Interoception: The Body Signals That Shape Thinking Before You Notice Them

Introduction

In a laboratory in Brighton, people sat in front of a screen and read words.

The words were ordinary. The timing was not. Each word appeared either at the exact moment the heart contracted, or in the gap between beats. The participants had no idea this was happening. As far as they could tell, words were simply appearing.

Later they were asked which words they remembered.

The words that had arrived on the beat were remembered worse [1].

That result is well over a decade old now, it has been cited hundreds of times, and it turns up in almost every popular piece about the mysterious sixth sense that lets you feel your own body from the inside. You may have read a version of it. What those pieces almost never mention is the number of people it was measured in.

Seventeen.

Seventeen participants finished that experiment. Two were dropped for equipment failure and an elevated heart rate. Three more did not complete the memory test. What is left is a careful, clever, genuinely interesting study with a final analysed sample of seventeen human beings.

Hold both of those thoughts at once, because the rest of this article lives in the space between them. The phenomenon is real. Later and larger work has found the same rhythm showing up in reward learning and in how well you can ignore a distraction. And the field that studies it is in the middle of an argument about whether its main measuring instrument measures anything at all.

This is not a debunking. Interoception is one of the most interesting things your nervous system does, and the evidence that internal bodily signals shape thinking is stronger than most people realise. But the honest version of the story is more useful than the tidy one, and nobody seems to be telling it. So here is what your body is actually doing to your mind, how much weight each claim will bear, and why three studies can all be about interoception and still be measuring three different things.

The Sense You Were Never Taught

You learned five senses at school. Sight, hearing, smell, taste, touch. All five point outward. They tell you about the world.

Interoception points the other way.

It is the sense of the internal state of your body: the pressure in your arteries, the stretch of your stomach, the carbon dioxide building in your blood, the temperature of your skin from the inside, the state of your bladder, the inflammation in your tissue. Some of it reaches awareness. Most of it never does. Your body is being measured, continuously, by a system you have no memory of installing.

The word is old. Charles Sherrington coined "interoceptive" in 1906, in The Integrative Action of the Nervous System, published by Yale University Press. He was drawing a boundary: exteroceptors face the world, proprioceptors report on limb position, interoceptors report on the viscera. It was a filing system more than a theory, and for most of the twentieth century that is what it stayed.

The modern field starts with a reframing. In 2002, Bud Craig argued that interoception was not a dusty category of visceral reflexes but a genuine sense with its own pathway, its own cortical destination, and a direct role in how feelings are made [2]. He followed it with a shorter companion piece the next year [3]. That reframing is why anyone is writing articles like this one.

The scope has widened since. A 2021 consensus statement written by thirteen authors out of a National Institutes of Health workshop defines interoception broadly, covering signals from the cardiovascular, respiratory, gastrointestinal, urogenital, immune and thermoregulatory systems, and it is careful to note that most of this traffic never becomes conscious at all [4].

That last point gets lost constantly. When people say "interoception" in casual conversation they almost always mean the small conscious part: noticing you are hungry, noticing your heart is racing. The system is much larger than the part you can feel. Most of it is doing regulation, not reporting.

You will sometimes see interoception called the sixth sense. It is a useful hook and it is wrong in two ways. It is not one sense, it is a family of channels that turn out not to correlate with each other very well. And proprioception has an equal claim to the title. Use the phrase to find the topic, then drop it.

How a Heartbeat Becomes Information

Here is the part that makes the memory result make sense.

Your arteries contain stretch receptors called baroreceptors. When the heart contracts and blood is forced out, the arterial wall stretches and those receptors fire. Between beats, when the pressure drops, they go quiet. This means the information your brain receives about your heart is not a steady stream. It arrives in pulses, once per beat, with silence in between.

That is the whole basis of the systole and diastole experiments. Systole is the contraction, when the signal is loud. Diastole is the gap, when it is quiet. If you time a stimulus to arrive during one or the other, you are effectively asking what happens to a piece of information that lands while the brain is being interrupted by its own body.

The signals travel by more than one road, and the roads run at very different speeds. Vagal fibres are the fast lane, carrying information from the heart, lungs and gut in something close to real time. Spinal pathways handle a different class of visceral information and take their own route upward. Hormones and immune molecules are slower again, sometimes by minutes, because they are not nerve signals at all but chemistry moving through blood.

That spread of timescales is easy to skim past and it matters. It means your brain is never receiving one coherent bulletin about the state of your body. It is receiving several, arriving at different times, describing the same body at slightly different moments. A 2021 review of the circuitry sets out how these converge [5], and it is a better guide to current thinking than the older single-pathway story.

Most of the traffic in that diagram takes the orange path. It never becomes a feeling. It adjusts your heart rate, your breathing depth, your blood distribution, and you are told nothing about it.

