Do the Five Love Languages Work? What the Studies Actually Measured
In the preregistered test, fewer than half of people had an identifiable primary love language. The quiz is zero-sum, and that is most of the problem.
Before the article. A love language is a description of a preference, not a bill one partner can present to the other. When it is used as a rule — you owe me this, you clearly do not love me because you keep failing to speak it — it has the shape that the UK statutory guidance on controlling or coercive behaviour describes as making and enforcing rules the other person must follow, with penalties for falling short, and pressure that continues after a no is coercion rather than a mismatch of dialects. Find A Helpline, in the sources at the bottom, lists free and confidential services by country. In the United States the National Domestic Violence Hotline is 1-800-799-7233, or text START to 88788, 24 hours a day.
With that said. Gary Chapman's The 5 Love Languages has sold over 20 million copies, and almost nobody meets its argument as an argument. They meet the quiz: four minutes, thirty paired statements, a result, a label. This page is mostly about what that quiz can and cannot measure, because that turns out to be where the scientific trouble starts — not in the idea that people differ in how they like to be loved, which nobody disputes.
The short version of what the research shows. The quiz is forced-choice: every point one language gains, another loses, so it produces a ranking rather than a measurement. When researchers put that ranking against the same person's independent ratings of all five, it did not predict them — in 83 couples, the forced-choice answer matched the person's own highest-rated language six times out of ten comparisons. Depending on which instrument you use, receiving gifts is the primary language of between zero and four per cent of people, or of more than half. In a preregistered study of 499 people, fewer than half had an identifiable primary love language at all. A bottom-up rebuild across three samples found that solutions with seven to ten languages fit better than five. Matching has failed in every well-powered test, up to and including 954 couples. And the behaviours that keep predicting relationship quality — verbal affirmation, undivided time, support for a partner's own goals, showing up when things are hard, owning it when you get it wrong — are not a channel anybody has to be assigned.
So this is not a debunking and not a defence. It is sixteen sections on what has actually been measured, which page of results reports it, and what the numbers look like when you read the papers instead of the summaries. Our editorial policy explains why every empirical claim here carries a linked source.
Before the rest of this page: when speaking their language becomes an obligation
Almost everyone who looks this up is looking it up in good faith, usually about themselves — whether the quiz result means anything, whether they are failing at something, whether it explains why two reasonable people keep missing each other. That is an ordinary question and the rest of this page is about what the research can and cannot say about it.
One thing has to go first, because the framework has a second life that the research literature never discusses. A love language is a description of a preference. It is not an entitlement, and it turns into something else the moment it is used as a bill: my love language is physical touch, so you owe me this; you clearly do not love me, because you keep failing to speak it. Stated that way, a preference becomes a rule that one person sets and the other has to satisfy, with a penalty for falling short.
That structure has a name outside relationship psychology. The UK statutory guidance on controlling or coercive behaviour lists, among controlling behaviours, making and enforcing rules the other person is expected to follow, and using punishments to secure compliance. Nothing in the description requires anyone to have meant harm. A requirement about affection, sex, money, or how someone must spend their time does not become reasonable because it arrived with a quiz result attached — and pressure that continues after a no is coercion rather than a mismatch of dialects.
If that is closer to your situation than to your curiosity, the helplines are free and take about a minute. Find A Helpline, in the sources at the foot of this page, lists free, confidential services by country. In the United States, the National Domestic Violence Hotline is 1-800-799-7233, or text START to 88788, 24 hours a day. The same distinction, with the research behind it, is set out in what monitoring a partner actually does.
With that said: what follows is the measurement story, and it is more interesting than either side of the usual argument.
What Chapman actually claims, and where most people meet the claim
Gary Chapman's The 5 Love Languages was published in 1992 by a marriage counsellor describing patterns he had seen in his own counselling. Its claim has three parts, and the review literature treats them separately because they can fail separately.
First, each person has one primary love language — the channel they rely on most to express love and to feel it. Second, there are five of them: words of affirmation, quality time, receiving gifts, acts of service, physical touch. Third, most relationship trouble comes from partners speaking different ones, so the fix is to learn and speak your partner's. The strength of the third claim is easy to underestimate. Chapman writes that the five are as different as Chinese from English, and the governing metaphor is a love tank that fills only when the right channel is used.
The scale of the thing matters for one reason: it is the reason the measurement question is not academic. When Denise Polk and Nichole Egbert tested the claims in 2013, they recorded sales of over seven million copies. The publisher's own quiz page, read on 4 November 2026, now says over 20 million copies sold, in what the 2024 scientific review puts at 50 languages, with the review also citing a figure of 30 million people having taken the quiz and a $20 million Australian government relationship-education programme built partly on the framework.
And this is the part worth holding on to. Hardly anyone meets the three claims as claims. They meet the quiz: four minutes, a result, a label. That quiz is a specific instrument with a specific design, it is proprietary — the official page warns readers to beware of unauthorized quizzes, tests, imitations and points them to an authorised premium assessment — and its design turns out to be where almost all of the scientific trouble starts. Not the idea. The instrument. The rest of this page is mostly about that difference, which is the same difference that makes our own attachment style test show two continuous scores instead of handing out one of four labels.
The first page of results, opened on 4 November 2026
This site audits the pages it competes with rather than guessing at them, so here is the measurement, with the method stated.
Google returned its unusual-traffic challenge to the headless browser used here (IP logged at 14:40 UTC), which is the thirteenth time that route has closed; captchas are not solved here, so the page one measured is DuckDuckGo's for do the five love languages work, United States. In order: Simply Psychology, a Psychology Today post titled Love Languages: What Science Actually Says About Relationships, a second Simply Psychology page on a second domain, a therapist's blog, Health.com, a second Psychology Today post (The Five Love Languages Are for Everyone), Cosmopolitan, two small psychology sites, and Verywell Mind.
