A person completes eight weeks of mindfulness training and says, afterward, that they notice bodily sensations more clearly, worry less about discomfort and can return attention to the body with greater steadiness.
Has their interoception improved?
The answer depends on what interoception means in that sentence.
A 2025 Scientific Reports meta-analysis by I. N. Treves and colleagues addresses a specific version of the question: whether mindfulness meditation training changes self-reported interoception in randomized controlled trials.
That qualifier is the centre of the paper.
The meta-analysis does not establish that meditation makes people objectively better at detecting heartbeats, respiratory resistance or gastric signals. It synthesizes questionnaires about how participants report noticing, attending to, interpreting and relating to internal bodily experience.
What the study did
The authors preregistered a meta-analysis of randomized controlled trials examining mindfulness meditation training and self-reported interoception.
The final synthesis included 29 randomized controlled trials with 2,191 participants. The sample was predominantly female and relatively young on average. The interventions included mindfulness-based programmes, body-focused approaches and other meditation-related training formats.
The Multidimensional Assessment of Interoceptive Awareness, or MAIA, was the most commonly used measurement family.
Across studies, mindfulness training was associated with a small average improvement in self-reported interoception. The reported overall standardized effect was approximately Hedges' g = 0.31, with a 95 percent confidence interval from about 0.21 to 0.42.
The authors reported low-to-moderate heterogeneity and did not find evidence of publication bias in their analyses.
This is a meaningful result.
It is also easy to overstate.
A small effect can still matter
An average standardized effect around 0.3 is not transformative at the individual level.
It means that, across the included trials, intervention groups tended to report somewhat greater interoceptive awareness than control groups after training.
Small effects can matter when interventions are low risk, scalable or cumulative. They can also disappear when translated into a different population or outcome.
The effect size therefore should not be converted into the claim that mindfulness "reconnects people with their bodies" as a universal fact.
The result is specific to included programmes, participants, controls and self-report measures.
Mindfulness-based programmes showed a somewhat larger estimate
The authors reported a larger average effect for mindfulness-based programmes than for the pooled set overall, approximately g = 0.41.
Other moderator analyses did not identify robust differences across several study characteristics such as dosage or clinical status.
Moderator null results need care.
Failure to detect a moderator does not prove that the factor never matters. Meta-analyses can have limited power for subgroup questions, especially when categories contain relatively few trials or interventions differ in ways that are difficult to code.
The paper supports a broad average effect more strongly than it supports fine-grained prescriptions about which exact mindfulness format works best for whom.
Why self-report matters
Self-report is sometimes dismissed as "subjective" as if that made it scientifically unimportant.
For a construct such as whether a person worries about bodily sensations, trusts them, notices them or can direct attention toward them, subjective report is part of the phenomenon.
The MAIA family was designed precisely because the relationship to internal sensation includes attentional and interpretive dimensions that cannot be read from an ECG.
If mindfulness reduces fear of bodily sensation or increases perceived ability to sustain attention, that can be a real psychological change even if heartbeat-detection accuracy does not change.
The error lies in changing labels after the result is obtained.
Self-reported interoception should remain self-reported interoception.
Why objective accuracy is a separate question
Performance-based interoception tasks compare judgments with physiological events or controlled perturbations.
The mindfulness literature on these tasks has been more heterogeneous. Prior work has not consistently shown large improvements in objective heartbeat perception from meditation training.
That pattern is not necessarily contradictory to the meta-analysis.
A person may learn to attend to bodily sensation with less worry, differentiate emotions better or feel more capable of using bodily information without becoming better at detecting each heartbeat.
The intervention may change attention and appraisal rather than receptor-level sensitivity.
Different measures can reveal different mechanisms.
Measurement concentration is both strength and limitation
The frequent use of MAIA creates comparability across studies.
It also concentrates the literature around one self-report architecture.
MAIA is multidimensional. Some subscales concern Noticing, others Not-Worrying, Attention Regulation, Emotional Awareness, Self-Regulation, Body Listening or Trusting. A total or composite change can hide which dimensions moved.
If mindfulness primarily changes how sensations are appraised rather than how often they are noticed, an aggregate interoception label can obscure the mechanism.
