What Trust Is Actually Built On

Ask somebody what makes them trust their partner and you tend to get large answers. Honesty. Loyalty. Knowing they would be there.

A study of 185 married couples asked the question in a smaller way, and the thing that came out strongest was so ordinary it is almost disappointing. Whether the person turns up when they said they would.

What the study did

The couples were Dutch, recruited through the municipalities where they had married, and this was the third wave of a longer study of them. Husbands averaged 34 and wives 31, they had been together around eight years, and they had been living together for about six.

Each partner filled in a long questionnaire at home, separately, with a trained interviewer present and instructions not to compare answers. It took about ninety minutes. Because two people in the same couple do not give independent answers, the analysis nested each person inside their couple rather than treating them as strangers who happened to be paired.

They were asked how much they trusted their partner, and how they saw their partner on three ordinary things: whether they forgave, whether they were reliable, and whether they were getting where they wanted to go in life. The reliability questions are as plain as they sound. One of the three is whether the partner is always on time.

What it found

All three were associated with trust. Reliability was the strongest by a distance.

Being seen as reliable had a standardised coefficient of 0.47 with trust. Forgiveness came in at 0.32, and getting somewhere in life at 0.27. All three were solidly significant.

The authors' actual interest was in what sits underneath. Their argument is that these everyday behaviours are read as evidence of self-control — that when you watch somebody keep a small promise, what you are really registering is that they can make themselves do things — and that it is this inference that produces trust. Perceived self-control was itself strongly associated with trust, at 0.44, and it partly accounted for the link between the behaviours and trust.

That held when they controlled for the obvious alternative explanations, including how committed the partner seemed and how much the person felt appreciated. It survived those controls for reliability and forgiveness.

An honest reading of the limits

This is a snapshot, despite coming from a longitudinal study. Everything in this analysis was measured at the same sitting. So "predicted" here means "was associated with", and the direction is not established. A person who trusts their partner may simply see them as more reliable. Both readings fit these numbers equally well, and the paper cannot separate them.

One person supplied nearly all of it. Perceived reliability, perceived self-control and trust were all reported by the same person at the same time. That arrangement inflates the relationships between them, because somebody in a good frame of mind about their marriage tends to answer everything more warmly. The authors do run a version substituting the partner's own report of their self-control, which helps, and it does not make the problem disappear.

Mediation from a snapshot is a weak inference. Testing whether one variable carries the effect of another requires knowing what came first, and here nothing did. This kind of analysis was standard practice in 2011 and is treated much more cautiously now.

Two of the four experiments rest on a shaky paradigm. The other studies in this paper manipulate the temporary depletion of a person's self-control, using the ego depletion procedure. Since 2011 that effect has been through large multi-laboratory replication attempts that found it close to zero. The married-couples study does not depend on depletion at all, which is why this article is built on that one, but a reader should know that the surrounding package has not aged well.

Narrow sample. Almost entirely Dutch, heterosexual, married, well educated, and recruited through marriage registries. The lab studies elsewhere in the paper are small, at 40, 43 and 48 people.

And reliability is not a complete theory of trust. A coefficient of 0.47 leaves most of the variance unexplained. This says reliability matters more than the other two things measured, not that trust is only reliability.

The counterpoint, which is newer and larger

There is an obvious story to tell from the finding above: trust is a running total, and every kept promise adds to it. A 2025 paper of 847 people gives good reason not to tell it that way.

Its starting point is an old awkwardness in this literature. Descriptive studies have repeatedly found that trust is already high very early — people a few months into a relationship report about as much of it as people who have been together for years. If trust were mainly accumulated evidence, that should not happen.

So these authors asked a different question. Rather than comparing how much trust people had, they compared what trust was made of, in 387 people together six months or less and 460 together five years or more. The broad shape of the construct held across both groups. The details did not: forcing the item loadings to be equal made the model fit significantly worse, and forcing the intercepts to match made it worse again.

The specifics are the interesting part. Items about a partner reacting well when you expose a weakness, and responding lovingly when you bring them a problem, carried much more of the general trust factor for people in long-term relationships than for people in new ones. Their argument is that trust does not grow by addition so much as it gets rebuilt around different material — early on it can rest on something closer to romantic confidence, and later it comes to rest on having been met well at the moments you were most exposed.

Two cautions on this paper too. It is cross-sectional and compares only the two endpoints, with nothing in between six months and five years. Its samples are predominantly white, and it inherits rather than establishes the finding that early trust is already high. There is also a technical irony worth naming: the test that failed here is the one you would need to pass before comparing average trust scores across the two groups, so this paper is not clean evidence that the levels are identical either.

What it does do is take the weight off the accrual story. Reliability tracking trust in a snapshot is one thing. Trust being a ledger that fills up as reliable acts are deposited is a different claim, and it is the one that does not survive contact with people who have only known each other a few months.

What this means for the fight you keep having

The useful thing here is what it does to the size of the problem.

If trust in your relationship has thinned, the conversation usually goes to large questions about character and feeling, which are hard to answer and harder to demonstrate. This finding points somewhere smaller and more testable: the strongest correlate of trust in these marriages was the unglamorous business of doing what you said you would do.

That reframes the fight about being late, or the thing that was supposed to get booked, or the call that did not happen. Those arguments look disproportionate from outside and often feel disproportionate from inside. On this evidence they are not obviously about the errand. Whether somebody can be relied on is the kind of thing partners work out from small, visible material, and this is the material.

What the evidence does not license is the ledger version of that thought, where trust is a balance that rises with each kept promise and falls with each broken one. The 2025 work above is the reason to be careful there. Being reliable looks like part of what trust is made of rather than a deposit into an account, and the difference matters if you have been keeping score.

Two further cautions, because the direction of the arrow is genuinely unknown. Becoming more punctual is not established as a way to manufacture trust, and this study cannot promise that it works. And if you are the one who feels untrusted, the implication is not that you should defend your character better. It is that the evidence a partner is drawing on is probably mundane and cumulative, and it is built from things that are, at least, visible.

Where a session with Elena fits

Elena is an AI coach, and there is no intervention in this research for her to deliver. This is an observational study of what trust travels with.

Where a session is useful is in getting a conversation about trust down to something specific. Couples talking about trust tend to talk in the largest available terms, and the argument stalls because nobody can prove a character trait. Elena is reasonably good at asking which particular things stopped happening, and when, which turns an unanswerable conversation into one with content in it.

Worth saying plainly what this does not cover. If trust broke through a betrayal rather than eroding through a hundred small things, that is a different situation with its own evidence, and our piece on what happened to couples five years after an affair is the better place to start. And if what is being called unreliability is actually a pattern of control — where the accounting only ever runs one way, or where being late becomes a reason to be afraid — that needs a human professional rather than a session.