Seven of the ten were downloaded and searched term by term. The result is not the one this genre usually produces, and it is only fair to report it: the strongest page is genuinely well sourced. Simply Psychology cites Flicker and Sancier-Barbosa, Bland and McQueen, Bunt and Hazelwood, Egbert and Polk, Hughes and Camden, Mostova, Surijah and the 2024 review, gives three real sample sizes in its body text (696, 499, 67), and is the only page of the seven that uses the word preregistered. Verywell Mind carries the right references too. One page labels itself an AI-assisted draft with an editorial spot review, which is more disclosure than most of the genre offers and is the kind of thing our own AI disclosure exists to say.
What none of the seven has is narrower and sharper, and it is the whole of this article:
- Not one of them contains the words forced-choice or zero-sum — the design feature of the quiz, and the reason the studies disagree with each other.
- Not one mentions the largest test ever run on the matching claim: 954 couples.
- Not one mentions the bottom-up study that asked how many love languages there are, and answered between seven and ten.
- Not one mentions the sixth love language that a separate team found in 2022, or the single randomised trial that exists.
- Not one reports the correlations between the five, which are the numbers that decide whether as different as Chinese from English can be true.
The Psychology Today piece titled What Science Actually Says is the sharpest case: it names no study, no author and no year anywhere in its text. That is the same pattern logged for this genre's invented numerical rules and for the divorce percentages that their own author retracted — except that here the sources are not missing. They are cited and then not opened.
The literature, counted on 4 November 2026
Before any argument about what the studies show, here is how many studies there are. The queries are written out so the count can be re-run rather than taken on trust, and they were run in more than one database on purpose, because one database was not enough the last two times this site counted a literature.
In Europe PMC, `"love languages"` returns 51 records and `"five love languages"` returns 22. In PubMed, `"love languages"` returns 9. For scale: the same Europe PMC query for breadcrumbing, a piece of dating vocabulary barely a decade old, returns 19, and the equivalent for micro-cheating returns 1. So the love languages have more literature than the vocabulary terms this site has covered recently, and vastly less than their reach would suggest. The 2024 review in Current Directions in Psychological Science found ten studies worth evaluating against a book that has sold more copies than most psychology textbooks have readers.
Two things follow, and they pull in opposite directions.
The first is that nobody should talk about this evidence base as though it were settled in either direction. Ten to twenty studies, mostly cross-sectional surveys, mostly self-report, is thin. A sentence like science has debunked the love languages is doing more work than the record supports, and so is studies show that matching makes couples happier.
The second is that the record is not empty, and the useful parts of it are specific. Three separate research groups have now put the framework's own instrument against continuous measures of the same preferences, in samples of 83 couples, 250 couples, 499 individuals and 954 couples, and the result is consistent enough to say something firm. What it is firm about is not whether words of affirmation matter. It is whether the quiz can find out which one is yours.
The quiz is zero-sum, and that is most of the problem
Chapman's instrument is the Love Language Personal Profile: 30 forced-choice items. Each item offers two statements and asks which is more meaningful. We hold hands against My loved one gives me a gift. You pick one. Your score for each language is the number of times it won.
Read that structure carefully, because everything downstream depends on it. The format cannot record that both options matter to you. It cannot record that both matter enormously, or that neither does. Every point one language gains is a point another one loses, which is what zero-sum means here. The output is therefore not a measurement of how much you value each channel. It is a ranking, and a ranking manufactured under a constraint that does not exist in life, where a partner can hold your hand and bring you something.
So researchers built Likert-type scales instead — rate each of the five independently, usually one to five. Egbert and Polk's 20-item Love Language Scale, four items per language, is the one most of the literature uses. And the moment you measure that way, the first thing you see is that people endorse all five. Across study after study the ratings are skewed high, with average scores hovering around 4 out of 5 for every one of the five languages. Which is unsurprising and important: asked independently, almost nobody says that compliments, time, help, touch or gifts are unimportant to them.
This is not a new problem in relationship research and it has a direct precedent. Attachment was first measured the same way — Hazan and Shaver's original test asked people to pick one of three descriptions — and the field moved to continuous dimensions for exactly this reason, because a forced choice compresses two independent quantities into one label and then the label gets treated as a type. Polk and Egbert cite that history explicitly as the motivation for their own study. The consequences of category thinking in attachment are set out in what the evidence says about attachment styles changing, and the same arithmetic — that a cut-off makes 3.9 and 4.1 look like different people — is spelled out on the attachment test's own page.
The strongest objection to everything in this section has to be stated, because it comes from inside the literature and it cuts the other way. Andrew Bland and Kand McQueen argued in 2018 that the real measurement problem is the opposite one: most researchers did not use Chapman's own inventory, built their own scales instead, and those scales drift from the model they claim to test — some of them with weak reliability coefficients. On that reading, the mixed results are an artefact of the substitutes rather than of the original. It is a fair point and it has one awkward implication for its own side: the study that did use Chapman's profile, on 981 people, is also the study whose results say that every one of the five languages predicted love and satisfaction, which is not what a model built on one primary channel predicts either.
The one study that put the quiz against the scale, and what it found
This is the best number in the whole topic and not one page on the first page of results reports it.
Denise Polk and Nichole Egbert recruited 83 couples, had each person answer a forced-choice question about the love language they most want to receive, and had the same person rate all five on the 20-item continuous scale. Then they asked the obvious question: does your forced-choice answer predict which of the five you score highest on?
It does not. Running one-way ANOVAs for each partner on each of the five dimensions produced no significant differences anywhere. A mixed-model MANOVA, with the forced-choice answer as the between-subjects factor and the five continuous subscales as outcomes, produced F(4, 160) = .49 for words, .36 for quality time, .48 for gifts, .51 for acts of service and .31 for touch — every one of them non-significant. In their words: stating a preference for touch in the forced-choice question did not correspond with higher scores on the touch subscale than people who had named something else.
The descriptive version is easier to feel. Across ten comparisons — five languages, two partner roles — the forced-choice answer lined up with the person's own highest continuous mean only six times (four of five for men, two of five for women). Among the men who picked physical touch on the quiz, the language they actually scored highest on was quality time.