The destination everyone talks about is the insula, a slab of cortex folded away inside the lateral sulcus. Craig's 2009 paper made it famous by proposing that the anterior insula is where bodily states get turned into subjective feelings, and where the self that has those feelings is assembled [6]. It is one of the most cited papers in the whole field.

It is also a simplification, and the correction matters. The insula is not a single functional thing.

Pool together the imaging studies that lit up some part of it and you do not get one region doing one job. You get clusters, with different connectivity and different roles [7]. A 2024 review starting from cells and circuits rather than from scanner images arrives at the same place from the other direction [8]. When two methods with completely different weaknesses agree, you can usually believe them.

What those subregions seem to be doing is integration rather than pure reception. Body signals arrive alongside emotional and external information, and the insula is where they are combined into something usable [9]. That is why anterior insula activity shows up both when people attend to their bodies and when they simply feel something [10]. The overlap is probably not sloppy measurement. It hints that the two are less separate than our vocabulary suggests.

If you want a rough analogy, think of how unhelpful it would be to say that a kitchen is where food happens. True, uninformative, and it hides every question you actually wanted answered about which part does what.

Hold on to one dissociation inside it, because it comes back later in this article. Paying attention to your body and being accurate about your body are not the same operation, and they do not depend on the same insular tissue [11]. Attention and accuracy come apart in the anatomy before they come apart in the psychology.

So when you read that the insula is your body's control room, treat it the way you would treat "the frontal lobe does planning". Directionally true. Structurally lazy.

There is one result in this section that does not fit the neat story, and it belongs here rather than in a footnote. If bodily feedback drives your sense of effort, then blocking that feedback should change how hard exercise feels. When that was tested, the perception of effort turned out to be largely independent of afferent feedback from muscles, heart and lungs [12]. The brain appears to generate much of that feeling from its own outgoing motor commands rather than from incoming reports. Keep that in your pocket. It is a useful corrective to the idea that the body simply tells the brain what is going on.

How the Field Got Here

The chronology matters, because the argument in this article is a chronological one. A concept gets defined, a measuring tool gets adopted because it is easy, a literature grows on top of the tool, and then somebody checks the tool.

Notice the shape. The big theoretical moves land between 2002 and 2015. The methodological reckoning lands after. That ordering is the reason so much of the popular coverage is confidently out of date: it is drawing on the exciting middle period and has not caught up with what happened next.

It is also a completely ordinary way for a science to develop, and reading it as a scandal gets it wrong. A field needs a measure before it can have findings. The first measure available is usually the cheapest one that works at all, because you cannot wait for a perfect instrument to start asking questions. Findings accumulate. The measure gets used far outside the conditions it was designed for. Eventually somebody with a large sample and a suspicious disposition checks whether the instrument does what everyone assumed.

What separates a healthy field from an unhealthy one is what happens after that check, not whether it happens at all. Here you get a critique with hundreds of participants, a technical rebuttal from the researchers whose work is implicated, a review sorting out which parts of the criticism stick, and a new measure built to fix the specific defect. That sequence took four years and it is all public. You can read every step of it.

What the Heart Forgets

Back to the experiment from the introduction, properly this time.

Words were presented under limited attentional resources, timed either to systole or to diastole. Memory for the systole words was worse. The effect was not uniform across people: it was attenuated in participants who scored higher on interoceptive sensitivity, and the people who scored low were more dependent on having detected the word with high confidence in order to remember it at all [1].

Read that second part again, because it is the interesting half and it gets dropped in every retelling. The finding was not simply "systole is bad for memory". It was that the cost of systole depended on how sensitive to your own heartbeat you were. The body was not doing the same thing to everybody.

Why would a heartbeat interfere with encoding a word? The leading explanation is that the burst of baroreceptor activity at systole is itself information competing for the same processing. There is a well-documented dampening of cortical excitability that follows the pulse. A word arriving in that window has to share.

But the story is not uniformly one of interference, and this is where it gets strange. When the same group timed fearful faces to systole rather than words, the fear stimuli were processed more intensely, not less [13]. Threat gets amplified on the beat. Neutral words get suppressed. The heartbeat is not a blanket dampener, it is a filter with a preference, and what it prefers looks a lot like danger.

Cardiac timing also reaches into emotional learning. A study of twenty-nine participants found that emotional learning and memory were shaped by cardiac cycle, by interoceptive accuracy and by personality together, rather than by cardiac timing alone [14]. And in a related line of work, feedback from the body was shown to shape recognition memory judgments, meaning the confidence you place in a memory is partly a bodily readout rather than a purely cognitive one [15].

