Does Couples Therapy Work? What the Trials Report, and Where the Six-Year Figure Came From
Couple therapy works, moderately, and fades for about half of couples. The six-year waiting figure is an unpublished thesis on fewer than 20 couples.
Before the article, the one thing that changes whether any of it applies. If what happens in your relationship involves fear, threats, being prevented from leaving, having your phone, your movements or your money controlled, or any physical harm, the research below is not about your situation: every effect size on this page was measured in couples who had been screened for that and excluded. The section near the end says why, and Find A Helpline in the sources lists free and confidential services by country.
With that said. There are two good answers to this question and they sound like they contradict each other. Couple therapy, as a thing in general, has a large effect at the end of treatment — a pooled Hedges g of 1.12 across 58 studies and 2,092 couples, against 0.12 for the waitlisted couples in the same trials. And as of October 2026, a systematic review applying the American Psychological Association's own classification criteria concluded that not one specific couple therapy model meets the standard for strong empirical support, because most trials in the field ran 15 or fewer couples per condition where 25 is the minimum.
Underneath that sit the numbers nobody quotes. How many couples recover, improve, stay the same, or get worse — including the deterioration rate, which appears in none of the twelve pages currently ranking for this query. What happens at two years and at five. What the effect drops to in ordinary practice rather than in trials. Where the 70-something per cent success figure comes from, and why the biggest version of it was pooled from four studies.
And one number in particular. The claim that the average couple waits six years before seeking help, repeated by four of those twelve pages and attributed to John Gottman, traces to an unpublished 1992 master's thesis on fewer than 20 couples that the researchers who went looking for it could not obtain a copy of. When they measured it properly, in 270 people who had been through couple therapy, the answer was 2.68 years.
Two questions wearing one phrase, and one case where this is the wrong page
One thing first, because it decides whether anything below applies. If what happens in your relationship involves fear, threats, being prevented from leaving, having your phone or your movements or your money controlled, or any physical harm, then the question on this page is not your question. Conjoint therapy for relationship distress is studied in couples who have been screened for exactly that and excluded, and the sensible next step is a person rather than an article. 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. The section near the end on when couple therapy is not the room says why this is a separate subject rather than a harder version of the same one.
With that said. Does couples therapy work is two questions that have different answers from different evidence, and almost every page answering it answers one and reports it as the other.
The first is about couple therapy in general: does sitting in a room with a trained third party, as a pair, move anything measurable? That has been tested a great deal, including in a meta-analysis of 58 studies, and the answer is yes, by a lot at the end of treatment.
The second is about specific branded methods — whether the Gottman Method or Emotionally Focused Therapy or whichever name is on the therapist's website is the evidence-based one. That has been tested much less well than the marketing implies, and in 2026 a systematic review applying the American Psychological Association's own classification criteria concluded that not one couple therapy model currently meets the standard for strong empirical support.
Both of those things are true at the same time, and holding them together is most of what this page is for. The third thing it is for is a number: the claim that the average couple waits six years before getting help. That figure turns out to have a traceable source, and the source is not what anyone thinks.
The short answer, with the numbers that are actually in the papers
Four sentences, each of which the rest of the article unpacks and sources.
At the end of treatment it works, and the effect is large. The biggest meta-analysis, 58 studies and 2,092 couples, puts the pre-to-post change in relationship satisfaction at a Hedges g of 1.12. Couples assigned to waitlists in the same studies did not significantly improve, at g = 0.12.
About half the couples are better and a substantial minority are worse. Pooling the two arms of the largest behavioural trial at termination, 130 couples: 48% recovered, 18% improved but still distressed, 21% unchanged, and 13% deteriorated. That last category is in none of the twelve pages currently ranking for this question.
The gains are not durable for about half of the people who get them. Five years after behavioural couple therapy, 33% of couples had recovered, 13% had improved but were still distressed, and 54% had deteriorated, dissolved, or not changed. Of the couples who were already worse at the end of treatment, 87% were still in that category five years later.
And in ordinary practice the effect is roughly half what the trials report. Across nine studies in real clinical settings the within-group effect size is about d = 0.45 to 0.50, against something above 0.80 in controlled trials, and about half of all couples do not finish.
None of that says don't go. The comparison that matters is not therapy against a perfect outcome, it is therapy against the untreated alternative, and the untreated alternative in these datasets is a waitlist that goes nowhere. What it does say is that the figures in circulation describe the best case at the best moment, and the honest answer is more conditional and more interesting than either it works or it doesn't.
The six-year figure is an unpublished master's thesis on fewer than 20 couples
Four of the twelve pages ranking for this query state that the average couple waits six years before seeking help, and attribute it to John Gottman or to the Gottman Institute. It is one of the most repeated numbers in the whole subject, and it has been traced.