And there is a stranger result in the same paper that deserves to be better known. Collapsing couples into matched, mismatched and partially matched, the significant finding was that matched and mismatched couples' reports of relational quality were closer together than partially matched couples' — which is not the pattern anyone's theory predicts, and is the kind of result that usually means the grouping variable is noisy. Every other test in the paper, the authors write, provided little empirical support for Chapman's notions.
The practical translation is unglamorous: the four-minute result is a snapshot of how you ranked thirty arbitrary pairings on one afternoon, not a stable fact about you. Two people taking it in different moods are not necessarily different types.
Gifts: between zero and four per cent, or more than half
There is a cleaner way to see the same problem, and it is the single most convincing number in the 2024 review, which assembled it by comparing results across studies rather than running anything new.
When samples are categorised by Chapman's forced-choice quiz, the proportion of people whose primary love language comes out as receiving gifts runs from zero to four per cent. When the same construct is assigned by whichever of the five a person scores highest on using a continuous rating scale, gifts comes out as the top language for more than half the sample.
Nothing about the people changed. Only the instrument changed, and the answer moved by fifty points.
The explanation is not mysterious once you have the zero-sum point. Asked to choose between a gift and an evening of undivided attention, most people choose the evening, because that is the comparison the forced-choice format imposes and because attention is the thing you can have more of. Asked to rate, separately, how much a thoughtful present means to you, most people rate it high, because it does. One question measures what you would trade away; the other measures what you value. The genre reports both as your love language and they are not the same quantity.
Polk and Egbert's sample behaves exactly this way: most of them were categorised as quality time or physical touch by the forced choice, while gifts had the highest mean on the rating scale. And it is worth noticing which direction the error runs in public. The quiz systematically under-detects gifts, so the least fashionable of the five — the one people are embarrassed to claim, the one that gets jokes made about it — is the one the instrument is least able to see.
This is the same failure mode as a cut-off that sorts a continuous score into a type, which is why this site publishes test results as numbers on dimensions and says so on every test page, including the conflict style test and the relationship satisfaction test.
Fewer than half of people have a primary love language at all
In September 2025 the question got the design it had been missing: a preregistered test, with the hypotheses filed before the data were collected.
Sharon Flicker, Flavia Sancier-Barbosa and Emily Impett studied 499 individuals in long-term, cohabiting relationships and asked two things. Do people have an identifiable primary love language? And if they do, does satisfaction with their partner's expression of that behaviour predict relationship quality better than satisfaction with the other four?
Both answers were no. Less than half of the participants had an identifiable primary love language — not chose a different one than expected, but had no single channel that stood out from their own ratings of the others. And among those who did, satisfaction with a partner's expression of it was no stronger a predictor of relationship quality or perceived partner love than satisfaction with any other loving behaviour. What predicted relationship quality was satisfaction across a wider range of behaviours.
That result replicated the same group's earlier test. In the 2024 study, 696 people recruited through CloudResearch — mean age 43.8, 70 per cent married, 97.3 per cent living with their partner — satisfaction with a partner's primary-language behaviour again predicted no better than satisfaction with the lower-ranked languages. The paper's own conclusion is blunt: Chapman's hypothesis was not supported.
Two honest notes about reading these. Both papers are closed access: the 2025 one has no open copy anywhere, and the 2024 one is free on the publisher's site but behind a bot check that would not open here. What is quoted above is from their abstracts, read on the publisher's record and on PubMed, which is why no effect sizes from either appear on this page. And neither study is a couple-level design: these are individuals reporting on their partners, with the usual limits that carries.
Still, the first claim is the one everything else rests on. If fewer than half of people have one primary language, then the instruction to find out your partner's cannot be followed for most couples, however willing both people are.
How many love languages are there? Between seven and ten
The 2024 review made a specific complaint: the five categories were derived top-down, from one counsellor's descriptions, and testing them properly would require a bottom-up approach — asking people what they actually do, then seeing how the answers group.
In September 2025 the same senior author co-authored exactly that study, and it is the most interesting paper in the topic. Flicker, Sancier-Barbosa, Clemons-Castanos, Field, Jackson-Zambon, Phelan and Impett ran three studies on people in long-term relationships in the United States — N = 696, 500 and 499 — rebuilding a love-languages measure from the items up with exploratory and then confirmatory factor analysis.
Their finding is not that the love languages are fictional. It is that there are more of them than five. Solutions ranging from seven to ten love languages fitted the data better than the five-factor solution and predicted relationship quality better. Their conclusion is that Chapman's model is overly narrow, and that a wider vocabulary would help therapists understand what clients are actually missing.
That reframes the whole debate, and it is why have the love languages been debunked? is the wrong question. The category system is not wrong in kind; it is too coarse. Five bins were drawn by hand in 1992 and the data want more of them. Earlier factor analyses had already hinted at this without converging: the published structures include three factors, four factors and five, and the 250-couple Indonesian study that used both a rating scale and a forced-choice scale reported that the five-factor solution was not supported and that love-language compatibility did not affect marital satisfaction.
Fairness requires the other side of the ledger too. Egbert and Polk's original 2006 validation ran a confirmatory factor analysis and found the five-factor fit superior to one-, three- and four-factor alternatives — so the five did survive the first serious test of their structure. What has happened since is that larger, better-designed attempts keep finding more factors rather than fewer, which is a different claim from the five do not exist.
The sixth love language turned out to live in your phone
Three years before the seven-to-ten result, a separate team ran the bottom-up test the other way — qualitatively, from people's own words — and nobody on the first page of results mentions what they found.
Rudy Pett, Priscilla Lozano and Sarah Varga collected 648 open-ended responses from 324 people in relationships, mean age 20.17, with relationships averaging about seventeen months. Four prompts: how you communicate love, a recent example, how your partner communicates love, a recent example. Then three coders worked through the responses using Chapman's own definitions, with reliability checks.