Follow that thought and it leads somewhere useful. If your bodily state is part of the context in which a memory is laid down, then arriving in a different state should make that memory harder to reach, which is precisely the effect known as state-dependent memory. Run the same logic on feelings rather than physiology and you get mood-congruent memory, where a low mood preferentially serves up low memories. Both are usually explained in psychological terms. Underneath, they are interoception doing the work.

The Beat Reaches Further Than Memory

If the cardiac cycle only affected word recall, it would be a curiosity. It does not.

In 2024 a study of thirty-two participants, after three exclusions for noisy recordings, found that timing along the cardiac cycle modulates the neural signals of reward-based learning [16]. The brain's response to whether an outcome was better or worse than expected, the prediction error that drives learning from reward, was shaped by where in the heartbeat that outcome landed.

In the same year, a study of forty participants looked at inhibition rather than learning [17]. Participants performed a stop-signal task while distracting information was timed to different cardiac phases. Distractors that arrived at systole were suppressed more effectively than distractors arriving at diastole. Your ability to ignore something depended, slightly, on when in your pulse it appeared.

Two studies is not a literature, and neither of those effects is large. But notice what they have in common with the memory result. Three different research groups, asking three different questions, using three different tasks, all found the answer changing depending on where in the pulse the stimulus landed. Nobody set out to find a general principle. It keeps turning up anyway, which is a better sign than a single dramatic finding would be.

The honest framing is that your heartbeat appears to be one of the rhythms your cognition is running against, along with breathing and the various cortical oscillations that are far better studied. It is not the main one. It was simply overlooked for longer, because the body was assumed to be the thing the brain was thinking about rather than part of the machinery doing the thinking.

Attention shows the same fingerprint from a different angle. Work using both scalp and intracranial recordings has separated the neural signatures of attending inward to the body from attending outward to the world [18], and a 2026 study identified beta-band suppression as a marker of heartbeat processing specifically [19]. These are the tools that will settle a lot of the arguments in this article over the next decade.

Then there is decision-making, where the popular story is at its most misleading.

The famous framing is the somatic marker idea: your body flags the risky option before your conscious mind works it out, and people with better access to those signals make better choices. When this was tested directly, the result was more honest and less marketable. Better interoceptive accuracy did not straightforwardly improve intuitive decision-making. It amplified whatever the bodily signal was pointing at. When the anticipatory signal favoured the advantageous option, good interoceptors did better. When it favoured the disadvantageous one, good interoceptors did worse [20].

Sit with that. The body is not a wise adviser. It is a loud one. Being better at hearing it helps only when it happens to be right.

A later study found interoceptive sensibility tuned risk-taking behaviour only when body-related stimuli were involved in the task at all [21], which further narrows the claim. And imaging work has shown the insula tracking how prior risk experiences shape later risky choice [22]. The mechanism is real. The self-help version of it is not supported.

There is a general version of this point. Your brain is constantly deciding which of the thousands of things reaching it are worth keeping, and bodily arousal is one of the tags it uses to make that call, which is a large part of how the brain chooses what to remember. The cardiac timing effects are that same selection process, caught operating on a timescale of milliseconds instead of minutes.

Three People, Three Different Kinds of Good

Now the idea that makes the rest of the literature readable.

Suppose you wanted to know whether someone is good at interoception. You could test them: sit them down, have them report their heartbeats, compare against an actual recording. That gives you a performance score.

Or you could ask them. Do you notice your body a lot? Are you in tune with your physical state? That gives you a self-report.

Or you could measure something subtler: when they are confident they are right, are they actually right? That gives you a measure of insight into their own accuracy.

In 2015, Sarah Garfinkel and colleagues ran all three in the same sample of eighty people and showed that they come apart [23]. They named them interoceptive accuracy, interoceptive sensibility and interoceptive awareness. The three dimensions were dissociable. They only corresponded meaningfully within the subgroup of people who scored highest on accuracy.

This is the single most useful thing in the field and it is almost entirely absent from popular coverage. Once you have it, contradictions dissolve.

Take anxiety. Some studies say anxious people are hyper-attuned to their bodies. Others say they are poor at reading them. Both can be true, because they are measuring different dimensions. An anxious person can score very high on sensibility, they are certainly paying attention, while scoring poorly on accuracy. Research on anxiety and depression found the two conditions pulling interoceptive accuracy in opposite directions [24], and further work showed accuracy behaving as an interactive function of anxiety-specific and depression-specific symptoms rather than tracking distress in general [25].

More recent work keeps finding the same split, and it keeps finding new places to put it.

Consider a question that sounds like it has an obvious answer. If you pay a lot of attention to your body, are you more accurate about it? Intuition says yes. Measurement says the two are independent [26]. A separate line of research built a dedicated scale for interoceptive attention and found the same dissociation waiting for it [27].

