small et al. (2007) showed this backfires directly: adding statistical impact information to an identifiable victim appeal reduced giving, because the deliberative framing attenuated the affective response that was sustaining donation. The opportunity is that scope insensitivity implies an approximate affective equivalence between low-impact and high-impact options in many relevant domains: people feel about equally good about donating to a low-impact and a high-impact charity, or about taking a symbolic vs. a substantive climate action. This equivalence means that interventions which redirect emotional engagement toward the higher-impact option do not require individuals to sacrifice felt reward. They can be at least welfare-preserving at the individual level, because the emotional experience of the behavior is not reduced, and in some cases welfare-improving where the intervention generates new affective meaning that the prior choice environment did not afford. The bundling design of Caviola and Greene (2023) is welfare-improving in precisely this sense: warm glow is preserved while a new competence glow is added. The more common case is welfare-preserving rather than welfare-improving: interventions redirect existing felt reward toward higher-impact options at constant individual hedonic cost. In both cases the social-level welfare gain is substantial because actual impact has increased. This is the ethical core of AP's strongest applications.

Affective paternalism and the cudge

AP builds on the philosophical and institutional framework of libertarian paternalism (Thaler and Sunstein, 2008) while introducing a distinct affective mechanism. Like nudging, AP is liberty-preserving: it does not restrict options, mandate behaviors, or apply financial incentives. Like nudging, it is softly paternalistic: it aims to guide behavior toward outcomes that serve individuals’ long-term interests and broader social welfare. But where nudging works by modifying the cognitive architecture of a choice (adjusting defaults, exploiting availability, changing the ordering or presentation of options), AP works by modifying its affective architecture: what the available options feel like, how emotionally compelling they are, and what felt consequences they promise.

The term affective paternalism has appeared previously with substantively different meaning. Dolan, Krekel, and Swanke (2024) use it to characterize the role of policymakers’ own affective states in biasing the paternalistic decisions they make on behalf of others, treating affect as a feature of regulator psychology. Blumenthal (2007) uses emotional paternalism to characterize interventions aimed at protecting individuals from emotional distress or affective error. AP as developed here is distinct from these prior uses in subject matter and aim. Where Dolan et al. address affect on the supply side of paternalistic decision-making and Blumenthal addresses affect as something to be protected against, AP addresses affect as something to be designed for: it centers on the affective experiences of citizens making welfare-relevant choices and treats affect as a constructive resource for behavioral intervention rather than a bias in regulators or a vulnerability requiring protection.

AP is also distinct from the affect heuristic on which it draws. The affect heuristic (Slovic et al., 2002) describes how people consult feelings as information when judging risks and benefits. It is a descriptive theory of judgment, and AP draws on this mechanism but operates at a different level: it supplies a normative and design-oriented framework for behavioral public policy, specifying when affective interventions are appropriate, what configurations of affect they should target, and what ethical constraints govern their deployment. The affect heuristic describes how people use feelings; AP develops the policy implications of that mechanism.

The operative tool of AP is the cudge, a term that combines cue and nudge to mark the mechanistic distinction. A nudge alters the structure of a choice environment to trigger cognitive shortcuts. A cudge alters the affective meaning and motivational intensity of options by embedding emotionally relevant cues in the decision context. The distinction is mechanistic, not merely descriptive: because the mechanisms differ, the behavioral trajectories they produce also differ. Choices mediated by affective reactions and made because the option genuinely felt right are more likely to be repeated and sustained than choices made because a cognitive shortcut was exploited (Schneider and van der Linden, 2023; Västfjäll, forthcoming).

The mechanistic distinction between nudges and cudges is illustrated through a number of selected examples across domains where both intervention types have been tried at policy scale or in well-identified experiments. Table 1 presents four such cases. In each, the behavioral target is held constant and the contrast is between a standard nudge approach that operates on cognitive architecture or choice frictions and a cudge that operates on the affective content of the choice environment.