The first result is the strongest piece of pro-Chapman evidence in the literature, and it should not be skipped because it is inconvenient: all five categories came through clearly. Words of affirmation appeared in 296 responses (45.7%), quality time in 215 (33.2%), acts of service in 164 (25.3%), and gifts and physical touch in 106 each (16.4%), with coder agreement from κ = .66 to κ = .94. The authors describe this as evidence confirming the validity of the typology.
The second result is the one worth the page. A sixth theme emerged that none of Chapman's five covered, and they named it check-ins: brief, asynchronous, partner-centred messages asking whether you are all right. A good morning text. Texts to make sure that I am awake, eating, getting home safe. Texted me after my test to see how it went. It appeared in 100 responses (15.4%) — nearly as often as gifts or physical touch — with κ = .68 on a separate reliability check of 82 responses. The authors argue it is distinct from quality time on three grounds: it is momentary rather than immersive, it is information-seeking, and it runs one way rather than as a shared interaction. And the majority of these check-ins happened over text messaging.
That is a 1992 taxonomy meeting the medium most relationships now run on. It also lands on something this site has measured from the other direction: what travels well through a thin channel and what does not, in the research on texting and what it actually distorts. One more detail, unexplained by anything in the paper: check-ins were the only theme more common among men (25.5%) than women (14.1%).
Matching: every properly sized test, and the one that found something
The third claim — that couples do better when partners speak each other's primary language — is the one people actually act on, and it is the one that has been tested most. Laid out by sample size, the record is remarkably consistent.
- 83 couples (Polk and Egbert, 2013): little empirical support; the significant grouping result ran in a direction no theory predicts.
- 67 heterosexual couples (Bunt and Hazelwood, 2017): limited evidence for alignment, and self-regulation contributed more variance in satisfaction than alignment did — the ability to manage one's own behaviour mattered more than the match.
- A cluster analysis of couples' love-language distributions (Bland and McQueen, 2018): no support either for the claim that matched couples report higher satisfaction.
- 250 couples (Surijah and colleagues, 2021): love-language compatibility did not affect marital satisfaction.
- 696 individuals (Flicker and Sancier-Barbosa, 2024): the primary language predicted no better than the others.
- 954 couples (Chopik and colleagues, 2025): the largest test ever run, and the most thorough. Mean age 55.42, mean relationship length 27.39 years. They tried difference scores, response surface analysis and machine learning across all possible pairings of languages, and found virtually no evidence that matching explained additional variation in outcomes above and beyond each partner's personality traits and attachment orientations.
Against that stands one study that did find matching effects, and it has to be reported properly rather than waved away. Olha Mostova, Maciej Stolarski and Gerald Matthews surveyed 100 heterosexual couples — 31 nationalities, relationships from six months to 24 years — and measured, for each person, the gap between the way they wanted to receive love and the way their partner expressed it, summed across all five languages. Smaller gaps went with higher relationship satisfaction and higher sexual satisfaction. The authors are careful: the design cannot establish direction, and they say so, suggesting future work test whether matching causes satisfaction, results from it, or shares a cause with it.
And notice what their measure is. It is not do you two have the same primary language. It is a continuous, five-channel discrepancy sum — which is also a measure of how much expressing your partner does overall, since somebody who expresses a lot of everything will show small gaps on everything. That is the review's central objection to the positive findings: they may be detecting the main effect of being loved attentively rather than any effect of matching. Two of their incidental results are worth keeping: love-language mismatch correlated r = .01 with relationship length in both sexes, so couples do not converge over time; and women scored higher than men on four of the five receiving scales, with the largest gaps for quality time and words of affirmation.
The null here is an instance of a wider one. Matching partners on a measured characteristic — any measured characteristic — has a thin record: across 23,250 married people in three countries, how similar two partners were on personality explained less than half a per cent of their relationship satisfaction once each person's own traits were in the model. What the research on whether opposites attract actually reports goes through that result, the one study that disagrees with it, and how large the resemblance between real partners turns out to be.
Why the five correlate too highly to be separate languages
There is a quieter number that decides the as different as Chinese from English claim, and no page on the first results page gives it.
If the five languages were genuinely distinct channels — if being a touch person meant something different from being a words person — then a person's ratings of them would be loosely related at best. Someone high on one might be low on another. That is what distinct means statistically.
They are not loosely related. In Egbert and Polk's validation data the correlations among people's ratings of all five run r = .54 to .75. In Chopik's 954-couple study the highest correlation between two of the languages reached r = .795 for expression and r = .774 for preferences, which the authors describe, in their own word, as problematically high.
A correlation of .79 between two supposedly separate constructs means they are largely the same thing measured twice. The plain reading is that what the five items are picking up, most of the time, is a single underlying disposition — something like how much affection is flowing in this relationship, or how responsive and attentive this partner is in general. Chopik's team draw exactly that conclusion: it is worth revisiting whether love languages tap distinct relationship preferences at all, or whether they reflect broader responsiveness and positive relationship behaviour regardless of matching.
That is a much better fit with what the wider literature already says. The strongest organising idea in close-relationships research is not channel-matching but perceived partner responsiveness: the sense that the other person understood what you said, valued it, and cared. The evidence for that construct is set out in the research on how people actually open up, and the reason matching is a weak bet in general — not just here — is visible in the neighbouring literatures too, where matching on conflict style or on language style turns out to help in some contexts and not others rather than uniformly. The conflict version of that is in what actually predicts recovery after a fight.
What did predict, in the studies that stopped asking people to rank
The most useful thing in this literature is not the demolition. It is what came out the other side when researchers stopped asking people to pick a favourite and simply asked which behaviours predicted relationship quality.
In the 2024 study of 696 people, two of the five did better than a person's own primary language at predicting both perceived partner love and relationship satisfaction: words of affirmation and quality time. In the preregistered 2025 study of 499 people, with the expanded item set, the robust predictors were verbal affirmations, encouragement for individual pursuits, support during difficult times, and accountability.
Look at that second list again. Two of the four are not love languages. Encouragement for individual pursuits is support for a partner's own goals and autonomy — the thing the 2024 review specifically pointed out was missing from a framework built on a homogeneous sample of married, religious, mixed-sex couples who probably shared traditional values. Accountability — owning it when you get something wrong — is not on Chapman's list either, and it is closer to repair than to affection.