The source is Buongiorno, J. (1992), Wait time until professional treatment in marital therapy, Catholic University of America, unpublished master's thesis. That is the entire provenance of the figure.
William Doherty, Steven Harris, Eugene Hall and Aimee Hubbard went looking for it and published what they found in the Journal of Marital and Family Therapy in 2021. They could not obtain a copy of the thesis from the university where it was written. They contacted the author, who no longer had a copy either, but who described the study by email in January 2020: the sample was non-divorced couples who had sought couple therapy together from three to four private-practice therapists, each partner completing a questionnaire independently, and — in her own words — the sample was not large, she could not remember the exact number, but it was fewer than 20 couples. The paper's discussion describes it as 20 couples, 40 people, recruited through three or four therapists in Washington DC.
So: a figure from the early 1990s, from a single city, from an unpublished thesis that cannot be read, with a sample smaller than most pilot studies, carrying a sentence that is now repeated in books, in peer-reviewed journals, and on the websites of thousands of practices.
The pattern is one this site keeps running into from different directions, and the comparisons are exact. The three-day texting rule and the 3-6-9 rule are numbers with no source at all. The thirty days of no contact is a number that disagrees with itself by a factor of seventeen across the pages prescribing it. This one is different and worse in a specific way: it has a real source, the source is findable, and nobody went and looked.
How it spread: a book in 1999, then the journals, under three different names
The route from a thesis nobody can read to an established fact is documented, and it is worth following because it is how this genre reproduces.
As far as Doherty and colleagues could establish, the first citation was Gottman's book The Marriage Clinic, in 1999, and it appears again in The Science of Trust in 2011, on page 425, in the form research indicates that the average waiting time of couples who seek professional help for serious relationship problems is about six years. From there it entered the peer-reviewed literature. Doherty's paper lists citations in work by Doss and colleagues in 2003 and 2009, Hubbard and Harris in 2020, Nightingale and colleagues in 2019, Roddy and colleagues in 2019, and Trillingsgaard and colleagues in 2019, plus three books and at least four theses and dissertations.
And it is cited under three different author lines. The correct one is Buongiorno, 1992. The thesis supervisor, Clifford Notarius, is frequently added as a coauthor, giving Buongiorno and Notarius, 1992, and in some papers the order is reversed to Notarius and Buongiorno, 1992.
Both of the wrong forms are checkable in published reference lists, which is the part that makes this concrete rather than an accusation. Hubbard and Harris's 2020 review in Contemporary Family Therapy lists it as Buongiorno, J., and Notarius, C. (1992). Jarnecke and colleagues' 2020 paper on veteran couples lists it as Notarius C and Buongiorno J. (1992) … Unpublished paper, Catholic University of America, and cites it in the body of the text as Notarius and Buongiorno, 1992, as cited in Gottman and Gottman, 1999 — that is, with the authors reversed, the document type changed from a master's thesis to an unpublished paper, and the citation taken at second hand from a trade book.
Searched on 8 November 2026, the thesis title returns zero records in Europe PMC and the figure appears in no indexed paper as a primary finding. The six years is not a research result that the field has been building on. It is a sentence that has been passed along.
The large-sample study found 2.68 years, and most people inside two
Doherty and colleagues did not only trace the figure. They went and measured it, in the first large-sample study of the question, and the answer is less than half of what everyone repeats.
They recruited through Amazon Mechanical Turk in 2018 and 2019, screening for people who had been in therapy specifically for relationship problems with a licensed mental health professional rather than a member of the clergy. N = 371, split into 270 who had attended couple therapy with their partner and 101 who had attended individual therapy for relationship problems. Ages 23 to 72, mean 36. Seventy-six per cent heterosexual, 53% men. The largest group of therapists were marriage and family therapists at 37% and clinical psychologists at 23%. Participants had seen an average of two therapists and attended an average of eight sessions.
The two measures were single questions, drawn from a list of 20 specific relationship problems: how long the problems had been damaging the relationship at the time of the first session, and how long the person had considered seeking help before it.
For the 270 in couple therapy, the mean interval from the onset of serious problems was 2.68 years (SD = 2.09). The distribution matters more than the mean:
- 64% had been dealing with serious problems for two years or less
- about 22% for three to four years
- about 11% for five years or more
The time spent thinking about it before booking was 1.87 years (SD = 1.29), with more than 80% deciding inside two years. The 101 who went to individual therapy for the same problems looked almost identical — 2.46 years of problems, 2.11 years of deliberation — and independent-samples t-tests found no significant difference between the two groups on either measure. The authors had expected couples to take longer, on the reasoning that two people are harder to get through a door than one, and they did not.
