What it means to feel heard, and how researchers measure it

In plain words

Psychologists study the sense that someone has really taken in what you said under two related names, perceived responsiveness and the newer, separately measured feeling heard. Responsiveness is usually split into feeling understood, validated and cared for, and this review asks how each part is measured and what happens when researchers change it on purpose. Two cautions run through the evidence. When one study split feeling heard into five components, people rating their own conversations did not tell them apart, and the feeling depends partly on the person who has it. The long-term health links come from married and cohabiting American adults, and the effects are small. One experiment found that replies written by AI made people feel more heard than replies from online strangers, and less heard once they were told the reply came from AI.

One feeling with three names

In research on close relationships the central idea is perceived partner responsiveness, the extent to which a person feels understood, cared for and appreciated by their partner (Slatcher et al., 2015). The Midlife in the United States survey (MIDUS) measures it with three questions about how much a spouse or partner cares about the respondent, understands the way they feel, and appreciates them. The same authors describe these three items as matching the components usually named in the literature, understanding, validation and caring.

Roos et al. (2023) set out to define the everyday phrase, feeling heard. Drawing on work on relationships, health care, procedural justice and communication, they proposed five components: having a voice, and receiving attention, empathy, respect and common ground. They wrote 16 items and tested them on 194 Dutch adults who recalled online conversations, some in which they had felt heard and some in which they had not. The five components did not come apart. The estimated covariances between the five latent factors ranged from .69 to .99, and the authors kept eight items that measure a single factor, the Feeling Heard Scale.

In a preregistered survey of 1,000 adults in the United States, the eight items held together well (omega .93) and the scale correlated .66 with a measure of perceived responsiveness. Among the related measures included, it was the strongest predictor of whether people said they would avoid another conversation with the same person. It also overlapped heavily with a scale of conversational intimacy, the two latent factors correlating at .89, which the authors concede may mean the two constructs are redundant. Both surveys asked people to rate a conversation they remembered, so they show how ratings hang together and say nothing about what the listener actually did.

That shapes the rest of this review. Understanding, validation and care are useful for describing what a listener can get right or wrong. Yet in the one study read here that tested proposed components of feeling heard, raters reported a single overall impression, and MIDUS averages its three items into one score, so in this review the parts can be pulled apart only by what an experiment chooses to change.

Understood: scripted sentences and a quarrel

Morelli et al. (2014) recorded 19 undergraduates at the University of California, Los Angeles describing good and bad events from their lives. A week later, in a scanner, the students were told that other students had watched the videos and picked three sentences in reply. The authors had written the sentences, some conveying understanding and some not. Feeling understood went with activity in the ventral striatum and middle insula, regions earlier tied to reward and social connection, while not feeling understood went with activity in the anterior insula, a region tied to negative affect. The sample was small and the replies were scripted, and the authors themselves point out that their understanding sentences also carried validation, so the study does not separate the first part from the second.

Gordon and Chen (2016), read here as an abstract, asked whether conflict damages a romantic relationship mainly when people feel misunderstood. Across seven studies, using cross-sectional, experimental, daily-life and laboratory designs with couples, conflict went with lower satisfaction among people who saw their partner as less understanding and not among those who felt understood. The authors report that the pattern was not explained by the people who felt understood being more understanding themselves, nor by their being happier in the relationship before the conflict. Of the three parts, being understood is tested in the widest range of designs among the studies read for this review, though the abstract does not say what the experimental studies manipulated or give sample sizes.

Validated: a message on a screen

To validate someone is to convey that their reaction makes sense. Kuo et al. (2022) tested this with 126 undergraduates in Toronto. Each listened to two short recorded scenarios about being rejected, rated their emotions, and then saw feedback on the emotion they had rated highest. The validating message said that 90% of other people had reported the same emotion, and the invalidating one said only 10% had. Every participant saw both, in counterbalanced order, while heart rate, skin conductance and a continuous rating of feeling were recorded.