That is a strange result. Attention is usually how you get better at a perceptual task. Watch closely and you see more. Here, watching closely does not appear to make you right.

Some researchers have concluded that the problem is partly linguistic. The labels themselves have been used inconsistently enough across studies that papers can appear to disagree when they are simply naming different things [28]. Others have gone further inside self-report and argued that even sensibility needs splitting, because noticing your body and regulating your body through that noticing are separate styles [29].

You can see where this is going. Every time somebody looks closely at a dimension of interoception, it turns into two dimensions.

There is a second dissolution hiding here, and it is even less well known. People talk about interoception as a single talent, as though someone could be an interoceptive athlete. When cardiac awareness was compared against sensitivity to gastric function in the same people, the two did not predict each other [30]. Being good at feeling your heart tells you very little about whether you can feel your stomach.

And the questionnaires have the same problem. A 2022 study comparing several established interoception questionnaires found, bluntly, that they do not assess the same construct [31]. Two papers can both report "interoceptive awareness" and be measuring different things, using different instruments, in different people.

One more piece of housekeeping before the difficult section. Interoceptive accuracy declines with age [32]. Whatever this capacity is, it is not fixed for life.

The Problem With Counting Your Heartbeats

Now the part that changes how you should read everything above.

For decades, the standard way to measure interoceptive accuracy has been the heartbeat counting task. It could not be simpler. You sit still. You are told to count your heartbeats silently, without taking your pulse, for a set interval. Your count is compared against the real number from an ECG. Close counts mean good interoception.

An enormous amount of the interoception literature rests on that task. And in 2018 two papers landed on it hard.

The first, working with a sample of five hundred and seventy-two people, reported a set of problems that are difficult to explain away [33]. The correlation between actual and reported heartbeats was low. The accuracy scores overwhelmingly reflected under-reporting rather than error in both directions, meaning the task was largely sorting people by how much they undercount. Worst of all, the scores rose mechanically at slower heart rates. A person with a calm resting pulse scores as a better interoceptor without perceiving anything more clearly than anyone else.

The second, in a sample of one hundred and twenty-three, made the argument from a different direction and reached a blunter conclusion in its title: the heartbeat counting task largely involves non-interoceptive processes [34]. You can score well by knowing roughly what your resting heart rate is and doing arithmetic. Nothing about that requires feeling anything.

Think about what that implies for a study that reports "high interoceptive accuracy predicted X". If the accuracy measure is partly indexing beliefs about your own heart rate, and partly indexing a slow pulse, then some of these findings may be about beliefs and cardiovascular fitness rather than about perception.

The self-report side has its own history. The MAIA questionnaire was published in 2012 [35] and revised in 2018 to fix psychometric weaknesses in the original [36], and it has since been validated in clinical populations [37]. It is a good instrument for what it measures. What it measures is sensibility, which is to say what you believe about your body. It was never a substitute for performance.

Brain-based measures sidestep the reporting problem, which is why they are gaining ground. Imaging work has mapped the neural correlates of turning attention inward and related them to dimensional body-awareness measures [38], and heartbeat-evoked potentials give a response you do not have to ask anyone about [39]. They come with their own analytic difficulties, but they do not depend on a participant's arithmetic.

The Rebuttal

An article that stopped there would be doing the same thing it criticises: presenting one side confidently.

In 2020, Daniel Zimprich, Lisa Nusser and Olga Pollatos published a direct comment on the 2018 critique [40]. Their argument is technical and it is not a hand-wave. Interoceptive accuracy scores are ratio variables, and several of the correlations the critics presented as evidence of a broken measure are, they argue, what you should expect from a ratio variable behaving normally. On that reading the task has real limitations, but the specific case against it was overstated.

The field's own stocktake arrived in 2022, in a short piece in Trends in Cognitive Sciences that tried to sort out what actually drives performance on the counting task [41]. Its answer is that performance is a mixture: trait-like characteristics, expectations about what the experimenter wants, and prior beliefs about one's own heart rate all contribute alongside anything perceptual. That is a warning label rather than a dismissal.

The constructive response has been to build something better. A 2022 paper introduced a heart rate discrimination task using psychophysical methods, which estimates accuracy and precision as separate quantities rather than collapsing them into one score [42]. That separation matters. A person can be systematically biased and highly consistent, or unbiased and wildly noisy, and the old task gave both the same number.

Where does this leave you as a reader? Roughly here. When a study uses the counting task and finds a large, replicated effect that also shows up with other methods, take it seriously. When a study uses the counting task alone and reports a modest correlation with a personality variable, hold it loosely. That is not scepticism for its own sake, just the reading strategy the evidence supports.