The limits are real and the authors state them. It is a convenience sample, better educated and with more Asian participants than the United States as a whole. The participants are individuals, not couples. They were asked to recall events that may have been many years earlier. The two measures were newly written for the study and had not been validated anywhere else. And everyone in it eventually went, so it says nothing about the people whose answer to how long did you wait is still running.
The one sample that waited longer, and what its standard deviation shows
In fairness to the six years, there is a study that found something close to it, and Doherty's paper names it rather than leaving it out — it surfaced after their own work had gone through peer review.
Alyssa Jarnecke and colleagues, in Couple and Family Psychology: Research and Practice in 2020, took 97 opposite-sex couples presenting to a Veterans Affairs medical centre, 85.6% of them married, and asked each partner how long it had been since they realised there were problems in the relationship. The published abstract reports that partners tend to wait approximately four to seven years.
The reason for that range, rather than a single number, is the interesting part. Men reported 3.73 years (SD = 4.36). Women reported 7.39 years (SD = 11.64). The difference approached but did not reach significance, t(29) = −1.83, p = 0.079, and the authors attribute the failure to the size of the standard deviations rather than to the absence of a gap.
Look at that second standard deviation. It is larger than its own mean, which means the distribution of answers is nothing like a bell curve around seven years — it is a pile of short waits with a long tail of very long ones. An average computed on that shape is a poor description of anybody, which is exactly the problem with reciting a single figure for this quantity in the first place.
Two further things in the same paper are more useful than the headline. Partners had made multiple attempts to improve the relationship before booking anything, a mean of 1.79 attempts for men and 2.40 for women, so the waiting is not inaction. And the longer someone had waited, the more optimistic their expectations for therapy were, which is the opposite of what the six-year story implies about people who come late.
Why this sample differs from Doherty's is not settled. Military and veteran couples may be less receptive to couple therapy, the cohorts are thirty years apart, and the question was worded differently. What cannot be said is that either number is the average couple's wait, because the two best measurements of it disagree by a factor of two and neither sample is representative of anything.
Why the “too late” implication is the harmful part
Doherty and colleagues are explicit that their main objection is not the figure's weak provenance. It is what gets attached to it.
Their paper records what they found on practice websites: the six years paired with an analogy comparing it to waiting six years to see an oncologist. The inference the analogy invites is that couples who waited are foolish or irresponsible, and — the part that actually matters — that by the time they arrive there is not much to be done. The authors say they are concerned that this negative picture puts couples off seeking the help they need, and that repeated often enough it undermines the hope of therapists and clients alike.
The evidence does not support the pessimism, and the clearest demonstration is the trial most people would expect to show it. Andrew Christensen, David Atkins, Brian Baucom and Jean Yi followed 134 chronically and seriously distressed married couples for five years after about eight months of treatment. Not couples with a recent rough patch: the sample was selected for chronic, serious distress, which is to say exactly the population the six-year story says has left it too late. Pre-to-post effect sizes were d = 0.90 and d = 0.71 for the two models, and at five years, relative to where they started, d = 1.03 and d = 0.92. Half of one group and 45.9% of the other showed clinically significant improvement at five years.
Separately, the biggest meta-analysis of couple therapy tested whether baseline distress moderated how much couples gained, and found that greater distress at intake went with larger gains, not smaller ones. Severity at the door is not a prognosis.
There is a real finding about timing underneath all this, and it is the opposite shape to the folklore: the people with the most pessimistic forecasts about their own relationships are often the ones whose forecasts are least accurate, which is covered in what has actually been measured about moving on. If you are weighing whether it is too late, the number that was supposed to tell you is a thesis nobody has read.
What the largest meta-analysis reports, and what the waitlists did
Now the efficacy question proper, and it starts with the single most informative study on it.
McKenzie Roddy, Lauren Walsh, Karen Rothman, Stacey Hatch and Brian Doss published a meta-analysis in the Journal of Consulting and Clinical Psychology in 2020 that set out to fix what the earlier ones left out: multiple treatment models rather than one, several research designs rather than only randomised trials, long-term outcomes, and recent studies. More than 70% of the studies they included had not appeared in any previous meta-analysis.
Across 58 studies, 40 unique samples and 2,092 couples:
- Pre-to-post within-group change in relationship satisfaction: Hedges g = 1.12, CI [0.92, 1.31].
- The same figure for couples assigned to a waitlist: g = 0.12, CI [−0.04, 0.29], not significant.
- Significant effects also on self-reported communication, observed communication, emotional intimacy and partner behaviours.
- Gains generally maintained over short- and long-term follow-up.
