A heart races, the stomach contracts, breathing changes and the face grows warm. The experience may be named fear, anger, anticipation, embarrassment or excitement.
The bodily changes do not arrive with labels attached.
That gap between internal physiology and emotional meaning sits at the centre of Interoceptive Mechanisms and Emotional Processing, the Annual Review of Psychology article by Benedict M. Greenwood and Sarah N. Garfinkel. Published in volume 76 in 2025 after appearing online as a Review in Advance in October 2024, the paper synthesizes a field that has expanded rapidly but often uses the word interoception for different levels of the same chain.
The review is important because it does not ask only whether bodily signals matter for emotion.
It asks which signals, which dimensions of perception, which neural processes, and which forms of interpretation are involved.
What kind of source this is
This is a peer-reviewed narrative review in Annual Review of Psychology.
It is not a single experiment, randomized trial or meta-analysis. Its evidentiary value comes from integration: the authors organize results across physiology, neuroscience, behavioral tasks and clinical research.
That source class matters.
A review can reveal convergence, conceptual gaps and recurring methodological problems. It cannot make a small or unreplicated underlying literature stronger simply by summarizing it.
Greenwood and Garfinkel are unusually explicit about that boundary. Their conclusion notes that some clinical findings rest on small numbers of studies, limited samples or effects that have not yet been replicated. They call for preregistration, open-science practice, replication and richer interoceptive measurement.
The review should therefore be read as both a map and a warning against premature certainty.
The paper's central architecture
The review treats interoception as multidimensional.
One level concerns the nature of afferent bodily signals. Internal organs and physiological systems generate information through cardiac, respiratory, gastric and other pathways.
A second level concerns how accurately those signals are detected or discriminated.
A third concerns neural processing: how internal information is represented, integrated and weighted in relation to other inputs.
A fourth concerns higher-order interpretation: beliefs, attention, appraisal and context.
Emotion can be influenced at every level.
This architecture prevents a common simplification in which "interoception" is treated as one trait that is either high or low.
Bodily signals do not determine emotion by themselves
The review is situated in a long tradition linking bodily change and emotional experience, but it does not reduce emotion to visceral input.
A physiological change becomes emotionally meaningful within a system that includes prediction, attention, prior learning, external context and conceptual interpretation.
The same cardiovascular arousal can accompany multiple emotions. The same emotion can occur with different physiological patterns. Internal signals influence emotional processing without functioning as one-to-one labels.
That is a crucial evidence boundary.
Interoceptive contribution is not physiological determinism.
Accuracy is not enough
Popular descriptions often assume that better detection of bodily signals should produce better emotional functioning.
The review makes the problem more complex.
Accurate sensing may be useful when the signal is relevant and interpreted well. But intense attention to internal sensations can also become maladaptive, particularly when it is coupled with threat expectations or catastrophic interpretation. A person can detect a bodily change and misattribute its cause. Confidence can be poorly calibrated. One interoceptive domain may be relatively precise while another is not.
The adaptive pattern may therefore involve several capacities at once: sufficient signal access, flexible attention, calibration, and contextual interpretation.
More is not automatically better.
Clinical literatures reveal different failure modes
Greenwood and Garfinkel review evidence across anxiety, depression, schizophrenia, post-traumatic stress disorder, emotionally unstable personality disorder and autism.
The value of placing these literatures side by side is not to propose one interoceptive deficit underlying all diagnoses.
It is almost the opposite.
Different conditions may involve different combinations of bodily signal processing, attention, belief, precision, confidence and interpretation. Anxiety may involve heightened attention to threat-relevant bodily changes without globally superior accuracy. Depression can involve altered bodily experience and motivational state. Autism research has produced heterogeneous results depending on task and construct. Schizophrenia and trauma-related conditions raise different questions about self, salience and prediction.
The diagnostic label does not tell us which interoceptive mechanism is altered.
That is why mechanism-specific measurement matters.
Measurement is one of the review's most important messages
The review pushes against reliance on a single task.