There is also an ordering that holds across studies which disagree about everything else, and it is offered here as a reading of three sets of numbers rather than as any single paper's finding. In Hughes and Camden's 981 participants — the one study in this set with substantial lesbian and gay male samples, and the one that used Chapman's own profile — the perception that a partner used your preferred language well predicted love and satisfaction in every case, but not equally: words of affirmation came top (R² = .26 for love, .32 for satisfaction) and gifts came bottom (.17 and .11). The same two languages top the list in the 2024 and 2025 studies that found against matching. So the pro-matching study and the anti-matching studies agree on the ranking even while disagreeing about the mechanism.
Which produces an oddity worth sitting with: gifts is the language the rating scales score highest and the one that predicts least. Those are not contradictory. People genuinely value tokens of thought, and valuing them does not make them the load-bearing part of a relationship. The behaviours that carry weight look a lot like the ones in the research on couples who stay together and in what builds attraction over a long stretch of time.
The only randomised trial that exists, and what it tested
Every section above is correlational. Somebody had to try the obvious experiment — take couples, teach half of them the framework, and see what happens — and as of today exactly one randomised controlled trial of a love-languages intervention appears in the indexed literature.
Guanhua Chen and colleagues published it in BMC Psychology in July 2025. Forty couples: twenty in China, twenty in Malaysia, randomly assigned by a computer-generated sequence administered by an assistant outside the intervention, with ten couples in the treatment arm and ten in the control arm in each country. Couples already in counselling were excluded. The treatment was a Love Language Game: a structured card-based programme adapted from a commercial game, in which partners identify and practise each other's preferred languages. The control arm got a different relationship-enhancement workshop series.
The groups were comparable at baseline (t = −0.464, p = .644). Afterwards, relationship satisfaction rose in all four of the treatment subgroups, by between 0.11 and 0.61 scale points, while the control subgroups moved by −0.07 to 0.24. The two-factor ANOVA gave a significant pre-post effect (F = 33.73, p < .001, η² = .32) with the treatment group improving more. The authors' own limitations are the right ones: the sample is small, and the follow-up was brief enough that nothing can be said about whether the gains lasted.
So: a real trial, a real difference, and almost no weight. Ten couples per cell, a single post-test, four subgroups carved out of eighty people, and no measurement of whether the mechanism had anything to do with matching. The control group also did a structured relationship activity and also mostly improved, which is the comparison that matters and which points the same way as everything above: what a love-languages exercise reliably delivers is a structured conversation about what each person wants — two partners sitting down, with a prompt, to say what they need and hear what the other needs. That is a real intervention with a real literature behind it, and it does not require the five categories to be true.
It is also, almost certainly, why so many people say the book helped their relationship. The review says the same thing in its own voice: the framework gives partners an occasion to reflect on, discuss and respond to each other's needs.
The balanced diet, and why a better metaphor is still only a metaphor
Having concluded that the three claims do not hold, the 2024 review does something unusual for a critique: it offers a replacement, because it accepts that the public will use some metaphor and would rather the metaphor were accurate.
Their proposal is a nutritionally balanced diet. A language is something you either speak or do not, which is why the original metaphor implies that the wrong channel is wasted effort. A diet implies something different: people need a range of nutrients, you can survive missing one, and the best state of health comes from having them all. So the advice becomes make sure your relationship is nutritionally balanced, and if something feels missing, say which thing — rather than identify your partner's single channel and commit to it.
The authors fill in the nutrients with the research behind each: time together, expressed appreciation, affectionate touch, practical help and support, being made to feel special — plus support for a partner's own goals and autonomy, which is not on Chapman's list at all. They allow for context, which the language metaphor cannot: physical affection matters more under stress, and partners higher in attachment avoidance, who find it hardest to trust that they are cared for, benefit more than others when a partner expresses gratitude. That last result, from five studies by Park, Impett, MacDonald and Lemay, carries a detail that belongs on this page in particular: what buffered insecurely attached people's daily dissatisfaction was perceived gratitude, not the partner's own report of having expressed it. The channel is not the unit of analysis. What landed is. That is the case for tailoring without the case for typing, and it sits beside what this site has reported on supporting a partner who pulls away and on supporting an anxious one.
And then the honest caveat, which the review does not quite make: a better metaphor is not a finding. The balanced diet has not been tested as an intervention, there is no diet quiz, and no trial has compared try to balance all five against learn your partner's one. It is a more defensible summary of the existing evidence, which is a genuine improvement and a different thing from being proven.
The nearest thing to a measured version is older and less catchy. Laura Stafford's revised relationship maintenance scale, built from what people say they actually do to keep a relationship satisfying, resolves into seven behaviours: positivity, openness, assurances, social networking, task sharing, conflict management and advice. Three of those have no equivalent among the love languages, and two of them — managing conflict, and bringing a partner into your social world — are the ones this site keeps finding in the data on relationships that last.
What none of this can tell you
The limits go in the body rather than in a footnote, because the most common use of a page like this one is to settle an argument it cannot settle.
Almost all of it is self-report, and most of it is cross-sectional. One questionnaire, one sitting, with people describing their own preferences and their partner's behaviour. Nothing in that design separates my partner expresses love in the way I like from I am happy with my partner, so I rate everything they do generously.
The largest study is a conference abstract. The 954-couple test is published in a conference supplement, so beyond the correlations quoted above there are no effect sizes to report and no full methods section to check. It is the most thorough test in existence and the least complete publication in this list, and both of those are true at once.
Two of the key papers could not be read in full. The preregistered 2025 study has no open copy and the 2024 one is behind a bot check; everything quoted from them is from their abstracts, which is why no effect sizes from either appear here. The 2018 cluster-analysis paper is closed too, and is cited on this page only for an argument it makes, not for any number.