Future meta-analytic work could benefit from subscale-level synthesis when enough trials report comparable data.
What randomized trials add
Randomization strengthens causal inference compared with cross-sectional comparisons between meditators and non-meditators.
Long-term meditators differ from non-meditators in many ways: motivation, health behavior, education, personality, expectations and prior experience. Randomized training studies reduce some of those selection differences.
They do not remove every source of bias.
Participants usually know whether they are receiving mindfulness training. Expectancy can influence self-report. Control conditions vary in credibility and contact time. Attrition can be selective. Many samples are not demographically representative.
The meta-analysis is stronger than anecdote and stronger than simple cross-sectional association.
It is not immune to the limitations of behavioral intervention research.
The sample matters
The pooled participants were predominantly women and, on average, relatively young adults.
That does not invalidate the result.
It limits generalization.
We should be cautious about assuming identical effects in older populations, adolescents, men underrepresented in the trials, people with severe medical illness or cultural settings in which body attention has different meanings.
Interoception is not independent of language, health status or social learning.
A global claim requires a global evidence base.
Correction history
Scientific Reports published a correction on 23 December 2025.
The correction addressed an author's name, Joanne Qina'au, rather than the statistical results or conclusions.
This matters for provenance. A canonical knowledge library should record corrections even when they do not change substantive findings.
Correction tracking is part of research literacy, not an admission that a paper has failed.
Does more self-reported interoception mean better mental health?
Not automatically.
Some forms of internal attention can be adaptive, while hypervigilant monitoring can amplify distress. The MAIA was designed partly to distinguish qualities of bodily attention rather than assuming that more noticing is always beneficial.
The meta-analysis itself concerns change in self-reported interoception, not a universal mental-health outcome.
If an intervention increases Noticing but also increases worry in a subgroup, the practical meaning differs from an increase in Attention Regulation or Trusting.
The quality of attention matters alongside its quantity.
Active controls are especially important here
A waiting-list comparison asks whether mindfulness training differs from doing little or nothing during the same period. An active control asks a harder question: whether mindfulness differs from another credible activity with similar time, attention and social contact. Self-report outcomes are especially sensitive to this distinction because participants know what they practiced and may have expectations about how they are supposed to change.
The included trials used different control conditions. The pooled effect therefore reflects a family of causal contrasts rather than one identical experiment repeated 29 times. Reading the forest plot as if every study asked exactly the same question would overstate uniformity.
Follow-up duration matters
Post-intervention change tells us that reports differed after training. It does not automatically establish persistence. Some practices may produce a temporary increase in attention that fades when structured training ends; others may alter habits that persist. Long-term follow-up was not uniform enough across the literature to make durability the central conclusion of the meta-analysis.
For programme design, maintenance should therefore be measured rather than assumed. A capacity that disappears when reminders stop is different from one that becomes available under ordinary pressure.
Adverse or uncomfortable experiences should not be erased
Mindfulness is often framed as low risk, but inward attention can intensify distress for some people, particularly when bodily sensations are associated with panic, pain, trauma or illness. The meta-analysis is not an adverse-event review and cannot establish a complete safety profile.
This matters ethically. A positive average self-report effect does not mean every participant benefited or that mandatory body-focused practice is appropriate. Institutions need opt-out pathways, adaptation and referral boundaries rather than treating a pooled effect as universal permission.
Statistical significance and practical significance
The confidence interval excludes zero, supporting a reliable average effect in the analyzed literature. Practical importance is a separate question. An effect of roughly one third of a standard deviation can be meaningful at population scale and modest for an individual. It also does not tell us which questionnaire dimensions changed enough to alter daily behavior.
A strong implementation claim would connect score change to observable functioning, persistence and the intended capacity. The meta-analysis provides one link in that chain, not all of them.
Subscale synthesis could change the story
Because MAIA contains qualitatively different dimensions, future analyses may find that mindfulness affects some components more than others. Attention Regulation and Not-Worrying, for example, could plausibly respond differently from simple Noticing or Trusting. Pooling across total scores and varied subscales can blur those distinctions.
This does not invalidate the current meta-analysis; it defines a next question. A subscale-level synthesis would need enough trials using comparable versions and reporting complete statistics. Until then, the most defensible conclusion remains at the broader self-report level.