Heartbeat counting, confidence ratings, self-report questionnaires, neural measures and organ-specific perturbations do not answer the same question. A result in cardiac interoception should not silently become a result about the entire internal body.
This is especially important in clinical science, where a mechanistic claim can become an intervention target.
If a study says "interoception is impaired" but used one contested heartbeat task, the treatment implication is weak. If a study identifies a reproducible problem in a specific domain and shows that changing it alters a clinical outcome, the claim becomes much stronger.
The paper therefore supports a move toward varied assessments across bodily axes rather than one proxy repeated everywhere.
The review and predictive accounts
Contemporary interoception research is strongly influenced by predictive-processing and active-inference frameworks. These models describe perception as an interaction between sensory evidence and expectations or priors, often weighted by estimates of precision.
The review uses these ideas to help organize emotional and clinical findings.
But a framework is not an empirical result by itself.
Predictive language can become so flexible that almost any outcome is redescribed after the fact. The useful version specifies what prediction, signal or precision term would differ, how it would be measured and what competing model would predict instead.
The review's emphasis on multidimensional measurement is therefore essential to keeping theory testable.
Treatment implications are possibilities, not established prescriptions
The review discusses the possibility that interventions could target interoceptive mechanisms. This can include exposure to internal sensations, attention training, biofeedback, contemplative practices or methods that alter interpretation.
The important word is mechanisms.
A treatment should not be justified merely because it is "body based." It should specify what process is expected to change and how that change will be measured.
Clinical evidence is uneven across conditions and intervention types. Some findings are promising, some are correlational, and some lack replication.
The review does not license a general claim that interoceptive training treats mental illness.
What the review contributes to human-capacity research
The paper offers a strong conceptual lesson beyond clinical science.
A useful human capacity is rarely one variable.
Consider the capacity to notice rising stress during a difficult meeting. It may depend on detecting a respiratory or cardiac change, directing attention without being captured by it, distinguishing arousal from danger, recalling contextual information and choosing an action consistent with the actual situation.
Calling all of this "body awareness" hides the architecture.
The review suggests that research and practice should preserve the chain from signal to perception to interpretation to action.
That is directly relevant to education, leadership and human-AI interaction, where people are increasingly asked to make judgments under uncertainty while external systems accelerate the pace of response.
A review is also an argument about categories
One of the less visible contributions of a major review is taxonomic. By placing accuracy, sensibility, neural processing and higher-order interpretation beside one another, the authors implicitly reject papers that use one word for all of them. That matters because literatures can appear inconsistent when they are actually measuring different layers.
A clinical study using self-report can be informative about attention or belief without contradicting a performance study that finds no sensory-accuracy difference. A neural result can show altered processing without telling us whether a patient experiences the signal consciously. The review helps make these apparent contradictions legible.
Clinical translation needs a causal bridge
A mechanism is clinically useful only if changing it changes an outcome that matters. Correlation between an interoceptive measure and symptom severity is the beginning of a causal question, not the end. The stronger translational chain would show that the mechanism precedes change, can be manipulated, and mediates improvement in a replicated design.
Many interoception findings have not yet reached that standard. This is why the authors' call for replication and better measurement is not a procedural footnote. It determines whether a proposed mechanism is ready to become a treatment target.
Multiple bodily axes may reveal different clinical stories
Cardiac tasks dominate parts of the literature because heartbeats are easy to record. Yet respiratory, gastric and other internal systems may be more relevant to particular symptoms. Panic can involve breathing sensations; eating-related phenomena may require gastric or metabolic measures; chronic pain raises still other channels.
A future clinical science of interoception will probably look less like one general test and more like a profile across modalities and operations. The review points toward that architecture without claiming it has already been standardized.
Precision language needs operational definitions
Terms such as prediction error and precision can sound mathematically exact even when a study does not measure them directly. A review can legitimately use them as organizing concepts, but individual empirical papers should specify how the proposed computational quantity is operationalized.