Table 1. Nudge and cudge approaches to four behavioral targets.

In each case the cudge operates on the affective content of the choice environment rather than its cognitive structure or friction profile.

The pattern across the four cases is the same. Cudges work by structuring integral affect such that the felt content of the available options tracks their actual consequences and the individual's reflective values. Nudges work by structuring the cognitive or structural environment around the options so that cognitive shortcuts produce the desired selection. The two approaches are complementary rather than substitutes, but they are mechanistically distinct, and the conditions under which each will succeed differ accordingly.

The most compelling real-world demonstration of cudging at scale is the donation bundling platform developed by Caviola and Greene (2023). Rather than asking donors to override their emotional attachment to personally meaningful causes in favor of more effective alternatives, which is a cognitively demanding and emotionally costly ask, the platform offered donors the option to split their donation between their preferred charity and an expert-recommended high-impact cause. The result was a 76% increase in effective giving without any reduction in total donations or reported satisfaction, and the platform raised more than USD 1.5 million in its first fourteen months of operation. What makes this a cudge rather than a choice architecture intervention? The bundling design is in a sense a classical nudge that works by reducing the friction of choice for donors who already wanted to split and now can do so more easily. But it does not predict an effect on donors who would not have considered splitting at all. We conceive it likely that the large increase in charitable giving through bundling works for both those who wanted to split and those that did not have that preference to begin with, because it generates an affective experience (competence glow alongside warm glow) that neither the personally-meaningful-only option nor the highly-effective-only option could provide. The choice environment offers a felt content that did not previously exist. No defaults are altered, no options are removed, and no information is added; what changes is the affective texture of the giving decision and, with it, the motivational pull of the high-impact component. Cudges of this kind do not reduce friction toward a pre-existing preference; they create new felt content that reshapes what the individual prefers in the moment of choice.

Convergent evidence: the integral–incidental pattern in behavioral insights trials

The Behavioural Insights Team (BIT) and comparable government behavioral science units have conducted several hundred randomized controlled trials in policy settings since 2010, a substantial portion of which involved interventions that AP would classify as affective rather than purely cognitive. This evidence base provides a useful test of the framework. AP predicts that interventions which engage integral affect (feelings genuinely responsive to the actual properties of the decision; Västfjäll et al., 2016) will outperform those which engage incidental affect (feelings normatively unrelated to the decision), even when the surface features of the interventions appear similar.

This prediction extends a well-established distinction in affect-and-decision research. Integral affect denotes feelings that arise from the decision target itself: its actual properties, consequences, and meaning for the individual making the choice. Incidental affect denotes feelings whose source lies elsewhere, such as ambient mood, recent unrelated events, or cues introduced into the environment that have no normative bearing on the choice. The empirical signature is consistent: integral affect provides reliable input to choice because the signal tracks what is at stake, while incidental affect either fails to integrate when its source is recognized as normatively irrelevant (Pham, 1998) or distorts choice when its source is misattributed to the decision (Västfjäll et al., 2016). AP's contribution is to convert this descriptive distinction into a design principle for behavioral public policy: the framework treats integral affect as the appropriate intervention target on both efficacy grounds (integral signals carry decision-relevant information that engages reflective values) and ethical grounds (integral interventions work with the individual's actual values rather than redirecting choice through normatively unrelated cues). The empirical prediction is therefore sharper than it first appears: two interventions that look similar at the level of surface features (both deploying emotional content, both targeting the same behavior, both operating through the same channel) will produce different outcomes depending on whether the affective content stands integral or is normatively detached from the decision being made. The four BIT trials below test this pattern in policy settings.