The Body As a Guess

There is a theoretical shift running underneath all of this, and it changes what the word "accuracy" even means.

The older picture is a reporting one. The body sends signals up, the brain receives them, and good interoception means receiving them clearly. On that view, accuracy is a matter of signal quality, like having good hearing.

The newer picture inverts it. The brain is running a model of your body and continuously predicting what state it should be in. What travels upward is not the raw report but the mismatch between prediction and arrival. Feeling your heartbeat is closer to your brain's best guess about your heartbeat, corrected by evidence, than to passive reception.

The account was set out for interoception in 2015 [43], building on earlier work that linked interoceptive prediction to something more basic than perception: the sense of being present in your own body at all [44].

Since then it has been pushed in two directions. One is computational, formalising what the gut is doing as an inference problem with actual mathematics attached rather than a metaphor [45]. The other is architectural, asking what a hierarchy of predictions would have to look like inside the insula for any of this to work [46].

Then there is a third direction, and it is the one that surprised me most.

An infant cannot regulate its own physiological state. It gets cold, or hungry, or overwhelmed, and it cannot fix any of that alone. What it has instead is a caregiver who notices and responds. On this account, learning to interpret your own internal signals is not a solo perceptual skill that matures on schedule. It is something you are taught, by someone else reading your body and acting on it, long before you can do either [47].

Whether that is right is an open question. It is a good illustration of how far this field has travelled from Sherrington's filing cabinet.

The practical consequence is sharp. If perception of the body is a prediction weighted by confidence, then two people with identical bodily signals can have different experiences because they weight the evidence differently. One account describes individual differences in interoceptive accuracy as differences in bodily precision, meaning how much confidence the system assigns to incoming signals rather than how strong those signals are [48]. Work on how top-down and bottom-up processing interact during interoception points the same way [49].

This is also where the article stops. Predictive processing is a large idea with implications far beyond the body, and it deserves its own treatment rather than a subsection here.

The Body That Makes a Self

There is a claim in this field that sounds mystical and is not. It is that interoception is part of how you come to be a self at all.

Start with something simpler than selfhood: ownership. You experience your hand as yours. That sounds like it needs no explanation until you learn how easily it breaks. Body ownership can be manipulated in a laboratory in minutes, and the neuroscience of body representation has been rebuilt several times as those manipulations accumulated [50]. Whatever produces the feeling that this body is yours, it is a construction, and constructions can be interfered with.

Interoception enters because it is the one signal stream that never changes address. Your visual field changes when you turn your head. Your hands leave your sight. The tug of your own heartbeat does not go anywhere. If the brain is looking for something stable to build a self around, the viscera are the only candidate that is always present and always yours.

That is roughly the argument behind the interoceptive account of presence, the sense of being here, in this body, now, rather than observing from somewhere. On that account presence is what it feels like when your brain's predictions about your internal state are being confirmed, and dissociative experiences are what happens when they stop being confirmed.

There is anatomy that fits. Work mapping the brain structures involved in cardiovascular arousal alongside interoceptive awareness found overlapping territory, which is what you would expect if the machinery that regulates the body is also the machinery that produces awareness of it [51]. And a 2024 review frames the insula explicitly as an interface, a place where sensation, emotion and cognition are not three separate streams arriving at a junction but a single set of computations wearing three different names depending on which discipline is looking [52].

I want to be careful here, because this is the part of the field where the writing gets loosest. The evidence that interoceptive signals contribute to bodily self-experience is decent. The leap from there to "your sense of self is made of your heartbeat" is a leap, and you should treat anyone who makes it without hesitating as someone selling something.

What is fair to say is narrower and still interesting. Your experience of being a continuous person is not assembled purely from memory and thought. Some of the raw material is a body reporting on itself, mostly below the threshold where you could ever inspect it, and when that reporting is disrupted the experience of being yourself changes in ways people describe as difficult to put into words.

When the Signal and the Story Come Apart

Interoception became a clinical topic quickly, because a lot of psychiatric conditions look, from the outside, like problems with reading the body.

Start with the claim you have most likely encountered: that autistic people have poor interoception. It is repeated widely and it is too simple.

The claim did not come from nowhere. Several studies reported reduced interoceptive ability in autistic participants, in reviews of the area [53] and in direct testing of children [54]. On the face of it the picture looked consistent.

Then a group asked a question nobody had properly controlled for. Alexithymia is difficulty identifying and describing your own emotions. It is not autism, but a separate trait that happens to co-occur with autism at high rates, which means any study of autistic participants is quietly also a study of people with elevated alexithymia unless somebody measures it.

Somebody measured it. When both were in the model, the interoceptive deficit tracked alexithymia and not autism [55]. A companion paper pushed the argument further and proposed that alexithymia is best understood as a general deficit of interoception rather than a purely emotional one [56].

