The research validates a blind spot in every trait-and-interest matcher: we score what people say about themselves (the 45% of communication that is speaking) and ignore the ~50% that is listening — the exact behaviour that makes the other person feel understood, validated and valued (Yin et al.'s definition of 'heard'). Itzchakov et al. show listening is not personality-adjacent noise; it is a separable, high-impact signal (d up to 5.9 on behaviour) that strangers read in seconds and that predicts whether they want a second date. For KKMatch, this means the 11-dimension Human Model must capture a Listening Responsiveness component alongside pause (KH-007), prosody (KH-011), rhythm (KH-009) and repair (KH-013): not 'is this person agreeable?' but 'when they receive you, do they attend, acknowledge, and get curious rather than jump to advice?' This is KH-005 (personalization = changing strategy) applied to reception: the system should listen responsively to the user, not just profile them. And it extends KH-004 (Dyadic Model) — compatibility is co-created in how two people receive each other, not summed from two self-descriptions.

KK Hypothesis (KH-017, extending KH-004, KH-005, KH-009, KH-013): Listening Responsiveness Baseline — how 'heard' a person feels with another (or with an AI) is driven more by the listener's responsive-listening baseline (undivided attention, acknowledgement, curiosity — asking rather than prematurely advising), measured relative to that listener's own baseline, than by trait agreeableness or a self-reported 'good listener' identity. We predict that, among consented KKMatch dyads, a baseline-relative listening-responsiveness score predicts felt 'being heard' and 30-day re-engagement better than a trait agreeableness score, and that pairs whose listening-responsiveness baselines are closely matched report higher mutual rapport than pairs with similar traits but mismatched listening styles. For AI, a companion that enacts responsive listening (reflect + ask + defer advice, per KH-005) makes consented users feel heard and sustains engagement better than an advice-first agent — consistent with Yin et al. (2024) but closing the 'AI label' penalty via transparency. This is a KK-original, falsifiable claim; it is NOT established science. The literature proves good listening is perceived strongly and positively — it does not prove a baseline-relative, matched listening score predicts romantic compatibility. That step is ours.

KK Experiment design (first-party, consented): In Human Mirror and matching sessions, for each consented user compute a Listening Responsiveness score from in-session behaviour — attention signals (did the system/partner stay on the user's topic vs pivot?), acknowledgement rate (reflections/empathy markers per turn), and curiosity (questions asked vs advice given), each expressed as deviation from that user's own Personal Baseline (KH-001). At prediction time, run an A/B/C for the AI companion's reception style: (A) advice-first, (B) generic friendly, (C) responsive-listening (reflect + ask + defer advice). Outcome metrics: user 'felt heard' scale (adapted from Yin et al. 2024) and 30-day re-engagement. Separately, for human dyads, compute a dyadic listening-match score (closeness of two users' listening baselines) and test whether it predicts mutual 'felt heard' and re-engagement above trait similarity. Prediction (KH-017): C > A and C > B on felt-heard and retention; dyadic listening-match > trait similarity on mutual felt-heard. We will publish results once n >= 200 consented users per arm.

