How do you know if your partner is cheating?

The largest synthesis of deception-detection research pooled hundreds of studies and found average accuracy of around 54 percent, where 50 would be guessing[1]. That figure barely moves for people who know the liar well, and it does not improve with confidence: individuals who are certain they can tell are not measurably better than those who are not. Familiarity helps less than expected partly because it works both ways, since a long-term partner also knows precisely which explanations you accept. There is a related asymmetry that makes the everyday situation worse than the laboratory one. In a study, roughly half the statements are lies. In a relationship the great majority of what a partner says is true, so most of the suspicion you generate is aimed at ordinary behaviour, and each false alarm costs something real.

The standard checklist items, avoiding eye contact, guarding a phone, becoming defensive, sudden attention to appearance, come from intuition rather than evidence. The comprehensive review of behavioural cues to deception found that very few cues are reliably associated with lying, and that those which are tend to be weak and inconsistent across situations[2]. Nothing resembling a tell that a person can read in real time survives that literature. There is a further problem specific to checklists, which is that every item has a much more common innocent cause. Phone privacy is now normal behaviour. Working late is usually work. A new interest in the gym is usually a new interest in the gym. A list assembled from behaviours that are individually unremarkable will match almost any relationship if you are already looking, which is why reading one at two in the morning feels like confirmation and is not.

This is the part that rarely gets said, and it is not a research finding so much as arithmetic applied to one. Whatever figure you accept for how many people are ever unfaithful, the proportion who are unfaithful right now, in this particular month, is much smaller. That means the starting probability for any given worry is low. When a weak signal is applied to an uncommon event, most of the positives it produces are false ones, and this holds no matter how strongly the signal feels like proof in the moment. It is the same reason a screening test with good accuracy still produces mostly false alarms for a rare condition. Practically, it means a single ambiguous observation should move your belief very little, and that the feeling of certainty which follows a night of pattern-matching is not information about your partner. It is information about how long you spent looking.

There is decent work on what predicts infidelity at population level, including relationship dissatisfaction, certain personality dimensions, sexual compatibility and prior history[4]. It is tempting to read that as a diagnostic checklist and it does not function as one. A predictor that raises the likelihood across thousands of people can still be present in a faithful relationship and absent in an unfaithful one. Using population-level predictors on a single person is the same error as reading the signs lists, wearing better clothes. The useful thing these findings offer is different: they describe conditions that make infidelity more likely, which is information about the state of a relationship and worth attending to regardless of whether anything has happened.

The obvious response to uncertainty is to look for evidence, and the research on partner surveillance suggests this rarely lands where people hope. Studies of interpersonal electronic surveillance find it associated with jealousy and relational uncertainty rather than with their resolution[3]. The mechanism is not mysterious. Surveillance produces ambiguous material, ambiguous material requires interpretation, and an anxious person interprets anxiously, which generates the need for more looking. It also changes the relationship independently of what it finds. A partner who discovers they have been investigated is dealing with a breach of their own, and if nothing was happening you have now created the first real problem in the sequence. This is separate from whether it is ever justified, which is a different question with its own answer.

Three things are available even though certainty is not. You can ask directly, accepting that you may be lied to, because a direct question at least puts the matter on the record and changes what continued concealment requires. You can attend to the relationship rather than the evidence: whether you are still being told things, whether affection has changed, whether you are being avoided. Those are observable and they matter whether or not anyone is unfaithful. And you can decide in advance what you would do under continued uncertainty, which is the state most people in this position actually remain in. Deciding that is more useful than another hour of searching, because it is the only part of the situation you control.

Nothing here means suspicion is always misplaced. People are sometimes unfaithful, partners sometimes do sense it, and being told that detection is unreliable is not the same as being told you are wrong. The claim is narrower and it is about evidence rather than truth: a feeling of certainty produced by observing behaviour is not reliable grounds for a conclusion, whichever way it points. That cuts against reassurance as much as against alarm, since the same research implies you also cannot verify that nothing is happening. There is a second asymmetry worth naming. Most of what gets called intuition in this context is pattern-matching against a story you have already started telling, and it feels identical from the inside whether the story is right or wrong. The people who turned out to be correct and the people who put a good relationship through months of interrogation had the same experience at the time. That is precisely why the feeling cannot be used as the deciding evidence, and why what you do next should depend on what you can live with rather than on how sure you feel.