- Moderator analyses were mostly non-significant, with one exception: greater baseline distress went with larger gains.
The waitlist line is the one to keep. A pre-to-post effect size on its own tells you very little, because people seek help when things are at their worst and some of what follows is regression toward their own average. The waitlist comparison is the control for that, and in these studies untreated distressed couples went essentially nowhere over the same interval. That is consistent with a separate observation in the literature that relationship distress rarely remits on its own.
The authors name their own main limitation and it is a serious one: the pooled sample is exclusively opposite-sex couples. Whatever the evidence base shows, it does not currently show it for same-sex couples, and the 2026 systematic review discussed below closes by calling for large-scale studies with more diverse couples precisely to establish how far any of this generalises.
If what you want is the one-line version of this section: at the end of treatment, couple therapy does something large, and the thing it is being compared against does nothing. The complications all live downstream of that.
Recovered, improved, unchanged, worse: the four-way split nobody publishes
An average effect size tells you the mean movement of a group. It does not tell you how many people in that group got better, and that is the number readers actually want. It exists, and it is not on any page ranking for this question.
Thomas Bradbury and Guy Bodenmann's review in the Annual Review of Clinical Psychology in 2020 sets it out, trial by trial, as four categories: fully recovered, improved but not recovered, unchanged, and deteriorated or dissolved.
At the end of treatment, pooling behavioural and integrative behavioural couple therapy across 130 couples: 48% recovered, 18% improved but still distressed, 21% unchanged, 13% deteriorated. In the behavioural arm specifically, 134 couples: 44% recovered, 17% improved, 21% unchanged, 18% dissolved or more distressed than when they started. In an emotionally focused sample of 28 couples: 61% recovered, 11% improved, 25% unchanged, 4% deteriorating.
Two years after the end of treatment, behavioural couple therapy, 60 couples: 37% fully recovered, 23% improved, 15% unchanged, 25% deteriorated or dissolved. Emotionally focused, 28 couples: 39% recovered, 7% improved, 21% unchanged, 32% deteriorated or dissolved.
Five years after, behavioural, 61 couples: 33% recovered, 13% improved but still distressed, 54% deteriorated, dissolved or unchanged. Integrative behavioural, 58 couples: 31%, 19%, 50%.
Read down that list and the shape of the thing becomes visible. Roughly half the couples who go through an empirically supported couple therapy are meaningfully better at the end. Roughly 40% are still meaningfully better two years later. By five years it is about a third, and the largest single category is the one containing everyone for whom it did not hold.
The samples are small — 28, 58, 60, 61 couples — and that is a property of this entire literature rather than a criticism of these particular trials. It is also the reason a later review downgraded every model in the field, which is two sections below.
The honest summary is the one Bradbury and Bodenmann put in their own abstract: 60 to 80% of distressed couples benefit from the leading approaches, the effects are weaker in real clinical practice, they dissipate for about half of all couples, and they may be explained by factors common to all the models rather than by anything specific to any of them.
Where the 70-something per cent comes from
Seven of the twelve pages ranking for this query quote a figure in the seventies. The exact claims vary — 70%, 70 to 75%, 70 to 73% recovery, 70 to 80% of couples report feeling happier — and two of them add that about 86% or 90% show significant improvement. Almost none says which study it is.
There are three different things being blurred together, and they are worth separating because two of them are defensible and one is nearly thirty years old.
The first is a comparison, not a success rate. A 2022 review in Family Process states that the average person receiving couple therapy is better off at termination than 70 to 80% of individuals not receiving treatment. That is a statement about where a treated person sits in an untreated distribution, which is a standard way of expressing an effect size. It is not the proportion of couples who got better, and when it is reported as 70 to 80% of couples improve it has been changed into a different claim.
The second is a real recovery rate from a real meta-analysis. Paul Spengler, Nicholas Lee, Stephanie Wiebe and Andrea Wittenborn published the first comprehensive meta-analysis of Emotionally Focused Couple Therapy in 2024, pooling randomised trials, quasi-experiments and dissertations. Across 20 studies and 332 couples: pre-to-post d = 0.93 CI [0.75, 1.12], pre-to-follow-up d = 0.86, and EFT against viable alternative couple interventions d = 0.44, CI [0.03, 0.85]. They conclude that about 70% of couples will be symptom free at the end of treatment. So the 70% has a source, and it is specific to one model, measured at termination, in a pooled sample averaging about 17 couples per study. Note also where the lower bound of that comparison against other therapies sits: 0.03, which is a hair above nothing.