For example, confidence, neural gain, variance and attentional weighting are related to precision in different models but are not interchangeable measurements of one hidden variable. The field will advance when computational vocabulary narrows hypotheses rather than merely modernizing descriptions.
Development and lifespan are still underrepresented
Interoceptive processes change with physiology, learning and age. Childhood, adolescence, pregnancy, menopause, cardiovascular aging and chronic illness can alter both the bodily signal and the opportunity to learn its meaning. Yet much mechanistic research relies on restricted adult samples.
A mature emotional-interoception science therefore needs lifespan designs and more diverse physiological contexts. Without them, a model calibrated on young healthy adults may be mistaken for a general architecture of human emotional processing.
Culture changes interpretation without making physiology arbitrary
People learn different vocabularies and norms for bodily feeling. Cultures vary in whether attention to internal state is encouraged, medicalized, moralized or ignored. These differences can shape reporting and appraisal even when underlying physiological pathways are shared.
The review's multidimensional framework is compatible with this complexity. Culture can influence the interpretive layers without implying that biology is infinitely flexible. The empirical task is to identify which layer varies, not to choose between "biological" and "cultural" explanations as if only one can be true.
Institute synthesis: interoceptive literacy is relational
The Institute uses interoceptive literacy to mean the capacity to work responsibly with internal bodily information by distinguishing signal, attention, interpretation, confidence and context.
This is Institute synthesis. It is not terminology introduced by Greenwood and Garfinkel and should not be presented as scientific consensus.
The review provides evidence for why such distinctions matter. The Institute extends those distinctions into a capacity framework.
In this framing, the goal is not maximal sensitivity.
It is a more calibrated relationship with internal evidence.
Institutional implications
Organizations increasingly use wearable and affective data to infer stress, engagement or readiness.
The review's multidimensional account should make such inferences more cautious, not more ambitious.
A heart-rate pattern is not an emotion label. A self-report is not objective accuracy. A neural signal is not a person's meaning. Group-level associations do not automatically support individual prediction.
If institutions use biosensing, they need construct validity, consent, data minimization and clear limits on interpretation.
The scientific complexity of interoception is an argument against physiological surveillance masquerading as self-knowledge.
What the evidence does not show
The review does not show that there is one interoceptive deficit shared across psychiatric diagnoses. It does not show that stronger bodily awareness is always adaptive. It does not establish that heartbeat accuracy represents all interoception. It does not demonstrate that predictive-processing models are uniquely correct. It does not show that interoceptive interventions are established treatments across the disorders discussed.
It also does not convert associations between internal sensing and emotion into a diagnostic tool for individuals.
The authors' own conclusion emphasizes limited samples, sparse literatures and the need for replication in several areas.
That restraint is part of the paper's contribution.
Why this paper belongs in the canonical library
A canonical knowledge library needs more than definitions. It needs field-level sources that show how concepts are actually assembled and where evidence is thin.
Greenwood and Garfinkel's review does this for interoception and emotion.
It connects bodily afference, perceptual dimensions, neural processing, interpretation and clinical research while refusing to pretend that all parts of the map are equally mature.
For the Institute, its deepest value is methodological.
The paper makes it difficult to speak responsibly about "listening to the body" without asking what is being listened to, how it is measured, and what turns a sensation into a meaning.
Research references
- Greenwood BM, Garfinkel SN. Interoceptive Mechanisms and Emotional Processing. Annual Review of Psychology. 2025;76:59-86. DOI: 10.1146/annurev-psych-020924-125202. Review in Advance published 18 October 2024. https://www.annualreviews.org/content/journals/10.1146/annurev-psych-020924-125202
- Garfinkel SN et al. Knowing your own heart: Distinguishing interoceptive accuracy from interoceptive awareness. Biological Psychology. 2015;104:65-74. PMID: 25451381.
- Khalsa SS et al. Interoception and Mental Health: A Roadmap. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging. 2018;3(6):501-513. DOI: 10.1016/j.bpsc.2017.12.004.