Twenty million copies sold before anyone tested the five love languages

Love languages get discussed as a single idea. As a testable proposition they are three claims stacked on top of each other, and Emily Impett, Haeyoung Gideon Park and Amy Muise set them out precisely in a 2024 review in Current Directions in Psychological Science[3].

First, that each person has one primary love language, a way of receiving care that matters to them more than the other four. Second, that there are five of them, and that five is enough to describe how people give and receive love. Third, that couples do better when partners identify and speak each other's primary language.

Each one depends on the one before it. If people do not have a single dominant preference, there is nothing to match on. If the five categories are not distinct, the quiz is sorting people along a dimension that does not exist. So the claims have to be tested separately, and over the past two decades they have been.

Impett and her colleagues' summary of that literature is that there is remarkably little of it, and that what exists does not strongly support any of the three.

The most direct test arrived in 2025. Sharon Flicker and Flavia Sancier-Barbosa recruited 696 adults, average age 43, most of them married and almost all living with a partner, and asked a narrow question with an unambiguous answer[4].

Does satisfaction with your partner's behaviour in your primary love language predict relationship quality better than satisfaction with their behaviour in the languages you ranked below it?

It does not. A person's top-ranked language was no better at predicting relationship satisfaction, or how loved they felt, than the ones they had placed lower. The ranking carried no extra information.

Two categories did stand out, but not in a way the framework predicts. Words of affirmation and quality time predicted feeling loved and being satisfied better than a person's own primary language did, whatever that primary language happened to be. The useful signal was in which behaviour it was, not in whether it matched the label the person had selected for themselves.

This is the result that does the most structural damage, because the quiz is the entry point to everything else. The whole apparatus assumes that sorting yourself into a category tells you something specific about what you need. On this evidence, the sorting step is close to inert.

Where did five come from? One counsellor's notes. Nobody has since arrived at the same five independently, and the attempts to check the structure statistically have not converged.

The study most often cited in the framework's favour is a fair one. Nichole Egbert and Denise Polk found that the five categories corresponded to relational maintenance behaviours that communication researchers had already identified and measured[5]. That matters: it suggests Chapman was describing something real rather than inventing it. But later factor analyses of love language scales have returned three-factor, four-factor and five-factor solutions depending on the sample and the item wording, which is the pattern you see when the underlying structure is blurry rather than crisp[3].

The most interesting test was not statistical at all. Rudy Pett, Priscilla Lozano and Sarah Varga asked 324 people to describe in their own words how they gave and received love, then coded 648 open-ended responses without forcing them into Chapman's five boxes[6].

Most of what people described did fit. But a sixth category came out of the data that Chapman's list does not contain, which the researchers called checking in: brief, one-directional messages asking how the other person is doing, with no request attached.

That result cuts both ways and is worth sitting with. The categories are not arbitrary, and people really do describe love in roughly these terms. But five is not a closed set. Five is where one person's notes happened to stop.

This is the claim people actually act on, and it is the one where the evidence is genuinely mixed rather than simply negative.

Against it: Selena Bunt and Zoe Hazelwood studied 67 couples and found only limited evidence that alignment on love languages promoted satisfaction[7]. What did predict satisfaction was self-regulation, the capacity to deliberately adjust your own behaviour. Partners who could change how they expressed affection did better than partners who happened to match, which points somewhere quite different from the book's advice. Flicker and Sancier-Barbosa found no advantage for the primary language in a sample ten times the size[4]. Impett and colleagues' reading of the field is that couples matched on primary language are not reliably more satisfied than mismatched ones[3].

For it: Andrew Bland and Kand McQueen clustered the love language profiles of 100 couples and reported that the more congruent a couple's combination of preferences, the less relationship distress they described[8]. And Olha Mostova, Maciej Stolarski and Gerald Matthews studied 100 heterosexual couples and found that mismatch correlated with lower relationship satisfaction and lower sexual satisfaction, for both partners[9].

The Mostova study is the strongest single piece of evidence in the framework's favour and deserves a careful reading rather than a wave. What it measured is not quite what the book recommends. Their mismatch score was the gap between what one partner preferred to receive and what the other partner actually tended to express. That is not the question of whether two people share a primary language. It is the question of whether your partner does the things you want. Those two come apart, and the second would be expected to predict satisfaction under almost any theory of relationships.

Its limits are the ones the authors state plainly. The design is cross-sectional and everything is self-reported, so the arrow may run backwards: people who are happy with their partners tend to rate them as more attuned. A hundred couples answering a questionnaire on one occasion cannot separate those two stories.

Here is the part the framework gets right, and it is not a consolation prize.

Chapman noticed something true. Partners routinely express care in a form the other does not register, and both then quietly conclude they are not loved. One person is paying bills and fixing things and reads that as devotion. The other is waiting to be told, in words, that they matter. Neither is withholding anything. The signal is being sent and not received, and the resulting hurt is completely genuine on both sides.

Relationship science has a name for the underlying variable, and it is not a taxonomy. It is perceived partner responsiveness: the degree to which you believe your partner understands what matters to you, values it, and acts on it. It is among the most reliably predictive constructs in the field. Emre Selcuk and colleagues followed a national American sample for ten years and found that partner responsiveness predicted a later rise in eudaimonic wellbeing, meaning a sense of purpose and self-realisation rather than moment-to-moment mood, and that it did so partly by reducing how hard everyday stress landed[10]. And in the largest pooled analysis of relationship predictors yet assembled, covering 11,196 couples, feeling appreciated by a partner ranked among the strongest predictors of relationship quality[11].

Notice what responsiveness is not. It is not a category you belong to. It is not fixed. What someone needs on a bad Tuesday is not what they need on holiday, and responsiveness is the business of tracking that difference, which is precisely the thing a permanent label makes harder.

Impett and colleagues suggest retiring the language metaphor and replacing it with a nutritional one[3]. Relationships need several things at once, the way a diet does, and no single one substitutes for the rest. Their argument is that the metaphor determines the behaviour. "Language" implies a native tongue you were born into and a second one you might laboriously learn. "Nutrient" implies you need all of them, in varying amounts, forever.

No, and the honest answer is more useful than either of the confident ones.

What the framework does well is supply a vocabulary and a licence to use it. "I feel more loved when you plan things for us" is an enormously more workable sentence than "you don't care about me," and most couples do not get to the first sentence unaided. Anything that reliably produces it has done real work, whatever its factor structure looks like.

What to put down is the machinery bolted on around it. You probably do not have one language; you value all five and the ranking does not predict much[4]. The list is not exhaustive, so a category that fits you badly is a limitation of the list rather than something odd about you[6]. Your partner scoring differently is not a compatibility problem and certainly not a diagnosis. And knowing each other's results is not the intervention. Doing the thing is.

There is also a specific failure mode worth naming, because it is common. A love language can become a way of declining to be responsive: I am an acts of service person, so I am not going to say the thing you have asked me to say. That inverts the entire point. The framework exists to help people notice what their partner needs, not to issue exemptions from providing it.

Used as a conversation starter, it is a good one. Used as a personality test, it sorts people along a dimension that repeated attempts have failed to locate.