Why Trust Accumulates Slowly

Trust, but verify.

— A Russian proverb

In the first round, a person repays a debt on time. That may mean reliability. It may mean wanting a second, larger loan. After the tenth round there are fewer interpretations left. Trust is often described as an inner feeling: either you have it or you don’t. In economic terms it is a forecast. One person expects another to meet an obligation even though they could gain in the short term by breaking it. Such a forecast needs data. The evidence base usually grows gradually and can change sharply after an event that reveals a new mechanism.

Psychology distinguishes predictability, dependability and faith. In new relationships and in long-established ones, the overall level of reported trust can be similar; what changes is the basis for the judgement. Early on, willingness to take a risk and faith in the partner do much of the work; later it is the generalisation of diagnostic episodes. What accumulates slowly isn’t necessarily the feeling of trust, but its evidence base.

The last round

In a one-shot game, cheating can pay. If the partners will never meet again, today’s violation doesn’t change tomorrow. In repeated interaction, every action affects future responses. Robert Axelrod ran tournaments of strategies in the repeated prisoner’s dilemma. The simple tit-for-tat strategy started with cooperation, then repeated the partner’s previous move, and turned out to be one of the most successful given that set of opponents and those tournament rules. Its success can’t be transferred to any repeated interaction; it showed that under certain conditions four properties can be combined: don’t defect first, answer defection, forgive quickly, and stay comprehensible.

People like transferring this model to relationships: be kind as long as they are kind to you, and respond symmetrically. In a world without mistakes the advice works better than it does in a real kitchen. People forget, arrive late, mishear and read intention into accident. Punishing every failure can start an endless chain of retaliation even between well-meaning parties. In a noisy environment, strategies with some tolerance work more reliably: distinguish a one-off mistake from a repeating pattern, allow room for repair, but don’t turn forgiveness into a free licence.

One of the best-known trust experiments is set up almost insultingly simply. The first participant receives money and decides how much to pass to a stranger. The amount passed is multiplied several times over. The second participant can then return any part of it — or keep it all. There is no legal obligation, no future meeting, and reputation is hidden. The rational version with fully self-interested participants predicts zero transfer: the second will return nothing, so the first has no reason to send anything.

In the experiment by Joyce Berg, John Dickhaut and Kevin McCabe, people sent money anyway, and many recipients returned part of it. When participants were shown the history of previous players’ behaviour, the link between trust and reciprocity grew stronger. Even a one-off anonymous situation carried a social norm; added history turned a vague expectation into a better-informed bet.

A single round doesn’t show how reliable a person is across different situations. It shows how many different motives hide under the word “trust,” and how quickly willingness to take a risk responds to information about the partner, to the rules and to social expectations. For predicting a relationship, repetition matters more: whether behaviour is consistent across several situations, and whether it changes once there is a chance to exploit someone else’s vulnerability.

Repeated life separates these mechanisms better. Returning a favour only when observed says something about supervision. Keeping a promise when breaking it would be profitable and unnoticed says something about reliability. Admitting a mistake before it is discovered says something about the value placed on one’s own reputation. Trust grows with the number of situations in which behaviour actually distinguished between the possible types; days spent together add little on their own.

At the start of an acquaintance, promises inevitably run ahead of the data. “I’m always here,” “you can rely on me” — these are claims about behaviour in situations that haven’t happened yet. Trust is built through a series of small tests that nobody usually calls tests. A person warns you they’ll be late. Returns a borrowed thing. Doesn’t use an admitted weakness in the next argument. Keeps a boring promise that will earn no admiration. Each episode updates the forecast. A small stake matters because a mistake is still reversible. Entrusting someone with your home, your money or your child’s safety straight away is a fast way to get a lot of information and a poor way to manage risk.

This is the principle of gradually raising commitments. It shows up in lending, insurance and international politics. The resemblance to those problems doesn’t cheapen relationships: incomplete information calls for sequence everywhere.

A major violation can change the forecast quickly, because observations differ in diagnostic strength. A forgotten carton of milk is usually noise; a secret account, systematic lying or the breach of an explicitly agreed boundary reveals a different mechanism and forces a recalculation of past episodes. What follows is a separate task of repair: acknowledge the damage and produce new data through changed behaviour.

Control produces obedience, not trust

Checking a phone, sharing location and imposing bans can reduce the room for violation. They produce observable obedience, but say little about behaviour when nobody is watching. Control replaces part of internal reliability with an external constraint, and deprives the couple of new data about voluntary choice.

Trust requires the option of acting otherwise; without it we are seeing the security of a mechanism, not a forecast about a partner.

Long relationships create a local reputation. Partners learn who remembers the payments, who goes silent in conflict, who comes back to the conversation, who exaggerates threats. This reputation is often more accurate than general personality traits, because it applies to one specific interaction. The danger comes when an old forecast becomes insensitive to new data. “He’s always like that” makes it possible to miss an improvement; “she never would” makes it possible to ignore a violation. Trust should be stable but updatable: full certainty shouldn’t be demanded before a history has accumulated, and an accumulated history can’t be treated as a guarantee that cancels observation.

Reliability matters more than the intensity of promises, repair more than being faultless, and repetition more than one fine episode. Trust accumulates slowly not from a shortage of feeling, but because it has to tell steady behaviour from performance and pattern from noise. It can collapse fast, when a new fact reveals a mechanism that had stayed hidden.

Main sources

Axelrod, R., & Hamilton, W. D. (1981). The evolution of cooperation. Science, 211(4489), 1390–1396; Trivers, R. L. (1971). The evolution of reciprocal altruism. Quarterly Review of Biology, 46(1), 35–57.

Berg, J., Dickhaut, J., & McCabe, K. (1995). Trust, reciprocity, and social history. Games and Economic Behavior, 10(1), 122–142.

Rempel, J. K., Holmes, J. G., & Zanna, M. P. (1985). Trust in close relationships. Journal of Personality and Social Psychology, 49(1), 95–112; Camanto, O. J., & Campbell, L. (2025). Trust in close relationships revisited. Journal of Social and Personal Relationships, 42(9), 2516–2544.