The results do not line up neatly with theory. After validation, heart rate fell, skin conductance rose and the continuous rating did not change significantly. After invalidation, ratings moved in a positive direction and heart rate also fell, which the authors suggest may reflect relief once the imagined scene ended, or surprise. The effects of validation also differed by which emotion had been validated and by how much difficulty people reported in regulating their emotions. Here validation was a two-line message produced by software, about what other people felt, and the authors list the student sample and the imagined scenario among their limits. The study shows that validation can be delivered in a controlled way in a laboratory, but it does not show that being validated by someone who matters to you works the same way.

Cared for: training a listener

In this review, care is approached through a listening experiment, the one registered report among the experiments read. In a registered report, the hypotheses and analysis plan pass peer review before any data are collected. Moin et al. (2024) paired 606 adults, recruited by student researchers at the University of Reading, into 303 pairs of strangers. Before a six-minute conversation about the speaker's character strengths, held by video call, listeners watched about 13 minutes of training in either high-quality listening or, for comparison, careful speaking, while speakers watched a nature film.

Speakers paired with a trained listener rated the listening as better (d = 0.38), and the listeners rated themselves higher too, but independent coders who watched 130 of the recordings found no difference in listening quality. Of the three conversation outcomes, the one that moved was positivity resonance, a shared sense of warmth and mutual care during the conversation, at d = 0.20; trained listeners also reported more intention to keep listening this way (d = 0.25). The speakers' feeling of being authentic and their state anxiety did not change. The authors call the manipulation partly successful, the effects were small, and the topic was a pleasant one on which both groups rated the listening high. Taken as a test of the cared-for component, the study found a small rise in warmth that both people shared and no change in the speaker's own state.

Ten and twenty years later

The long-term evidence comes mostly from MIDUS and its three items. Slatcher et al. (2015) followed 1,078 married or cohabiting adults, 95.1% of them White. Responsiveness reported in 1995 to 1996 predicted a steeper fall in cortisol across the day, measured in saliva about ten years later, after controls for demographics, depressive symptoms, agreeableness, marital risk and the emotional support people gave. It did not predict total cortisol output. Part of the association ran through a decline in negative affect over the decade, and the authors call the effects small.

Stanton et al. (2019) linked the same items to deaths among 1,208 adults over 20 years, during which 100 died. Neither the level of responsiveness nor its change over the first ten years predicted mortality directly. An indirect path from the level, through how much people's negative mood rose on stressful days, vanished once relationship and personality covariates were added, whereas a path from change in responsiveness, through that same daily reactivity, to mortality survived every covariate with a small coefficient. The authors note that the data are correlational, the sample is almost entirely White and the deaths are few.

Outside the family, Kluger et al. (2024) pooled 664 effect sizes in a registered meta-analysis of perceived listening at work and found an average correlation of .39 with work outcomes. They warn that 75% of the studies relied on self-report. With the data resting only on self-report removed, the correlation with job performance was .21, against .36 overall, and only one study had used an experiment to test whether listening changes measured performance. Taken together, these sources show sizeable associations when one person reports everything, smaller ones when outcomes are measured separately and small ones over decades, with few experiments behind any of them.

The perceiver's share

Perceived responsiveness is a judgement, and the person making it contributes to it. Canevello and Crocker (2010) followed first-year roommates who had never met before college, 115 pairs across a semester and 65 pairs across three weeks. A student's own goal of supporting the roommate predicted how responsive that student was, and students who saw themselves as responsive also saw their roommates as more responsive, which the authors call projection. What one roommate reported doing also appeared in the other's perceptions, more clearly in daily reports than in weekly ones. The samples were mostly women, nothing was manipulated, and the authors acknowledge that their measures overlapped a good deal. Gordon and Chen (2016) found that being more understanding oneself did not explain their effect, so the two lines of work are compatible, with the perception carrying something of the perceiver and something of the partner.