The samples are narrow and mostly mixed-sex. Polish and multinational online volunteers, Indonesian couples, American crowdworkers, American undergraduates, older American couples. Four of the five matching studies are mixed-sex only. The one study with substantial lesbian and gay male samples is also the one that found positive results, which nobody has followed up.
And there is almost nothing on doing. One small trial, no follow-up, no test of the mechanism. Every page on this subject, this one included, ends with something that sounds like advice. The advice is an inference from correlations, offered as one.
What survives is narrow and worth having. The quiz produces a ranking, not a measurement, and the ranking does not reproduce the same person's own ratings. Fewer than half of people have one primary channel. There are probably more than five channels. Matching has failed in every well-powered test. And the behaviours that keep predicting relationship quality — telling someone plainly what you value about them, giving them undivided time, backing their own ambitions, showing up when things are hard, owning it when you get it wrong — are available to anyone regardless of which pair of statements they picked on a thirty-item quiz. Every empirical claim here carries a linked source, which is the promise made in our editorial policy, and anything this site has had to retract is listed on the corrections page. The opposite case is worth seeing beside this one: the ick is a popular relationship idea with a single underpowered study behind it, where the love languages are a popular idea with thirty years of tests. More of this under Mindset.
Pulling it together
The honest summary is narrower than either slogan and more useful than both.
The quiz does not measure what people think it measures. It ranks thirty forced pairings, and the ranking does not reproduce the same person's own ratings of the same five things: across the one study that tested it directly, the forced-choice answer and the highest continuous score agreed six times out of ten, and the five subscale tests all came out non-significant. The clearest demonstration is gifts, which is the primary language of nought to four per cent of people by quiz and more than half by rating scale, depending only on how you ask.
The category system is not fictional, it is too coarse. The bottom-up work finds seven to ten factors rather than five, and a separate qualitative study found a sixth that Chapman could not have anticipated in 1992: brief check-in messages, mostly by text, reported almost as often as gifts or physical touch and more often by men than women.
The matching claim is the one to drop. It has failed at 83 couples, at 67, at 250, at 696 individuals and at 954 couples, and the one study that found it measured a continuous discrepancy sum that is hard to separate from simply being loved attentively. What replaced it is better advice anyway: relationship quality tracks satisfaction across a range of loving behaviours, not alignment on one.
And the thing that survives is the thing people were getting from the book all along. Sitting down with a prompt and saying plainly what you want more of, then listening to the same from the other person, is a real intervention. It does not require either of you to have a type.
Frequently asked questions
Are the 5 love languages scientifically proven?
No, and the state of the evidence is more specific than that. A 2024 review in Current Directions in Psychological Science evaluated the research against Chapman's three core assumptions — that each person has one primary love language, that there are five, and that couples do better when partners speak each other's — and found little support for any of the three. Since then the picture has firmed up rather than reversed: a preregistered study of 499 people found that fewer than half had an identifiable primary love language, and a bottom-up factor analysis across samples of 696, 500 and 499 found that solutions with seven to ten languages fitted better than five. What is not in dispute is that the behaviours themselves matter. Telling someone what you value about them, giving them undivided time, helping, touching and giving all correlate with relationship satisfaction. The disputed part is the typology and the matching rule built on top of them.
Is the love language quiz accurate?
The quiz has a specific design problem rather than a vague one. It is 30 forced-choice items: you pick which of two statements is more meaningful, so every point one language gains, another loses. That cannot record that two things both matter to you, which is the ordinary case. Denise Polk and Nichole Egbert tested the consequence directly in 83 couples by giving people the forced choice and a 20-item rating scale covering the same five languages. The forced-choice answer did not significantly predict scores on any of the five rating subscales — the F values were .49, .36, .48, .51 and .31, all non-significant — and it matched the person's own highest-rated language only six times across ten comparisons. Among men who picked physical touch on the forced choice, the language they actually scored highest on was quality time. The result is a ranking of one afternoon's trade-offs, not a stable property of a person.
Do couples need to have the same love language?
The evidence says no, and this is the best-tested of Chapman's claims. Matching has failed in study after study: 83 couples (Polk and Egbert, 2013), 67 couples (Bunt and Hazelwood, 2017, where each partner's self-regulation explained more variance in satisfaction than alignment did), a cluster analysis of couples' distributions (Bland and McQueen, 2018), 250 couples (Surijah and colleagues, 2021), 696 individuals (Flicker and Sancier-Barbosa, 2024) and 954 couples (Chopik and colleagues, 2025), the last of which used difference scores, response surface analysis and machine learning across all possible pairings and found virtually no evidence that matching added anything beyond each partner's personality and attachment. One study of 100 heterosexual couples did find an association, but it measured a continuous discrepancy summed across all five languages, which is difficult to separate from how much a partner expresses in general.
How many love languages are there?
More than five, on the only evidence that has asked the question from the data upwards. Sharon Flicker, Flavia Sancier-Barbosa, Emily Impett and colleagues rebuilt a love-languages measure from the items up across three samples of people in long-term relationships in the United States — 696, 500 and 499 — and found that solutions ranging from seven to ten love languages fitted the data better than the five-factor solution, and predicted relationship quality better too. Their conclusion is that Chapman's model is overly narrow rather than wrong in kind. Earlier factor analyses had already failed to converge on five: published structures include three factors, four and five, and a study of 250 couples using both a rating scale and a forced-choice scale reported that the five-factor solution was not supported. In fairness to the original, Egbert and Polk's 2006 validation did find a five-factor fit superior to one-, three- and four-factor alternatives.
Is there a sixth love language?
One candidate has been published, and it is worth knowing about. Rudy Pett, Priscilla Lozano and Sarah Varga collected 648 open-ended responses from 324 people about how they and their partners express love, coded them against Chapman's own definitions, and found a theme none of the five covered. They called it check-ins: brief, asynchronous, partner-centred messages asking whether you are all right — a good morning text, a message after an exam, checking you got home safe. It appeared in 100 responses, 15.4% of the total, which is almost exactly as often as gifts or physical touch (16.4% each), with acceptable coder agreement. Most of those check-ins happened by text message, which is the sort of thing a 1992 taxonomy could not have included. It was also the only theme reported more often by men (25.5%) than by women (14.1%).