The third is the one being cited when the number climbs into the eighties and nineties, and it is from 1999. Susan Johnson, John Hunsley, Leslie Greenberg and Dwayne Schindler's meta-analysis in Clinical Psychology: Science and Practice is the origin of the highest effect size anyone reports for EFT, d = 1.31, pooled from four studies. Four. Bradbury and Bodenmann list it alongside the others precisely so the pattern is visible: the EFT estimates run 0.73 from 12 studies, 1.27 from 7, 1.31 from 4, and 2.09 from 9. The smaller the evidence base, the bigger the number, which is what small-sample literatures do.
It is the mirror image of the problem in what the research on whether opposites attract actually reports, where the papers were excellent and the coverage dropped the effect sizes. Here the numbers survive the trip and their denominators do not.
The 2026 review that downgraded every model, including the branded ones
In October 2026 Jeffrey Jackson, Luca Codecá, Richard Miller, Jessica Simpson, Abigail O'Leary and Samantha Renner published a systematic review in the Journal of Marital and Family Therapy that replicated and updated the standard 1998 evaluation of empirically supported treatments for couple distress, using the Chambless criteria — the framework the American Psychological Association uses to classify treatments.
Applied to 40 eligible studies, across 12 models, the conclusion was that no couple therapy model meets the standard for strong empirical support. Seven demonstrated modest support: behavioural, cognitive-behavioural, emotionally focused, Gottman, integrative behavioural, insight-oriented and strategic couple therapy. Five demonstrated weak support: imago relationship therapy, object relations, relationship enhancement, solution-focused and systemic couple therapy. And three widely practised models — Bowen family systems therapy, internal family systems therapy and narrative therapy — had insufficient research to evaluate at all.
The reason is not that the therapies failed. It is sample size. The Chambless criteria require at least 25 couples per condition after attrition, and the review found that the vast majority of trials in this literature had 15 or fewer. The authors go further: 25 per condition is itself adequate only when a large effect is expected, and comparing one active therapy against another needs roughly 70 couples per condition to be properly powered. Small samples do not only risk missing real effects; they inflate the ones they find, which is why a literature of underpowered trials reporting large effects should be read with the denominators in view.
Two details from the same review are worth having. The APA's own public-facing treatment site still lists Emotionally Focused Therapy as having strong empirical support for relationship distress, citing 19 studies — and the review notes that most of those studies are about populations with a specific condition rather than relationship distress itself: a partner with depression, PTSD, a sexual disorder or cancer, parents of children with autism or chronic illness, a history of childhood sexual abuse. Important work, and not evidence about the thing the label claims. Several of the others had no comparison condition.
And integrative behavioural couple therapy rests on exactly one eligible trial — the Christensen study with 62 to 66 couples per condition — while insight-oriented couple therapy rests on one trial from 1989, with nothing published on it since.
This site already reported the headline of this review in the piece on emotional maturity, where it corrected a claim about EFT that had been sourced to an institute rather than to a study. What belongs here is the mechanism: the downgrade is a statement about how couple therapy has been studied, not a verdict on what it does.
The field cannot agree with itself, and the reason is the criteria
Here is something that ought to be unsettling and is mostly just instructive. In 2022 a major review in Family Process stated that cognitive-behavioural, integrative behavioural and emotionally focused couple therapy each have sufficient evidence to be considered well-established treatments for relationship distress. In 2026 a systematic review concluded that none of them has strong empirical support. Neither is wrong. They used different rulebooks.
The 2022 decade-in-review in the same journal as the 2026 paper classified behavioural, emotionally focused and integrative behavioural couple therapy as well-established, using criteria from Southam-Gerow and Prinstein rather than the Chambless criteria. The key difference between the two frameworks is that the first has no sample-size requirement at all; its authors set their own inclusion floor at a minimum of 10 couples per condition, against the 25 the Chambless criteria demand. Run the same trials through one filter and three models are well-established. Run them through the other and nothing is.
That is not a scandal, and it is not a reason to disbelieve everybody. It is a specific, checkable fact about what the phrase evidence-based is doing when it appears on a therapist's homepage. The phrase is not a measurement. It is the output of a classification scheme, and there is more than one scheme.
What the 2026 review recommends, incidentally, is neither of these: it argues for the newer Tolin criteria, which synthesise evidence through meta-analysis rather than counting qualifying trials, and which weigh clinical significance rather than statistical significance. Under a meta-analytic framework the field's position would likely look better than it does under Chambless, because the meta-analyses are where the pooled samples finally get large.
The practical consequence for a reader is narrow and useful. Anyone telling you that one named method is the scientifically validated one is reporting a classification, not a finding — and as the next section shows, the head-to-head comparisons between methods are the weakest part of the whole literature. It is the same wedge as in what the research on the five love languages actually measured, where a branded framework's popularity ran a long way ahead of the tests of it.