The most direct evidence comes from the organ donation trial reported by Sallis, Harper, and Sanders (2018), one of the largest behavioral RCTs ever conducted in public policy (N = 1,085,322 UK road taxpayers). Seven messages were tested against a control prompt. The reciprocity message (“If you needed an organ transplant would you have one? If so please help others”) was the most effective, increasing registrations by 38% relative to control. Theoretically more revealing is that the condition adding a photograph of smiling people to a social norms message performed worse than the social norms message alone. Adding what was intended as an emotionally engaging cue actively reduced effectiveness. This pattern is precisely what AP's integral/incidental distinction predicts: the reciprocity message engages integral affect by making the symmetric obligation of organ donation viscerally present, whereas the smiling-person image introduces incidental positive affect that is normatively disconnected from the decision and is therefore discounted as a guide to action (Pham, 1998).

A complementary BIT trial in charitable giving (Behavioural Insights Team, 2013) illustrates the same distinction from the opposite direction. In a payroll giving experiment at HMRC, adding photographs of colleagues who were themselves donors approximately doubled enrollment rates. The same surface intervention (photograph plus prosocial framing) produced opposite outcomes in the two trials because the affective content stood in different relations to the decision: real colleagues whose own giving was being modeled in the HMRC case, generic stock-photo cues unconnected to the decision in the organ donation case. The standard nudging framework treats both interventions as “social norms plus emotional appeal” and cannot account for the divergent results.

The pattern extends beyond prosocial domains. In the court fines trial (Haynes et al., 2013), text reminders containing the recipient's name and the specific amount owed produced a 189% increase in the average amount paid relative to a generic reminder. The DVLA vehicle tax trial (Cabinet Office, 2012) demonstrated the mechanism with greater experimental control: holding channel, modality, and message length constant, notices including a photograph of the recipient's own vehicle increased payment rates six-fold relative to baseline, compared with four-fold for notices without the photograph.

Across the four trials, the interventions that succeeded most powerfully made the affective stakes integral to the specific decision being made, rather than associating generic, incidental positive affective cues.

Affect architecture and the CARE framework

AP's practical toolkit for designing affective interventions is organized around four central principles emerging from over five decades of affective science (Västfjäll et al., 2024): Create, Attenuate, Reinforce, and Eliminate. Each principle addresses a distinct configuration of the affective context of a decision, and each has a characteristic set of implementation strategies and boundary conditions. Table 2 provides a systematic overview; the text below develops each mode in relation to behavioral economics and public policy contexts.

Table 2. The CARE framework: four modes of affective intervention in behavioral public policy.

Each mode targets a distinct affective problem and operates through a distinct mechanism. Interventions may involve more than one mode; the table presents canonical applications.

Create

Many policy-relevant behaviors lack affective salience: the option is available, its benefits are understood, but nothing in the decision context generates felt motivation to act. Climate action is the paradigm case where high-impact behaviors such as shifting to plant-based diets or reducing air travel are cognitively understood but emotionally remote, and the gap between attitude and behavior is correspondingly large. Create interventions build affective associations calibrated to the actual consequences of the option. This can be achieved through narratives that make abstract consequences personally vivid, through imagery that evokes anticipatory emotional responses, or through reframing that connects the option to identities and values the individual already endorses. Karlsson et al. (2024) showed that prompting individuals to engage in vivid mental imagery of local climate consequences generated meaningful anticipatory concern and substantially increased intentions to adopt high-impact behaviors, without distorting the factual content of what participants knew. The affective response created was integral, meaning it was therefore genuinely about the option and its consequences, rather than incidental (Västfjäll et al., 2016). The boundary with Reinforce (below) hinges on baseline affect: Create applies where the affective response to the option is at or near zero, Reinforce applies where a measurable baseline response exists and the design target is its amplification or redirection.