How much the perception predicts differed between two countries, which the authors suggest may reflect cultural context. Tasfiliz et al. (2018) compared 3,079 married or cohabiting adults in the United States with 861 in Japan, using the same three items. Responsiveness predicted both hedonic well-being (life satisfaction and mood) and eudaimonic well-being (scales such as purpose in life, personal growth and self-acceptance) in both countries, more strongly in the United States, and the gap was wider for eudaimonic well-being. The Japanese sample was drawn only from the Tokyo area.

Two smaller results point the same way. In Moin et al. (2024), speakers reported an improvement in listening that trained coders could not see. In Morelli et al. (2014), students more sensitive to rejection tended to feel less understood by the same unsympathetic sentences, a marginal correlation (r = -.42, p = .07) in 19 people.

When the reply comes from a language model

Yin et al. (2024) asked 501 online participants to describe a difficult situation they were facing. Half were sent a reply written by another online participant who had been asked to make them feel understood. The other half were sent a reply from Bing Chat, given the same aim and a length matched to the human replies, whose medians ran from 52 to 63 words across batches. Each reply was then labelled at random as coming from a person or from Bing Chat. Among the 456 who rated their reply, those sent the AI's reply felt more heard (5.74 against 5.17 on a seven-point scale), and those told the reply came from AI felt less heard (5.13 against 5.81). The two effects were similar in size and did not interact. The human responders were strangers putting in moderate effort, and the study had no unlabelled condition, so it cannot say whether the AI label lowered ratings or the human label raised them; the authors make both points.

Rubin et al. (2025), read as an abstract, report nine studies with 6,282 participants in which empathic replies, all written by AI, were labelled as coming from a human or from AI. Replies attributed to humans were rated more empathic and supportive. The effect held across reply lengths, delays and models, and it was carried mainly by replies that stressed sharing feelings and care.

De Freitas et al. (2024), a working paper, randomly assigned 1,381 online adults to talk for 15 minutes with one of three chatbots the authors had built. The one designed to be empathic was rated far higher on feeling heard. In this working paper, self-rated loneliness was lower after the empathic chatbot than before, and in a mediation analysis comparing it with the other two chatbots, feeling heard accounted for more of the difference in that drop than the chatbot's perceived performance did. That outcome was measured minutes after the chat.

None of these studies measured anything like the cortisol and mortality outcomes above, which concern spouses and partners over years; the Yin et al. and De Freitas et al. studies measured their outcomes at the end of a single session. Our Technical Note 02 reports that the average length of Bloom's replies fell from 54 words to 11. The replies in Yin et al. (2024) were several times that length, and none of the studies read here reports testing replies as short as 11 words (the Rubin et al. abstract does not give the lengths it varied).

What is not yet known

No study read for this review changed one of understanding, validation and care while holding the other two fixed and then measured all three, so whether they are separate experiences or three descriptions of one remains open.

The health findings rest largely on one panel of mostly White American adults in long relationships. Whether responsiveness predicts cortisol or survival elsewhere, or in relationships other than marriage and cohabitation, has not been shown, and the mortality analysis rests on 100 deaths.

It is unknown whether the penalty for an AI label persists when someone keeps talking to the same system over months, and how people judge a reply whose source is not stated.

Whether feeling heard by software relates to anything measured months or years later, as the partner studies did, is untested in the work read for this review.

How the length of a reply affects whether a person feels heard has not been tested in the studies reviewed; Rubin et al. (2025) varied length, but its abstract reports only that the label effect held across lengths. Nor is it known why trained coders and speakers disagreed about listening in the one registered report, which leaves open how much of feeling heard tracks what the listener did.