Does the model matter? One four-year result says maybe, and it has not been replicated
The commercial question underneath does couples therapy work is usually which kind should we book. The evidence on that is thin, and it is thin in an interesting direction.
Markus Rathgeber, Paul-Christian Bürkner, Eva-Maria Schiller and Heinz Holling meta-analysed randomised controlled trials only — 33 studies, 2,730 participants, all measuring relationship satisfaction. At post-test the effects were medium: g = 0.60 overall, g = 0.53 for behavioural couple therapy and g = 0.73 for emotionally focused. At six months, smaller: 0.44 overall, 0.35 and 0.66. In the twelve-month data, which existed only for behavioural couple therapy, g = 0.06. Between the two methods there was no significant difference in effect size, and the authors flag possible publication bias.
The 2026 systematic review reaches the same place from a different angle: couple therapies were consistently superior to inactive conditions at post-treatment, and mostly equivalent to each other, with occasional short-term advantages for one model that had usually disappeared by follow-up. The authors name the relevant idea — the dodo bird verdict, from the line in Alice's Adventures in Wonderland that everybody has won and all must have prizes — and are careful to add that small samples and methodological noise could be hiding genuine differences rather than proving their absence.
Against that stands one result that has never been matched, and it deserves reporting properly rather than being waved away. Douglas Snyder and Robert Wills randomised 79 distressed couples to behavioural therapy, insight-oriented therapy, or a waitlist. At the end of roughly 20 sessions both treatments produced statistically and clinically significant gains over the waitlist, with no difference between them. At four years, 38% of the behavioural couples had divorced, against 3% of the insight-oriented couples. The authors argued that resolving the emotional conflicts each partner brought into the marriage changes their view of the other more durably than skills training does.
Two cautions, and then what to do with it. The 1991 follow-up drew a published commentary arguing that the therapists in the behavioural arm had not been using techniques integral to that model, which would make the comparison unfair. And insight-oriented couple therapy has had no new efficacy study published since, which is why the 2026 review classes it as modestly supported on the strength of a single trial from the 1980s.
So the fair answer to which kind is: nobody has shown that it matters much at termination, one old trial suggests it might matter a great deal four years later, and the question has been more or less abandoned rather than answered. What is left is the therapist, the fit, and whether both people actually come — which is an unsatisfying conclusion that happens to be what the evidence supports.
Durability is the weak point, and it fails in a predictable pattern
If the goal of couple therapy is lasting change, then the follow-up data is the only data that counts, and it is the thinnest part of the record. There are very few trials with four- or five-year follow-ups, and what they show is consistent.
Across the handful of controlled trials and one uncontrolled evaluation that looked four to five years after treatment, deterioration or divorce occurred for roughly 35 to 50% of couples. The 2022 Family Process review that reports that range also notes the single exception — Snyder and colleagues' insight-oriented trial, with a 20% deterioration-and-divorce rate at four years — and says plainly that it has not been replicated.
The decay is not random. Bradbury and Bodenmann describe it as following a predictable pattern, and the two halves of it point in opposite directions:
- Of couples who had deteriorated by the end of treatment, 87% were still in that category five years later. Bad outcomes are sticky, which is consistent with distress rarely remitting on its own.
- Of couples who reliably improved or fully recovered by the end of treatment, only about half were still in those categories five years later. The other half had reverted to where they started or gone further down.
There is a plain-language reading of this that is not a criticism of therapy. Relationships keep happening after the sessions stop. Lives change, children arrive, parents get ill, jobs end, and a couple who resolved one stage's problem will meet the next stage's. Most contemporary couple therapies include work before termination aimed at future difficulties, and the honest state of knowledge about those components is that their efficacy in forestalling later deterioration is unknown.
The practical consequence is that the useful question to ask a prospective therapist is not what their success rate is. It is what happens afterwards — whether there are booster sessions, whether coming back is expected rather than an admission of failure. On the evidence, a course of couple therapy behaves less like a repair and more like a treatment with a relapse rate, and the field says so about itself.
For what the longitudinal research says about which couples hold up over years without any intervention at all, what 137 studies found about couples who stay together is the companion piece, and none of what predicts it is a technique.
Trials are not clinics: what happens in ordinary practice
Everything above comes from efficacy research: manualised treatment, trained and supervised therapists, screened couples, randomisation. The question a reader is actually asking is about the therapist whose website they are looking at, and that is a different literature with worse numbers.
Effectiveness studies measure what happens in natural settings. Nine of them can be pooled, and the within-group pre-to-post effect size comes out at about d = 0.45 to 0.50, against something above 0.80 in controlled trials. Roughly half the effect.
