Establishing rapport: what the research actually says

Linda Tickle-Degnen and Robert Rosenthal's 1990 model is the one every serious literature since has used, and its value is that it decomposes something people treat as a single quality.

Mutual attentiveness. Both parties focused on each other and on the same thing. This is the component that fails first and most invisibly, because a divided attention is easy to conceal for about ten minutes.

Positivity. Mutual friendliness and care. This is the one people mean when they say rapport, and it is one of three.

Coordination. Behavioural synchrony, balance and predictability between the two parties: turn-taking that works, a shared pace, a sense that the other person is going to behave the way they did last time.

Naming three components rather than one changes the diagnosis. A conversation that feels off is usually not short of friendliness. It is short of coordination, or one party is not actually attending, and adding warmth to either of those problems makes it worse rather than better.

This is the finding that most advice on the subject silently ignores. Tickle-Degnen and Rosenthal found that the relative importance of the three components shifts over the life of a relationship. Positivity and attentiveness dominate early interactions. Coordination and attentiveness dominate later ones.

The practical consequence is large. Nearly every article on building rapport is written for the first-meeting case, which is where positivity does the heavy lifting: the warm opening, the small talk, the visible friendliness. Applied to an established relationship, the same behaviour reads as strange at best and as a preamble to bad news at worst.

With someone you meet weekly, rapport is carried by predictability and continuity. Starting where you left off, remembering the thing they were worried about, running the meeting the way you ran the last one. That is coordination, and it is unglamorous work that no one would describe as rapport-building, which is exactly why it gets dropped in favour of a warmer opening that helps less.

The strongest evidence that rapport is instrumental rather than merely pleasant comes from investigative interviewing, and it is worth taking seriously precisely because that field had every incentive to find that pressure worked.

Christian Meissner and colleagues' Campbell systematic review compared information-gathering approaches with accusatorial ones across field and experimental studies. Information-gathering preserved or increased true confessions while reducing false ones. Accusatorial approaches increased both, which is the worst possible combination: more admissions, less reliable. The later update reports information-gathering producing significantly more true confessions than direct questioning, with a combined odds ratio of 2.43, while accusatorial methods produced far more false confessions than information-gathering, with an odds ratio of 4.41.

The field study is Laurence Alison and colleagues' ORBIT work, which coded real police interviews with real terrorism suspects for interpersonal behaviour and for investigative yield. Correct use of rapport-based approaches was associated with a significant increase in yield. The follow-up finding is the more striking one: rapport-based behaviour in the style of motivational interviewing was far more effective at reducing counter-interrogation tactics, meaning silence, monosyllables, no comment and claimed memory loss, than any coercive alternative.

State the caveat, because it matters: these are police interviews with suspects, not user research sessions, and the transfer is an inference rather than a demonstration. What transfers cleanly is the mechanism. A person who is defended gives you less, and the behaviour that lowers the defence is not the behaviour that feels like control.

The clinical literature converges from another direction. Christoph Flückiger, A. C. Del Re, Bruce Wampold and Adam Horvath's meta-analytic synthesis of the therapeutic alliance covers 295 independent studies and more than 30,000 patients, and reports an alliance-outcome association of r = .278. In a field notorious for effects that do not replicate, that one has, repeatedly, for forty years.

Tanya Chartrand and John Bargh's chameleon effect is the study behind an enormous amount of sales advice, and the advice mostly gets it backwards.

What they established is that nonconscious behavioural mimicry occurs between strangers, and that it increases liking and the smoothness of the interaction. Read the first word again. The effect is about mimicry that neither party is managing. It is a symptom of coordination, not a technique for producing it.

Deliberately mirroring someone's posture is a different behaviour with different risks, and the main one is that it consumes the attention you were supposed to be spending on what they are saying. If you are monitoring their arms, you are not listening, and the attentiveness component is the one you have just spent to buy an imitation of the coordination component.

The useful reading is that synchrony emerges when both people are genuinely attending to the same thing. Fix the attention and the coordination tends to follow.

Two findings, and they point in opposite directions, which is why the honest summary is more useful than either alone.

On one side, the alliance meta-analysis found that for internet-based psychotherapy the alliance-outcome correlation was r = .275, essentially identical to the in-person figure. Whatever rapport does, it survives the screen.

On the other, Jeremy Bailenson's account of nonverbal overload sets out four mechanisms that make video effortful: an excessive amount of close-up eye gaze, the cognitive load of consciously producing and reading nonverbal cues that are normally automatic, the self-evaluation that comes from watching yourself all day, and constrained physical mobility.