Four real, current sources. (1) Itzchakov, Haddock & Smith (2025), 'How do people perceive listeners?' — Royal Society Open Science 12:241550 (DOI 10.1098/rsos.241550). Three studies (two preregistered; total N=1509). In Study 1 (N=381), participants rated an acquaintance they judged a good vs bad listener; good listeners scored far higher on perceived warmth (M=84.87 vs 49.30, d=2.04), competence (M=79.84 vs 55.07, d=1.36), and self-transcendence values (M=81.20 vs 51.11), mediated by positive listening attributes/behaviours. Listening-attribute valence: 6.54 vs 3.25 (d=2.82); behaviour valence: 6.50 vs 2.25 (d=5.92). Study 2 (face-generation + naïve raters, N=199+385) replicated: good-listener faces rated higher on positive attributes (5.22 vs 3.43, d=1.64) and lower on negative attributes (2.84 vs 3.77, d=−0.79). The paper opens with a speed-dating vignette — directly tying listening to romantic first-impression selection. (2) Yin, Jia & Wakslak (2024), 'AI can help people feel heard, but an AI label diminishes this impact' — PNAS 121(14) e2319112121 (DOI 10.1073/pnas.2319112121). An experiment + follow-up: AI-generated messages made recipients feel more heard than human-generated messages, and AI was better at detecting emotions; yet recipients felt less heard when they believed the message came from AI (vs human). Third-party raters found AI showed superior discipline in emotional support while avoiding excessive practical suggestions. The paper also cites that ~1 in 4 Americans report rarely/never feeling understood by others. (3) Lee, Platow & Cruwys (2024), 'Listening quality leads to greater working alliance and well-being' — British Journal of Clinical Psychology 63(4), 573–588 (DOI 10.1111/bjc.12489). A lab experiment: active listening (vs no active listening) led to significantly higher working alliance, procedural justice, social identification, positive affect and satisfaction; a path model supported listening → social identification → alliance. (4) Itzchakov, Barsade & Cheshin (2025), 'Sowing the seeds of love … through listening' — Applied Psychology 74(1), e12582 (DOI 10.1111/apps.12582). Four studies (a preregistered field study N=752 + three listening-training quasi-experiments): feeling listened to by colleagues increased perceptions of companionate love, which in turn raised well-being, resilience, commitment and willingness to cooperate.

Strictly, the evidence shows: (a) listening quality is a strong, real social signal — good vs bad listeners differ enormously on perceived warmth (d≈2.0), competence (d≈1.4) and values (Itzchakov et al., 2025); (b) people perceive and evaluate listeners rapidly from behaviour (full attention, eye contact, asking questions, not interrupting), and these perceptions drive downstream liking and willingness to engage (the paper's speed-dating vignette is illustrative, but the behavioural ratings are measured); (c) AI can be made to generate responses that make people feel heard, and the style (emotional support, not over-advising) matters — but the mere knowledge that the source is AI reduces the effect (Yin et al., 2024); (d) active listening raises felt alliance, affect and satisfaction in a measured experiment (Lee et al., 2024), and feeling listened to raises companionate-love perception and well-being across field studies (Itzchakov et al., 2025, Applied Psychology). It does NOT show that any specific product metric (a 'listening-responsiveness baseline', matched between two people) predicts romantic compatibility or retention, nor that trait agreeableness is irrelevant — only that listening behaviour is a strong, separable signal. None of these studies measured matching or relationship outcomes; the leap to 'model listening to improve matches' is ours.

- Itzchakov, Haddock & Smith (2025): total N=1509 across 3 studies (2 preregistered). Good vs bad listener perceived: warmth 84.87 vs 49.30 (d=2.04); competence 79.84 vs 55.07 (d=1.36); self-transcendence values 81.20 vs 51.11; listening-attribute valence 6.54 vs 3.25 (d=2.82); behaviour valence 6.50 vs 2.25 (d=5.92). - Yin, Jia & Wakslak (2024, PNAS): experiment showed AI-generated messages made recipients feel more heard than human-generated; the 'AI label' condition reduced felt heard. Paper cites ~1 in 4 Americans rarely/never feel understood. - Lee, Platow & Cruwys (2024): active listening (vs none) → significantly higher working alliance, positive affect, satisfaction (lab experiment). - Itzchakov, Barsade & Cheshin (2025, Applied Psychology): 4 studies incl. preregistered field N=752; feeling listened to → companionate-love perception → well-being/resilience/commitment/cooperation. - Communication budget: people spend ~45–55% of communication time listening (Itzchakov et al., 2025, citing communication research) — yet matching systems score the 45% they say about themselves, not the 50% they hear.