Can your love language change?
There is no study tracking the same people's love languages over time, so this cannot be answered directly — but two findings bear on it. The first is that the instrument is unstable by construction: because the quiz forces trade-offs, a change in which comparisons feel salient to you will change the output without anything about you changing. The second is that love-language mismatch within couples correlated r = .01 with relationship length in the one study that checked, in both men and women, so couples do not appear to converge over the years. If what the question is really asking is whether what you want from a partner shifts with circumstances, the research says it plainly: physical affection matters more under stress, and people with chronic insecurities benefit more than others from a partner's expressed gratitude. That is context, not conversion.
Does learning your partner's love language actually improve the relationship?
Only one randomised controlled trial of a love-languages intervention exists, and it is small. Guanhua Chen and colleagues randomised 40 couples — 20 in China, 20 in Malaysia, ten per arm in each country — to a structured Love Language Game or to a different relationship-enhancement workshop. The groups were comparable at baseline, and afterwards relationship satisfaction rose more in the treatment arm, with a significant pre-post effect. The authors note the small sample and the brief follow-up. What the trial cannot show is that matching was the active ingredient: the control group also did a structured relationship activity and also mostly improved. The most defensible reading, which the 2024 review reaches independently, is that the exercise works by giving two people an occasion to say what they need and hear what the other needs — which is valuable, and does not require the five categories to be real.
If the research does not support it, why did the book help so many people?
Because the parts of it that are true are the parts people act on. The behaviours Chapman named are real and they matter: in the one study that used his own inventory, on 981 participants, the perception that a partner used your preferred language well predicted both love and relationship satisfaction for all five languages. The book also gives couples a vocabulary and a reason to sit down and discuss unmet needs, which is one of the better-supported things anyone can do. What the evidence does not support is the machinery around that: one primary channel per person, exactly five channels, and matching as the mechanism. The 2024 review proposes replacing the language metaphor with a balanced diet — people need a range of nutrients rather than one — while being clear that this is a better summary of existing evidence rather than a tested intervention in its own right.
Sources
What this article draws on. For the broader bibliography see our sources page.
- Impett, E. A., Park, H. G., & Muise, A. (2024) — Popular psychology through a scientific lens: Evaluating love languages from a relationship science perspective. Current Directions in Psychological Science, 33(2), 87–92. The review this page is built around. Source of the three core assumptions, the cross-study comparison showing gifts at 0–4% by forced choice versus over 50% by continuous score, the rs = .54–.75 between the five, the 20-million-copy and 30-million-quiz figures, and the balanced-diet proposal. Read in full in the authors' accepted version on OSF.
- Polk, D. M., & Egbert, N. (2013) — Speaking the language of love: On whether Chapman's (1992) claims stand up to empirical testing. The Open Communication Journal, 7(1), 1–11. Open access, read in full. 83 couples; the forced-choice answer did not predict scores on any of the five continuous subscales — F(4, 160) = .49 words, .36 quality time, .48 gifts, .51 acts of service, .31 touch, all non-significant — and matched the person's own highest mean only six times across ten comparisons.
- Egbert, N., & Polk, D. (2006) — Speaking the language of relational maintenance: A validity test of Chapman's (1992) Five Love Languages. Communication Research Reports, 23(1), 19–26. The 20-item Love Language Scale used by most of the later research, and the one confirmatory factor analysis that found the five-factor fit superior to unidimensional, three- and four-factor solutions. Cited here from its abstract and from the review's report of its correlations.
- Flicker, S. M., Sancier-Barbosa, F., & Impett, E. A. (2025) — Revisiting the five love languages framework: Toward a more flexible model of love expression. Journal of Marital and Family Therapy, 51(4), e70078. The preregistered test: 499 individuals in long-term cohabiting relationships; fewer than half had an identifiable primary love language, and satisfaction with a partner's expression of it predicted no better than other loving behaviours. Verbal affirmations, encouragement for individual pursuits, support during difficult times and accountability were the robust predictors. Closed access with no open copy; quoted from the published abstract, so no effect sizes from it appear on this page.
- Flicker, S. M., Sancier-Barbosa, F., Clemons-Castanos, C., Field, S., Jackson-Zambon, S., Phelan, M., & Impett, E. A. (2025) — How many love languages are there? Examining Chapman's five love languages using a bottom-up approach. Journal of Marital and Family Therapy, 51(4), e70072. Three studies, N = 696, 500 and 499: solutions of seven to ten love languages fitted the data better than the five-factor solution and better predicted relationship quality. Cited from the published abstract.
- Flicker, S. M., & Sancier-Barbosa, F. (2024) — Testing the predictions of Chapman's five love languages theory: Does speaking a partner's primary love language predict relationship quality? Journal of Marital and Family Therapy, 51(1), e12747. 696 participants recruited through CloudResearch, mean age 43.8, 97.3% cohabiting: Chapman's hypothesis was not supported, and words of affirmation and quality time predicted perceived love and satisfaction better than the person's own primary language. Free on the publisher's site but behind a bot check that would not open here; quoted from the abstract.
- Chopik, W. J., Hickman, L., Weidmann, R., Purol, M., & Yang, H. (2025) — Lost in translation: Matching on love languages explains little variation in relational outcomes in mid and late life. Innovation in Aging, 9(Supplement_2). The largest test of the matching claim: 954 male/female couples, mean age 55.42, mean relationship length 27.39 years, using difference scores, response surface analysis and machine learning. Virtually no evidence that matching added anything beyond personality and attachment; expression correlations between languages reached r = .795 and preferences r = .774. Published as a conference abstract, which is why no further detail is quoted here.