Attenuate

Some behavioral patterns are sustained by affective responses that are disproportionate to the actual properties of the option, in either direction. Attenuate interventions reduce affective intensity where the existing signal is miscalibrated to the option's actual consequences, regardless of valence. The central question is not whether the affect to be attenuated is positive or negative but whether its intensity tracks the real properties of the choice. Two classes of application follow. First, where negative affect is disproportionately high and generates defensive avoidance or paralysis, attenuation reduces the maladaptive response while preserving the informative signal. Pairing high-threat health information with credible, concrete efficacy information attenuates avoidance without eliminating warranted concern (Witte and Allen, 2000). Second, where positive affect is disproportionately high and sustains behavior that is poorly aligned with welfare, attenuation reduces the felt reward that drives the misaligned behavior. The diffuse positive affect attached to low-impact symbolic climate action is a behavioral barrier of this kind: when individuals derive substantial felt reward from actions whose actual environmental impact is small, the felt equivalence between symbolic and substantive options can substitute affective closure for behavioral effect. Weakening this felt equivalence (rather than asserting it deliberatively) is a prerequisite for Create interventions targeting higher-impact behaviors to take effect. Attenuation is graded: it reduces the intensity of an affective signal while preserving its informational content where that content carries residual functional value.

Reinforce

Many beneficial behaviors already generate some positive affective response, but the response is either too weak to sustain effort over time or insufficiently connected to the highest-impact version of the behavior. Warm glow in charitable giving is the canonical example: people already feel good when they donate, and the Reinforce mode asks how this existing affective reward can be amplified and redirected toward more effective giving. The donation bundling evidence of Caviola and Greene (2023) demonstrates this logic directly: warm glow was not replaced or reduced but supplemented with a new source of positive affect, creating a felt experience that neither option alone could generate. Social feedback that makes the cumulative impact of an individual's climate choices personally vivid represents another Reinforce approach: the positive affective identity of the environmentally engaged person, already present for many individuals, is connected more concretely to specific high-impact behaviors. The key welfare property of Reinforce interventions is that they augment rather than substitute felt reward, making them among the most ethically tractable of the four modes.

Eliminate

Where Attenuate reduces an affective signal that retains some functional value, the Eliminate mode targets affective responses whose miscalibration is demonstrable and whose residual functional value is minimal or absent. The most demanding CARE mode targets the disruption of specific affective responses that are demonstrably miscalibrated, ones that reliably drive behavior away from welfare in ways that are not tracking the actual properties of the options being evaluated. Graphic tobacco warning labels represent the clearest real-world example: they are explicitly designed to attenuate and partially eliminate the positive affect (i.e., the glamour, social signaling, and immediate pleasure) that sustains smoking, by making the actual health consequences of the behavior viscerally and emotionally present at the point of consumption. Evans et al. (2015) showed in a randomized clinical trial that graphic warning labels elicit both affective and deliberative responses from smokers that text-only warnings do not, with the affective response (the felt aversiveness of the smoking cue) driving the behavioral effect. The intervention changes not what smokers know about tobacco risk (that information is already available) but how the act of smoking feels in the moment of decision. The Eliminate mode is ethically the most demanding, because it involves the deliberate disruption of an affective response that may retain some functional value even when it is miscalibrated. Interventions should target the specific maladaptive expression of the response rather than the underlying motivational system, and should be calibrated to the actual degree of miscalibration rather than applied wholesale.

Ethical considerations and the welfare case for AP

Affective interventions raise ethical concerns that are somewhat distinct from those raised by cognitive nudges, and the behavioral economics literature has not yet fully addressed them. The most fundamental is the question of manipulation: if a policy instrument works by shaping how options feel rather than by providing better information or removing friction, does it cross from legitimate guidance into illegitimate emotional exploitation?

AP's answer turns on the distinction between integral and incidental affect as intervention targets (Västfjäll et al., 2016). Integral affect, meaning feelings that are genuinely about the object of a decision and tracking its actual consequences and alignment with the individual's own goals, is the ethically preferable target because intervening to strengthen or redirect it brings the emotional signal into closer alignment with the real features of the choice. The warm glow of effective giving is not manufactured or misattributed; it is a genuine response to a perceived moral achievement, made accessible by a choice environment that allows it to be experienced. This contrasts with commercial advertising's characteristic use of incidental affect and arbitrary positive associations attached to products through aspirational imagery normatively unrelated to the product's actual properties. The distinction tracks a meaningful difference in the welfare properties of the intervention. Integral affect interventions work with the individual's actual values and goals; incidental affect manipulation often (but not always) works against them.