The German national study is the clearest single demonstration. Christian Roesler surveyed couple counselling across Germany with 554 couples, applying the same methodology as two earlier studies in the same country, and found remarkable consistency across all three: the improvements achieved in practice settings are about half the effect sizes achieved in efficacy trials, and there is no improvement at all in up to 50% of couples.
Two more facts from the effectiveness literature, both of which belong in any honest answer:
- About half of all couples in these studies do not complete therapy. Not as a dropout footnote — as the modal trajectory alongside completion.
- Of those who do, as many as 40% reach a non-clinical level of relationship adjustment by their last session, and 30 to 60% of them relapse to pre-treatment levels at some point afterwards.
Bradbury and Bodenmann's own summary sentence is worth quoting because it is more blunt than anything on the first page of search results: while the typical couple seeking treatment in an ordinary community setting probably does not benefit from couple therapy, the couple who actually receives the treatment likely experiences moderate gains in satisfaction, of unknown durability.
The gap between those two clauses is attendance. Half of the people who start do not finish, and the couples who finish are not a random sample of the couples who started — they are the ones who kept turning up, which is itself one of the strongest things anyone has ever been able to say about who gets better. That is not a reason to skip therapy. It is a reason to treat the decision to go as the decision to go for several months, and to notice that the single variable most under a couple's own control is the one the data keeps pointing at. And where the obstacle is upstream of all of this — one of the two will not come at all — the help-seeking research on that case, including the one randomised trial of doing the work alone, is in when one partner won't go to couples therapy.
The online programmes, which are the best-measured thing in the field
One of the questions the ranking pages ask themselves most often is whether online couples therapy works, and by an accident of research funding this corner of the subject has better evidence than the in-person therapies it is compared against — larger samples, nationally representative recruitment, pre-registered waitlist designs, and twelve-month follow-ups.
Brian Doss and colleagues tested the OurRelationship programme in the Journal of Consulting and Clinical Psychology in 2016. It is an eight-hour online programme adapted from an in-person couple therapy, with four fifteen-minute calls with project staff. 300 heterosexual couples, 600 participants, broadly representative of the United States on race, ethnicity and education, randomised to begin immediately or to a two-month waitlist.
Against the waitlist: relationship satisfaction d = 0.69, relationship confidence d = 0.47, negative relationship quality d = 0.57. On individual functioning, with the effects concentrated among people who started with a problem in that domain: depressive symptoms d = 0.71, anxious symptoms d = 0.94, perceived health 0.51, work functioning 0.57, quality of life 0.44.
The four-way breakdown, which is the figure to compare against the in-person ones from two sections ago: 57% recovered or improved, 36% unchanged, 7% deteriorated. That deterioration rate is lower than the 13% to 18% in the behavioural trials, and the likeliest reason is simply that the programme is shorter and lighter-touch than eight months of conjoint therapy.
It holds up. A follow-up contacted 668 individuals who had enrolled in either OurRelationship or a behavioural online programme, one year after finishing, in a low-income sample. Relationship gains were maintained with no significant linear or quadratic decline, at medium-to-large within-group effect sizes from pre-treatment to one year; individual functioning was maintained or continued to improve; 68.3% were still with the same partner; and there were no significant differences between the two programmes, which is the dodo bird verdict showing up again in a new format.
Two honest caveats. These trials come from the group that developed the programme, which is normal in intervention research and still worth knowing. And the 2016 comparison is against a waitlist rather than against in-person therapy, so nothing here establishes that online is as good as a therapist — only that it does considerably more than nothing, at a fraction of the cost, for a population that the in-person system reaches badly. Roughly one third of divorcing couples seek professional help at all.
Can couples therapy make it worse? Yes, and here is the rate
None of the twelve pages ranking for this query contains the word deterioration. One of them asks, in its own FAQ, whether couples therapy can make things worse. The answer exists in the numbers already on this page, and it is worth stating on its own because it is the question people are embarrassed to ask.
At the end of treatment, between 4% and 18% of couples in these trials are worse off than when they started — 13% pooling the two behavioural models across 130 couples, 18% in the behavioural arm specifically, 4% in a small emotionally focused sample, 7% in the online programme. Two years out the figure is 25% to 32%, though that category merges deterioration with relationships that ended, which are not the same outcome and are not always a bad one.
Three things make that less alarming than it reads, and one makes it more so.
The category combines deteriorated with dissolved, and a distressed relationship ending is sometimes the outcome both people were moving toward with or without a therapist in the room. These samples are selected for serious chronic distress, in which the untreated base rate of getting worse is not zero — the waitlist couples in the meta-analysis did not improve, and distress rarely remits spontaneously. And the trials cannot tell you whether a particular couple got worse because of the therapy or despite it, because nobody has run the design that would separate those.