Put together: rapport over video is achievable and more effortful, and the effort is spent on exactly the coordination component. Turn-taking has to be managed consciously because the cues that normally handle it are degraded. So the practical adjustments are coordination adjustments. Leave longer gaps. Say explicitly when you have finished a question. Do not fill a two-second silence that is really a half-second of latency. And turn off self-view, which removes one of the four mechanisms for free.

In a session with a participant you have never met and will never meet again, you are in the first-meeting case, so positivity and attentiveness are what you have.

The first four minutes decide it. Not because of anything magical about four minutes, but because the participant is deciding what kind of session this is: an examination, or a conversation in which they are the expert. Everything after runs on that judgement.

Say what you want out loud. That there are no wrong answers is a cliché because it is true and because participants do not believe it by default. What makes it credible is a specific version: that you are trying to find out where the thing is confusing, so confusion is the useful outcome, not a failure on their part.

Start broad. The funnel technique, broad open questions before specific ones, is a rapport instrument as much as a data-quality one, because an early specific question tells the participant that you have a checklist and that their job is to complete it. Nielsen Norman Group lists poor rapport first among the five facilitation mistakes in user interviews, and pairs it with the funnel for this reason. The mechanics are in open-ended questions.

When a participant is performing. The signal is answers that are complete, positive and slightly too fast. The move is not to press but to change the register: ask about a specific past occasion instead of a general opinion, since it is much harder to perform a memory than an attitude.

Observers in the room. Every additional person costs rapport, and silent observers cost the most, because a participant cannot read them. If stakeholders must attend, put them somewhere the participant is not watching, and say who they are and why they are there.

Indi Young's listening sessions are the most radical version of all this: deliberately not called interviews, participant-led, with the researcher's role reduced to demonstrating that the person is not being judged. Worth knowing about even if you never run one, because it makes clear how much of a standard interview is structure the researcher imposed.

Here you are in the second case, where coordination and attentiveness dominate, and almost all the popular advice is aimed at the wrong component.

Continuity is the mechanism. The one-to-one that opens by returning to the thing they raised three weeks ago does more for rapport than any amount of warmth, because it demonstrates the two components that actually matter at this stage: you were attending, and you are predictable.

Predictability is worth defending deliberately. Same slot, same length, not cancelled, and cancelled with a reason and a replacement when it must be. A regularly moved one-to-one communicates its own priority regardless of what is said in it.

The failure mode specific to this case is the meeting becoming a status report, at which point attentiveness has quietly become one-directional. If you are the manager and you are talking more than half the time, the mechanism has already broken.

Rapport is one of the few interviewing skills you can genuinely improve by reviewing your own conduct, and one of the few where in-the-moment self-assessment is worthless. A moderator cannot evaluate their own attentiveness while moderating, for the same reason a driver cannot watch themselves drive.

What is reviewable afterwards, from the text of a session alone, is more than people expect.

• Talk ratio. Roughly how much of the session was your words against theirs. This is the single most diagnostic number in a research session and it is visible from a transcript.
• Question form. Whether the questions were open, and whether the funnel narrowed or jumped to specifics in the third question.
• The recoveries. The moments where the other person went quiet or gave a one-word answer, and what you did next. Those are coordination failures, and what follows them is where the skill is.
• Whether you interrupted. A sentence that stops mid-clause followed by a question is a pattern you can see in a transcript and cannot see in your memory of the session.

All of it requires the session to still exist in text, which is where Earkeep fits. It records your day continuously on your own device, transcribes it there, and writes plain files you own, so the session is available afterwards in order with timestamps, without a separate recording step and without a bot in the call.

Two limits to state plainly rather than gloss. The record does not label speakers, so everything above is your own reading of your own session; in a two-person session the turns are almost always legible from context, and that is a fact about two-person conversations rather than a feature of the tool. And an always-on recorder in a session with a participant is a consent question, not a technical one: a participant session is recorded with the participant's knowledge, what is and is not kept is on privacy, and how to say it before you start is on telling people you're recording.

Rapport decides whether someone will tell you anything. Open-ended questions decide what you ask once they will, and empathic listening decides what you do with the answer. The three pages are written to be read together. In practice this lands in a research routine, which is Earkeep for UX researchers and for UX designers, and in a management routine, which is for people managers. The conversation you should not record, and why, is the skip-level meeting.