A further welfare grounding for AP comes from the criterion of asymmetric paternalism developed by Camerer et al. (2003b): a regulation is asymmetrically paternalistic if it creates large benefits for those who make errors while imposing little or no harm on those who behave rationally. This criterion provides a welfare-economic foundation for AP's strongest applications that is both rigorous and politically tractable. In scope-insensitive domains, affective equivalence between low-impact and high-impact options means that an individual who already directs their emotional engagement toward high-impact behavior gains nothing from a Create cudge (the intervention has zero cost to them), while the individual whose felt responses are miscalibrated to impact gains substantially. The asymmetry holds even more cleanly for Reinforce interventions: amplifying warm glow that is already present for an individual whose giving is well-directed leaves their experience unchanged, while generating a new competence glow for the individual whose giving is concentrated in lower-impact options produces both a new affective experience and a welfare gain. Camerer et al.'s criterion also clarifies the limits of the Eliminate mode: the Eliminate mode is most justifiable when the targeted affective response is demonstrably and specifically miscalibrated (as in the case of the glamour affect sustaining tobacco use, which arguably tracks no genuine benefit of the behavior) and least justifiable when the affective response carries residual functional value alongside its maladaptive expression. AP thus offers a theoretically grounded, welfare-economic extension of the asymmetric paternalism agenda: it identifies a class of affective interventions that satisfy the criterion's requirements while engaging mechanisms that the cognitive nudge toolkit cannot reach. The application of AP to professional advisory contexts, including medical shared decision-making, further illustrates both the promise and the ethical complexity of affect-informed intervention design (Tinghög et al., 2024).

A further welfare consideration specific to AP's strongest applications concerns the approximate affective equivalence between low-impact and high-impact options in scope-insensitive domains. When felt reward is approximately equal across these alternatives, as documented extensively in the giving and climate literatures, affective interventions that redirect engagement toward high-impact options do not reduce individual well-being. The individual experiences similar or greater felt reward while producing substantially greater social welfare impact. This is a rare property of behavioral interventions: it means that AP in scope-insensitive contexts can be welfare-improving at both the individual and social levels simultaneously, without the trade-offs between individual autonomy and social benefit that characterize many other paternalistic approaches. Legitimacy conditions drawn from the broader nudging ethics literature, including transparency of intent, proportionality to the behavioral objective, and alignment with individuals’ considered values (Hagman et al., 2015; Lades and Delaney, 2022), apply to AP and provide additional constraints that distinguish acceptable from unacceptable uses of Affect Architecture.

A final complication concerns the temporal structure of affective experience. Affective responses to a policy or behavior are not fixed quantities that can be measured once and treated as stable inputs to welfare analysis. They vary systematically across at least three distinct temporal phases: the affect people anticipate before an intervention (anticipated affect), the affect they actually experience during it (experienced affect), and the affect they remember afterward (remembered affect). These three are often poorly correlated, and each is subject to characteristic distortions. Anticipated affect is shaped by affective forecasting errors: people systematically overpredict the negative emotional impact of behavioral changes and underpredict their capacity to adapt (Loewenstein et al., 2003; Wilson and Gilbert, 2003). Experienced affect, while more accurate as a moment-by-moment hedonic record, is largely inaccessible to subsequent decision-making. Remembered affect is reconstructed at recall according to the peak-end rule: retrospective evaluations are dominated by the most intense moment and the final moment of an episode, with the duration and the overall hedonic integral receiving negligible weight (Fredrickson and Kahneman, 1993; Kahneman et al., 1993). This three-way divergence creates what might be termed a temporal paradox for affect-informed policy: should behavioral public policy target the anticipated affect that determines initial uptake, the experienced affect that constitutes the actual hedonic reality of the intervention, or the remembered affect that drives subsequent behavioral repetition and political support? Each criterion implies a different intervention design. AP does not resolve this paradox, but it idientifies it as a central design challenge for affect architecture and calls for empirical data on all three temporal phases to be collected before, during, and after behavioral interventions.