Mechanism studies should connect process and outcome
If mindfulness changes self-reported interoception, a mechanistic trial can ask whether that change mediates improvements in stress, pain or emotional functioning. Mediation is stronger when temporal ordering is clear and when the proposed mediator is measured independently of the final outcome.
Without that bridge, interoceptive change may be an accompanying feature rather than the reason an intervention helps. This distinction matters for programme design because resources should target mechanisms that actually carry benefit, not merely variables that move at the same time.
Replication should include less represented populations
The participant profile in the meta-analysis limits confident generalization. Future randomized studies should deliberately recruit more men, older adults, adolescents and culturally diverse samples, and should report medical and psychiatric factors that can alter bodily attention.
Diversity here is not only a fairness goal. It tests whether the mechanism itself generalizes across bodies with different histories, physiology and learned meanings.
Institute synthesis: practices may change relationship before sensitivity
The Institute uses the phrase relationship before sensitivity as a hypothesis for interpreting some contemplative findings.
This is Institute synthesis, not a conclusion stated by Treves and colleagues.
It means that a practice may first change how someone attends to, appraises or responds to internal sensation before it changes objective sensory thresholds – and objective thresholds may not change at all.
This is a useful capacity hypothesis because it prevents practice from being judged by the wrong outcome.
If the purpose of a programme is to reduce reactive worry and improve flexible attention, objective heartbeat counting may be a poor primary endpoint. If the purpose is to train detection of a respiratory load, self-report alone is insufficient.
The desired capacity should choose the measure.
Implications for programme design
Mindfulness is often added to education and workplace programmes with vague promises of self-awareness or resilience.
The meta-analysis supports a more bounded claim: mindfulness training can produce a small average improvement in how participants report aspects of interoceptive awareness.
That may be relevant to programmes concerned with attention to bodily state.
It does not justify mandatory meditation, clinical promises or performance evaluation based on self-report scores.
Institutional programmes should also remember that inward attention is not universally comfortable. Trauma history, panic symptoms, pain and medical conditions can change how body-focused practice is experienced.
Choice, adaptation and referral boundaries matter.
What the evidence does not show
The meta-analysis does not show that mindfulness meditation reliably improves objective interoceptive accuracy. It does not show that all meditation styles have equal effects. It does not establish that all MAIA subscales improve to the same degree. It does not prove clinical benefit for every population. It does not demonstrate that more inward attention is always adaptive.
The result also does not establish the mechanism. Changes could reflect attentional skill, appraisal, expectancy, reduced avoidance, increased vocabulary for bodily experience or combinations of these processes.
A small average self-report effect should remain exactly that until stronger evidence supports a larger claim.
Why the paper matters
The paper is useful not because it provides a dramatic verdict on meditation.
It provides a disciplined quantitative answer to a recurring claim.
Across randomized trials, mindfulness training appears to shift self-reported interoception modestly in a positive direction.
That conclusion is more credible than testimony and narrower than the wellness language that often surrounds contemplative practice.
For the Institute, the narrowness is a strength.
It shows how a practice can be studied without turning one outcome into a theory of the whole person.
Research references
- Treves IN et al. A meta-analysis of the effects of mindfulness meditation training on self-reported interoception. Scientific Reports. 2025;15:38889. DOI: 10.1038/s41598-025-22661-4. Published 6 November 2025.
- Correction to: A meta-analysis of the effects of mindfulness meditation training on self-reported interoception. Scientific Reports. Published 23 December 2025. DOI: 10.1038/s41598-025-32949-0.
- Mehling WE et al. The Multidimensional Assessment of Interoceptive Awareness (MAIA). PLoS One. 2012;7(11):e48230. DOI: 10.1371/journal.pone.0048230.
- Mehling WE et al. MAIA-2. PLoS One. 2018;13(12):e0208034. DOI: 10.1371/journal.pone.0208034.
- Khalsa SS et al. Interoception and Mental Health: A Roadmap. Biol Psychiatry Cogn Neurosci Neuroimaging. 2018;3(6):501-513. DOI: 10.1016/j.bpsc.2017.12.004.