Work using self-report instruments has since traced the same relationship [57]. That consistency across very different measurement approaches is what makes the finding hard to wave away as an artefact of one task.

The logic here is worth slowing down for, because it is a piece of scientific reasoning you can reuse. Two traits travel together. Everyone studies the more visible one. The effect gets attributed to the visible trait for years. Then somebody measures both at once and the effect moves. This is not a rare event in psychology and it is the single most common reason a confident finding quietly reverses.

The picture in autism turns out to be a dimensional one rather than a deficit one. When the three dimensions were measured in autistic participants, they did not all decline together. They came apart, and the size of the gap between them was related to anxiety [58].

That is a much more specific claim than "poor interoception", and it suggests a mechanism. If your body is sending signals you believe you are reading accurately, and you are not, the mismatch itself is a reason to be anxious. The gap is doing the work, not the level.

The same separation shows up when you look at empathy, where interoception and alexithymia make different contributions rather than one standing in for the other [59]. It even shows up in tissue. Structural imaging finds the sociocognitive network differences associated with autism and those associated with alexithymia occupying separable territory [60], which is about as concrete as a dissociation between two co-occurring traits can get.

That correction matters beyond accuracy. Telling a group of people that they cannot feel their own bodies, when the evidence points at a different and separable trait, is not a small mistake.

Elsewhere the findings are more stable, and the useful detail is that the conditions do not all look alike. A 2024 review covering anxiety, depression and psychosis found interoceptive differences in all three, with a different profile in each [61]. That is more informative than a blanket deficit would be, because it means the measure is discriminating rather than simply tracking how unwell someone is.

One study approached it from the perceptual side, testing sensitivity to changes in interoceptive signals rather than asking people to count anything, and found reduced sensitivity across depression, anxiety and substance use disorders [62]. Methodologically that is a better design than the counting task, for the reasons set out earlier in this article.

Depression has the deepest evidence base of the three. Imaging has repeatedly implicated the insula [63], and interoceptive awareness has been tied specifically to positive affect and decision-making in major depression [64]. The suggestive reading is that some of what depression does to motivation may run through a body that has stopped reporting clearly. That reading is not established, and I flag it as speculation rather than finding.

A caution applies to this entire section, and it is the kind that is easy to nod at and then forget by the next paragraph.

Almost all of this is cross-sectional. Researchers measure interoception in people who already have a diagnosis and compare them against people who do not. When a difference turns up, there are at least three explanations and the study usually cannot tell them apart. The interoceptive difference might contribute to the condition. The condition might change interoception. Or something else, medication being the obvious candidate, might be producing both.

You will notice how often the coverage of this topic picks the first explanation and presents it as though the other two had been ruled out. They generally have not been.

Two conditions deserve more than a passing mention, because in both of them the interoceptive story sits close to the centre of the diagnosis rather than alongside it.

Eating disorders came first, and for an obvious reason: hunger and satiety are interoceptive signals, so a condition organised around eating is already a condition about reading the body. The theory says those signals are faint or misread. It is intuitive, it has been repeated for years, and in 2026 somebody finally pooled the evidence.

The result was not what the theory predicted. A systematic review and meta-analysis of sixteen studies found no significant difference in cardiac interoceptive accuracy between people with anorexia nervosa and healthy controls [65]. One moderator did come out significant, and it is an interesting one: accuracy was lower in samples with shorter illness duration, which hints that any deficit may belong to the early stages rather than to the condition as a whole.

Then look at what those authors name as the main limitation of the literature they had just pooled. Almost every study in it used the heartbeat counting task, whose validity, in their own words, has been questioned. A meta-analysis of a whole clinical literature arrived at no clear effect and pointed straight at the instrument. That is the argument of this article turning up inside somebody else's paper, and it is the strongest single piece of evidence I found that the measurement problem is not a technicality.

Addiction is the second, and the framing there is sharper than "poor interoception". Craving is treated as an interoceptive state in its own right, something the body generates and the insula represents, rather than a purely cognitive wanting that happens to have physical symptoms attached [66]. On that reading, a craving is closer to thirst than to a decision.

Emotion research runs alongside all of it, and it is where the old James-Lange intuition finally gets tested rather than argued about. James thought you feel afraid because you notice your body reacting, not the other way round. That is testable now, and the test is whether people who read their bodies more accurately have measurably different emotional lives.

They do, though not in the direction a simple version would predict. Brain imaging shows interoceptive awareness changing how strongly an emotion is subjectively felt rather than merely whether it is detected [67]. And people who score higher on these measures are more susceptible to emotional influence in the first place, meaning better access to the body is not a form of emotional armour [68]. If anything it is the opposite.