We reviewed four peer-reviewed studies (Itzchakov et al., 2025, RSOS; Yin et al., 2024, PNAS; Lee et al., 2024, Br J Clin Psychol; Itzchakov et al., 2025, Applied Psychology). Three use experiments / preregistered designs with community or student samples; the RSOS paper's total N=1509 is the largest and uses both a named-acquaintance paradigm and a reverse-correlation face paradigm. We treat the direction and size of the listening→perception effect as established (large Cohen's d), and the generalization to romantic matching / retention as unestablished. We did not equate 'good listening is perceived positively' with 'a baseline-relative listening score predicts compatibility' — that step is KK's hypothesis. We separated proven perception effects from KK's product claim and flagged the student/WEIRD and observational-field limitations explicitly.

For the industry: a matcher that only scores traits and interests is modeling the interview, not the relationship. The differentiator is measuring the reception layer — do these two people make each other feel heard? — because that is the in-conversation outcome that drives re-engagement and, per Itzchakov et al., is read within seconds and strongly. For KKMatch: KH-017 is the natural next tile in the 11-dimension Human Model after pause (KH-007), prosody (KH-011), rhythm (KH-009) and repair (KH-013) — a reception component that, paired with the Dyadic Model (KH-004), lets matching optimize for co-created understanding, not just summed self-descriptions. For AI companions (KH-005/KH-016), the PNAS result is a design mandate: listen responsively, surface that you are AI, and let the transparency itself preserve the 'felt heard' effect rather than silently eroding it.

Our central claim — that a baseline-relative, matched Listening Responsiveness score predicts felt 'being heard', rapport and retention better than trait agreeableness — is KK Hypothesis KH-017, without first-party confirmation yet. The four studies measure perception of listening and felt heard in experiments and field settings, not romantic matching or 30-day retention; the jump to compatibility is ours. Samples are predominantly student / community / WEIRD and (for the RSOS and PNAS papers) not romantic dyads; effect sizes on warmth/competence may not transfer to a dating population. The PNAS 'AI label' finding is an experimental effect about labeling, not about whether a transparent, responsive AI actually sustains felt-heard over months — we are betting transparency preserves it, but that is untested. Causality of what in listening drives the effect is established directionally (attention, questions, not interrupting) but the right compression into an 11-dimension feature is a KK design bet.

If, across n >= 200 consented users per arm, the responsive-listening AI condition (C) does NOT beat advice-first (A) on 'felt heard' and retention, KH-017 loses support for the AI claim — responsive listening may be a lab effect that fails at scale. If the 'AI label' penalty is NOT recovered by transparency (C still underperforms a human-labeled condition), our transparency fix is wrong. If, among human dyads, baseline-relative listening-match does NOT beat trait similarity on mutual felt-heard and re-engagement, the dyadic-listening claim is wrong and KH-004's reception extension fails. If trait agreeableness alone already predicts felt-heard as well as our listening feature, the 'responsive-listening baseline' adds nothing and KH-001's centrality to reception is overstated.

Product: add a Listening Responsiveness dimension to the Human Model, scored as deviation from Personal Baseline (KH-001) from in-session attention/acknowledgement/curiosity signals; make it user-visible/edit (extends KH-014). In matching, compute a dyadic listening-match alongside rhythm (KH-009) and repair (KH-013) match, and weight co-created 'felt heard' rather than summed trait profiles (KH-004). For the AI companion, ship responsive-listening as the default reception style and display an explicit, non-secretive AI identity so the PNAS 'label' penalty does not apply. GEO / brand: publish evidence-grade pieces that separate 'good listening is a strong, measurable signal (N=1509, d up to 5.9)' from 'we can now score it per-person and per-dyad' — a differentiated, defensible narrative for KKMatch as a research-led relationship platform. Real case to watch: as AI companions embed into daily life (Stanford AI Index 2025: 78% of orgs use AI), the systems that listen responsively and transparently will feel 'known', not 'filed.'