- Pett, R. C., Lozano, P. A., & Varga, S. (2022) — Revisiting the languages of love: An empirical test of the validity assumptions underlying Chapman's (2015) five love languages typology. Communication Reports, 36(1), 54–67. Read in full. 648 open-ended responses from 324 participants: words of affirmation 296 (45.7%), quality time 215 (33.2%), acts of service 164 (25.3%), gifts and physical touch 106 each (16.4%), plus a sixth theme — check-ins — in 100 responses (15.4%), mostly by text message, and the only theme more common among men (25.5%) than women (14.1%).
- Mostova, O., Stolarski, M., & Matthews, G. (2022) — I love the way you love me: Responding to partner's love language preferences boosts satisfaction in romantic heterosexual couples. PLOS ONE, 17(6), e0269429. Open access, read in full. The one study finding matching effects: 100 heterosexual couples, 31 nationalities, relationships from six months to 24 years. Mismatch was a sum of discrepancies across all five languages, and the authors state that the design cannot establish direction. Mismatch correlated r = .01 with relationship length in both sexes.
- Hughes, J. L., & Camden, A. A. (2020) — Using Chapman's five love languages theory to predict love and relationship satisfaction. Psi Chi Journal of Psychological Research, 25(3), 234–244. Open access, read in full. 981 participants living with their partners, including substantial lesbian (174) and gay male (168) samples, and the only study in this set to use Chapman's own profile. Perceiving that a partner used your preferred language well predicted love (R² = .26 words, .23 quality time, .17 gifts, .25 acts of service, .24 touch) and satisfaction (.32, .24, .11, .20, .24).
- Bunt, S., & Hazelwood, Z. J. (2017) — Walking the walk, talking the talk: Love languages, self-regulation, and relationship satisfaction. Personal Relationships, 24(2), 280–290. 67 heterosexual couples. In the authors' own words, the results provided limited evidence that love language alignment promotes satisfaction, and self-regulation contributed greater variance in satisfaction than alignment did. Cited from the publisher's abstract.
- Bland, A. M., & McQueen, K. S. (2018) — The distribution of Chapman's love languages in couples: An exploratory cluster analysis. Couple and Family Psychology: Research and Practice, 7(2), 103–126. Listed by the 2024 review among the studies that failed to find matching support, and the source of the strongest counter-argument on this page: that most researchers used their own scales rather than Chapman's inventory, which may itself explain the mixed results. Closed access; cited only for arguments reported in sources that could be read in full.
- Surijah, E. A., Prasetyaningsih, N. M. M., & Supriyadi, S. (2021) — Popular psychology versus scientific evidence: Love languages' factor structure and connection to marital satisfaction. Psympathic: Jurnal Ilmiah Psikologi, 7(2), 155–168. 250 couples measured with both a rating scale and a forced-choice scale: the five-factor solution was not supported and love-language compatibility did not affect marital satisfaction.
- Chen, G., Chen, Y., Fan, Y., Zhang, Z., Chen, G., Wu, J., & Abidin, N. E. Z. (2025) — The impact of a love language game intervention on relationship satisfaction among Chinese couples in China and Malaysia. BMC Psychology, 13(1), 756. Open access, read in full. The only randomised controlled trial of a love-languages intervention: 40 couples, ten per arm in each of two countries, control group given an alternative relationship-enhancement workshop. Baseline t = −0.464, p = .644; pre-post F = 33.73, p < .001, η² = .32. The authors list the small sample and the brief follow-up as limitations.
- Stafford, L. (2010) — Measuring relationship maintenance behaviors: Critique and development of the revised relationship maintenance behavior scale. Journal of Social and Personal Relationships, 28(2), 278–303. The bottom-up alternative referenced in the final sections: seven maintenance behaviours — positivity, openness, assurances, social networking, task sharing, conflict management and advice — three of which have no equivalent among the five love languages.
- Park, Y., Impett, E. A., MacDonald, G., & Lemay, E. P. (2019) — Saying "thank you": Partners' expressions of gratitude protect relationship satisfaction and commitment from the harmful effects of attachment insecurity. Journal of Personality and Social Psychology, 117(4), 773–806. Five studies. Cited here for the finding that it was perceived gratitude, rather than the partner's own report of expressing it, that buffered insecurely attached people's daily dissatisfaction.
- The Official Love Language Quiz — 5lovelanguages.com, read on 4 November 2026 — the publisher's own page, cited for the figures and framing quoted in this article: over 20 million copies sold, a four-minute quiz, and the warning to beware of unauthorized quizzes, tests and imitations alongside an authorised premium assessment.
- Europe PMC and PubMed — The literature counts reported in this article, run on 4 November 2026 so that they can be re-run rather than taken on trust: in Europe PMC, "love languages" returns 51 records and "five love languages" returns 22; in PubMed, "love languages" returns 9.
- Psychology Today (January 2025) — "Love Languages: What Science Actually Says About Relationships", Psychology Today — the second result for this question on the page of results measured for this article, linked here because this article audits it. It names no study, no author and no year anywhere in its text.
- Find A Helpline — Free, confidential support services listed by country — the referral in the first section of this article, for the case where a stated preference has turned into a rule one partner enforces on the other.
Keep reading
- MindsetThe Habits of Couples Who Stay Together: What the Longitudinal Research MeasuredPersonality barely predicts who lasts. What two people do on ordinary days predicts a great deal - though less precisely than the famous numbers suggest.
- MindsetHow to Reconnect After a Fight: What the Repair Research ShowsResearch says little about how to fight well and a great deal about the hours afterwards. Most advice for those hours is untested or points the wrong way.
- MindsetThe Psychology of Jealousy: The Feeling, the Behavior, and the Line Between ThemThe feeling is ordinary and almost everyone has it. What a minority do with it is a different subject, and the research is clear about where the line falls.
- SignsSigns A New Relationship Has Real PotentialSeparating the predictors with longitudinal evidence behind them from the ones that are just folklore.
- MindsetLove or Anxious Attachment? How to Tell the DifferenceFrom the inside they feel the same. They come apart in your body, in how long reassurance lasts, and in whether the intensity needs uncertainty to survive.
- MindsetSigns You Have Actually Moved On (And Signs You Have Not)Moving on is not the absence of feeling. The research points at something more specific than that.