More striking is the priming result. Attending to your own body appears to change how you then process emotional material, an effect demonstrated across several neurodegenerative conditions rather than in one convenient sample [69].

Neurodegeneration and stroke have been unusually valuable here, for a slightly grim reason. They let researchers see what happens to feeling when specific tissue stops working, which is an experiment nobody could ethically run [70]. Dementia research has used the same opening to separate two things that look identical from outside: failing to recognise someone else's emotion because you cannot read the face, and failing because you cannot feel your own response to it [71].

That second possibility deserves a moment. It suggests that part of how you recognise what someone else is feeling is by feeling a small version of it yourself, and then reading your own body rather than only their face. If that is right, then a person who has lost access to their own internal signals would appear to have lost empathy, while what they had actually lost was the instrument they were reading it with. Those look the same from the outside and they are not the same thing at all.

Chronic pain has been examined through the same inference lens [72], which is a natural fit, because pain is the clearest everyday case of a bodily signal whose intensity is obviously not fixed by the tissue alone.

Step back from the diagnoses for a moment, because there is a pattern under them.

In every condition above, the problem is rarely that a signal has gone missing. It is that the signal and the story told about it have come apart. The body reports something, the interpretation placed on that report does not match it, and the mismatch is where the difficulty lives. That is why the same physiological arousal can be read as anticipation by one person and catastrophe by another, and it is why stress and cortisol reshape memory so unevenly between people who went through the same event. It is also the mechanism behind how emotions shape memory: what gets kept is not what happened but what your body made of it.

The gut deserves a mention here too, since it is the interoceptive channel with the most public attention right now and the one where the gap between evidence and enthusiasm is widest. The signalling from the gut to the brain is real and consequential. Very little of it is anything you can feel, which is the part usually left out.

Can You Get Better At It?

This is the question everyone actually wants answered, so it deserves a straight answer rather than an encouraging one.

The honest summary: there is some evidence that interoceptive measures move after training, the studies are small and short, and a real methodological problem sits underneath the whole literature.

The problem is circular. If you train people and then measure success with the heartbeat counting task, and the counting task partly indexes beliefs about your own heart rate, then a training programme that changes what people believe about their bodies will look like it improved their perception. That objection is not hypothetical. It falls straight out of the 2018 and 2022 critiques once you apply them to the training literature.

With that said, here is what has been found, and some of it is genuinely encouraging.

An eight-week body scan intervention improved measures of interoceptive processing [73]. Eight weeks is a serious commitment, which cuts both ways: it is long enough to expect real change, and long enough that plenty else in a person's life changed too. A randomised controlled trial of brief mindfulness training also reported changes in interoception [74], and the randomisation matters, because it is the design feature most of this literature lacks.

The work I find most useful is the least dramatic. Rather than asking whether resonance breathing improves interoceptive awareness, one line of research asked what cardiovascular mechanism could possibly connect the two [75]. Mechanism questions are slower and less quotable. They are also how you tell a real effect from a hopeful one.

Consider what a mechanism question buys you. If somebody reports that a practice improved a score, you have one fact and no way to check it beyond running the study again. If somebody can tell you which physiological quantity changed, and why that quantity should affect the score, then the claim becomes attackable in a dozen new ways. Attackable is good. A claim that can only be confirmed and never challenged is not doing much work.

The strongest counter to my own scepticism comes from two directions at once. A group programme combining mindfulness with cognitive behavioural therapy reported improvements in a primary care sample [76], which matters because a busy clinic is a much harder place to produce an effect than a university laboratory full of psychology undergraduates. And a study of interoceptive training reported changes in anterior insula circuitry [77]. That one goes some way to answering the objection I raised two paragraphs ago, because a questionnaire score can drift on its own but insular circuitry is harder to talk yourself into.

Interoceptive skills also form the explicit basis of at least one body-oriented therapy approach [78], so this is not a purely academic question. People are already being treated on the strength of it.

Now notice what is missing from every one of those results. None of them says that training your interoception will improve your memory, your decisions or your attention. That link has not been demonstrated, and given how narrow and conditional the cognitive effects turned out to be, there is no good reason to expect it. The distance between "an eight-week course changed a body-awareness score" and "you will think more clearly" is enormous, and it is routinely collapsed in coverage of this topic.

Underneath all of it sits a harder question. Any claim that practice changes a perceptual ability is a claim about how much the adult brain rebuilds itself, and the adult brain is more conservative than the popular version suggests. Eight weeks of anything is a short time to move tissue. It is a perfectly reasonable time to move a belief about tissue, which is the confound this whole section keeps running into.

How Small Are These Studies, Really?

Set the sample sizes side by side, because the pattern is the argument.