Predictions and future research

AP generates a set of testable predictions for behavioral economics research that are distinct from those derivable from the cognitive nudge literature. We highlight four.

First, in domains characterized by scope insensitivity, Create and Reinforce interventions should produce welfare gains at the individual level that are absent or negative for information-based or cognitive interventions. The donation bundling evidence is consistent with this prediction, but systematic experimental comparison of affective vs. cognitive approaches across domains (giving, climate, health) is needed to establish boundary conditions and effect sizes. The prediction is that cognitive approaches will underperform in domains where affective equivalence between options is strongest.

Second, Attenuation of low-impact affective substitutes should be a prerequisite for effective Create interventions in domains where symbolic action provides affective motivations for a behavioral goal. The felt equivalence of recycling and dietary change is a testable prediction: if this equivalence is attenuated experimentally, Create interventions targeting high-impact behaviors should show larger effects than when applied in isolation. This sequencing prediction has direct design implications for behavioral public policy campaigns.

Third, the welfare properties of affective interventions should vary predictably with the degree of affective equivalence between options. In domains where low-impact and high-impact options differ substantially in felt reward, where the high-impact option is genuinely more effortful or less pleasant, affective interventions face a harder task and may require welfare trade-offs. AP's ethical case is strongest where affective equivalence is documented; identifying these domains empirically is therefore a research priority.

Fourth, the durability of behavior change should favor affective over cognitive interventions where the mechanism of change involves genuine motivational engagement rather than exploitation of cognitive shortcuts. Schneider and van der Linden 2023; (see also Västfjäll, 2026) suggest this on theoretical grounds, but longitudinal experimental evidence is sparse. Establishing whether cudge-mediated behavior change is more durable than nudge-mediated behavior change, and under what conditions, is one of the most consequential open questions for future research to address.

Conclusion

Behavioral economics has given policymakers a powerful set of tools for addressing cognitive barriers to welfare-enhancing behavior. The field's next major extension is affective: developing a theoretically grounded, empirically tested, and ethically defensible toolkit for engaging the emotional dimension of decision-making in the service of high-impact behavior change. AP and Affect Architecture offer a starting point for this extension, grounded in many decades of affective science research and demonstrated at scale in the domain of charitable giving.

The theoretical contribution of this paper is not to displace cognitive approaches to behavioral public policy but to identify the class of problems for which they are insufficient and to specify what an affective approach requires. Scope insensitivity and impact neglect define this class: these are domains in which the affective context of choice is systematically misaligned with welfare impact, in which cognitive interventions have repeatedly underperformed, and in which affective interventions hold the prospect of large welfare gains at negligible individual cost. AP creates large benefits for individuals whose felt responses are miscalibrated to impact, while imposing no cost on individuals whose affective engagement with high-impact options is already well-calibrated. The cudge, affect architecture deployed in the service of liberty-preserving, welfare-enhancing behavior change, is the instrument that behavioral public policy needs for these domains.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. In preparing this manuscript, the authors made use of Claude (Anthropic, claude.ai), a large language model, to assist with language editing in sections of the manuscript. All content generated with AI assistance was reviewed, verified, and substantially revised by the authors. The authors take full responsibility for the accuracy of all claims, citations, and references.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Lory Barile, University of Warwick, United Kingdom

Reviewed by: Jana Sadeh, University of Southampton, United Kingdom

Manuele Citi, Copenhagen Business School, Denmark