References

1. Canevello, A., and Crocker, J. (2010). Creating good relationships: responsiveness, relationship quality, and interpersonal goals. Journal of Personality and Social Psychology, 99(1), 78-106. https://pmc.ncbi.nlm.nih.gov/articles/PMC2891543/
2. De Freitas, J., Uguralp, A. K., Uguralp, Z. O., and Puntoni, S. (2024). AI companions reduce loneliness. Harvard Business School Working Paper 24-078; arXiv:2407.19096. https://arxiv.org/abs/2407.19096
3. Gordon, A. M., and Chen, S. (2016). Do you get where I'm coming from?: Perceived understanding buffers against the negative impact of conflict on relationship satisfaction. Journal of Personality and Social Psychology, 110(2), 239-260. Abstract only. https://pubmed.ncbi.nlm.nih.gov/26523997/
4. Kluger, A. N., Lehmann, M., Aguinis, H., Itzchakov, G., Gordoni, G., Zyberaj, J., and Bakaç, C. (2024). A meta-analytic systematic review and theory of the effects of perceived listening on work outcomes. Journal of Business and Psychology, 39, 295-344. https://doi.org/10.1007/s10869-023-09897-5 (read at https://www.hermanaguinis.com/pdf/JBPlistening.pdf)
5. Kuo, J. R., Fitzpatrick, S., Ip, J., and Uliaszek, A. (2022). The who and what of validation: an experimental examination of validation and invalidation of specific emotions and the moderating effect of emotion dysregulation. Borderline Personality Disorder and Emotion Dysregulation, 9, 15. https://pmc.ncbi.nlm.nih.gov/articles/PMC9116024/
6. Moin, T., Weinstein, N., Itzchakov, G., Branson, A., Law, B., Yee, L., Pape, E., Cheung, R. Y. M., Haffey, A., Chakrabarti, B., and Beaman, P. (2024). The effects of listening on speaker and listener while talking about character strengths: an open science school-wide collaboration. Royal Society Open Science, 11(12), 221342. https://pmc.ncbi.nlm.nih.gov/articles/PMC11651903/
7. Morelli, S. A., Torre, J. B., and Eisenberger, N. I. (2014). The neural bases of feeling understood and not understood. Social Cognitive and Affective Neuroscience, 9(12), 1890-1896. https://pmc.ncbi.nlm.nih.gov/articles/PMC4249470/
8. Roos, C. A., Postmes, T., and Koudenburg, N. (2023). Feeling heard: Operationalizing a key concept for social relations. PLOS ONE, 18(11), e0292865. https://pmc.ncbi.nlm.nih.gov/articles/PMC10688667/
9. Rubin, M., Li, J. Z., Zimmerman, F., Ong, D. C., Goldenberg, A., and Perry, A. (2025). Comparing the value of perceived human versus AI-generated empathy. Nature Human Behaviour, 9(11), 2345-2359. Abstract only. https://www.nature.com/articles/s41562-025-02247-w
10. Slatcher, R. B., Selcuk, E., and Ong, A. D. (2015). Perceived partner responsiveness predicts diurnal cortisol profiles 10 years later. Psychological Science, 26(7), 972-982. https://pmc.ncbi.nlm.nih.gov/articles/PMC4504783/
11. Stanton, S. C. E., Selcuk, E., Farrell, A. K., Slatcher, R. B., and Ong, A. D. (2019). Perceived partner responsiveness, daily negative affect reactivity, and all-cause mortality: a 20-year longitudinal study. Psychosomatic Medicine, 81(1), 7-15. https://pmc.ncbi.nlm.nih.gov/articles/PMC6298854/
12. Tasfiliz, D., Selcuk, E., Gunaydin, G., Slatcher, R. B., Corriero, E. F., and Ong, A. D. (2018). Patterns of perceived partner responsiveness and well-being in Japan and the United States. Journal of Family Psychology, 32(3), 355-365. https://pmc.ncbi.nlm.nih.gov/articles/PMC5922804/
13. Yin, Y., Jia, N., and Wakslak, C. J. (2024). AI can help people feel heard, but an AI label diminishes this impact. Proceedings of the National Academy of Sciences of the United States of America, 121(14), e2319112121. https://pmc.ncbi.nlm.nih.gov/articles/PMC10998586/