The two tallest bars are the studies asking whether the measuring instrument works. The four shortest are the ones producing the findings that get written up in magazines. The one in the middle, at eighty, is the study that pulled the three dimensions apart. That asymmetry is not a scandal, and it is not unusual in cognitive neuroscience, where a within-subject design timed to the millisecond genuinely does not need hundreds of people to detect an effect.

But it does tell you how to read the coverage. The claims that travel furthest come from the smallest studies. The studies large enough to be confident about are mostly the ones checking the plumbing.

This is not a reason to dismiss the cardiac cycle findings. They have been replicated across different tasks by different groups, which is a stronger form of evidence than any single sample size. It is a reason to be suspicious of any article, including this one, that makes them sound settled.

The Paper That Says None of This Is a Thing

In 2025 a paper appeared in Frontiers in Psychology under a title that does not leave much room for interpretation: There is no such thing as interoception [79].

The argument is not that people cannot feel their own bodies. It is that the term bundles together processes that have little in common, and that bundling them obscures more than it reveals. Cardiac perception, gastric sensitivity, thermal sense, immune signalling and the perception of breathlessness are treated as a single faculty largely because they share a Sherringtonian filing category from 1906.

The evidence assembled elsewhere in this article is uncomfortably supportive of that complaint. Cardiac awareness does not predict gastric sensitivity. Questionnaires named for the same construct do not correlate. Accuracy, sensibility and awareness dissociate. Interoceptive attention and interoceptive accuracy have separable neural substrates. Every one of those results is a fracture line in a construct that is supposed to be one thing.

The counter-argument is that a category can be useful without being a natural kind. There is a shared anatomy in the ascending visceral pathways and a shared cortical territory, and grouping these signals has produced findings that would not have been looked for otherwise. Recent work extends the idea into how bodily experience grounds abstract concepts [80], which is not the sort of question you ask without the umbrella term.

I do not think this is resolved, and an article that pretended otherwise would be doing you a disservice. What you can take from it is a reading habit. When you see the word interoception, ask which channel and which dimension. If the answer is not in the paper, the paper may be about less than it sounds like.

What Your Body Is Actually Doing While You Read This

Let me put the pieces together on one concrete case, because the abstractions are easier to hold once they have somewhere to sit.

Say you are about to walk into an exam.

Your heart rate rises. That is not the anxiety, not yet, just an autonomic adjustment that happened without consulting you. Your baroreceptors are now firing more often, so the pulsed information reaching your brainstem is arriving at a higher rate, and the interruption pattern in your cortex changes with it.

That change is detected against a prediction. Your brain expected a certain bodily state for sitting-in-a-corridor and is receiving a different one. The mismatch is what propagates.

Now the part that decides your experience. That mismatch has to be interpreted, and the interpretation is not determined by the signal. The same elevated heart rate can be labelled as readiness or as dread. Which one you land on depends on context, on prior experience, on what you believe about what your body does under pressure, and on how much confidence your system places in the signal in the first place.

If you have high interoceptive sensibility you will notice all of this vividly. That does not mean your reading of it is correct. Sensibility and accuracy are different dimensions, and the vividness of the experience is not evidence about its accuracy. A person can be intensely aware of a racing heart and badly wrong about what it means.

Meanwhile, underneath the part you can feel, the timing effects are running. Words on the page in front of you are landing at different points in your cardiac cycle, and some of them are landing during systole. On the evidence available, that will make a small difference to what you retain. Small. Measured in laboratory conditions with tightly controlled timing, in samples of a few dozen people, using stimuli chosen to be difficult to detect.

That is the whole shape of this field in one paragraph. Something real is happening. It is happening at a scale you cannot notice and cannot use. And the people studying it are still arguing about how to measure it.

What To Take Away

Four things, and then the questions.

Your body is not a passenger during thinking. Signals from your heart, gut, lungs and tissues reach cortex continuously and shape emotion, memory and choice, mostly without ever becoming conscious.

The cardiac cycle has measurable effects on cognition. Memory, reward learning and inhibition all vary slightly with where a stimulus lands in your pulse. These findings come from small samples, they have been replicated across different tasks by different groups, and they say nothing about how you should study or make decisions.

"Good interoception" is not one thing. Objective accuracy, self-reported sensibility and metacognitive awareness come apart. So do channels: being good at feeling your heart does not make you good at feeling your stomach. Most confusing headlines about interoception are two dimensions being mistaken for one.

The field is arguing about its own foundations, in public, right now. Its main measuring instrument is under serious challenge, the defence of that instrument is technical and repays reading, better measures are being built, and at least one paper argues the whole concept should be broken up. A field checking its own work in public is not a field in trouble, and it is more than most of the coverage of it manages.