What makes it more serious is the stickiness. Of the couples who were worse at termination, 87% were still worse five years later, which is the single most durable finding in the follow-up data and points the wrong way.
The 2026 systematic review offers a mechanism in passing, while explaining why couple therapy trials have small samples: couple therapy can provoke difficult emotions and interpersonal conflict, which makes participants more likely to discontinue. Opening up what two people have been avoiding is the active ingredient and also the risk, and the literature treats those as the same thing because they are.
The practical version is not don't go. It is that getting worse is a real outcome with a published rate, that it is a reason to notice early rather than to persist on faith, and that a therapist who tells you the figure is close to zero is quoting something that no trial reports. Where the thing being avoided is a specific unresolved rupture rather than a general pattern, what the research says about repair after a fight covers the one model that has been tested on exactly that.
When couple therapy is not the room
This section is near the end for length, not for importance, and it repeats what is at the top of the page because it is the one part of this subject where being wrong has consequences.
Some of what gets brought to a couple therapist is not a relationship problem with two authors. Being told who you may see or message. Having your phone checked, or being required to share your location when it is not mutual and freely chosen. Being made to account for ordinary hours. Money controlled or withheld. Anger, punishment or silence arriving as a sanction when you say no to something. Being separated from friends and family a little at a time. Any fear of your partner, any physical harm, any sense that leaving would not be allowed.
UK statutory guidance on controlling or coercive behaviour names isolating a person from sources of support, monitoring their time and movements, controlling access to money and depriving them of independence as recognised markers of that pattern. It is not a communication problem and it is not a mismatch, and the framing matters practically: calling it either invites you to negotiate with it.
The research literature on this page is structurally silent about it, and that silence is informative. The 2026 systematic review states that studies of couples presenting with intimate partner violence were excluded from its evaluation, along with infidelity and infertility, because they define distinct clinical populations rather than a relationship-distress sample. That exclusion is also why no figure on this page answers the question of what happens to couples after an affair; the evidence that does exist on that is a nineteen-couple cell inside one of the trials above, and we go through it in whether couples stay together after infidelity. In other words, every effect size quoted above was measured in couples from whom these situations had been screened out. There is a specialist literature on conjoint treatment where violence is present, and the 2022 Family Process review cites evidence supporting some couple treatments for it — but it is a different question, requiring separate screening and separate expertise, and it is not what a general couple therapist's advertised success rate refers to.
If any of this is recognisable, the useful next step is a person rather than an article. Find A Helpline, in the sources below, 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. Where the pattern is jealousy rather than control, the line between the two is worked through in the psychology of jealousy; where it is the pull of a relationship that hurts and is hard to leave, what trauma bonding does and does not mean is a better starting point; and the mechanics of staying in something unhappy are in why people stay in unhappy relationships.
What the pages ranking for this question report, and what they leave out
Twelve pages returned for this query were downloaded as plain text and searched on 8 November 2026. Among them: Psychology Today twice, BetterHelp, Regain, Headway, Lifebulb, and six therapy practices.
Across all twelve:
- Zero link a DOI or a journal page.
- Zero give the sample size of any study they describe.
- Zero contain the word deterioration in any form.
- One contains the string et al.
- Nine mention Gottman. Eight mention Emotionally Focused Therapy.
- Seven quote a figure in the seventies as a success rate.
- Four repeat the six-year waiting figure, three of them attributing it to Gottman or the Gottman Institute.
- None mentions Buongiorno, Notarius, Doherty, the 2.68-year finding, Rathgeber, Roddy, Christensen or Bradbury.
Two things are worth saying in fairness, because a tally like this is easy to use dishonestly. One page is noticeably better than the rest: it cites a named review, cites the 2024 EFT meta-analysis with its 20 studies, admits that the 86% and 90% figures that appear elsewhere on its own site measure something different from what they are used for, and asks in its own words whether couples therapy can make things worse. And several of these pages are written by practising therapists whose clinical description of what happens in a session is better than anything a literature review can supply. The failure is not of expertise. It is that the numbers are quoted from each other rather than from the papers, and the papers are free.
Because they mostly are. Of everything cited on this page, the 2026 systematic review, the 2022 field review, the 2020 veteran study, the Annual Review chapter and the 2016 online trial are all readable in full at no cost. The research note that traced the six-year figure is open access in translation. What is behind a paywall is about half of the primary trials, and where that is the case this article says so in the source list and quotes only what the published abstract states.
That is the same gap this site has documented in the real numbers on modern dating, where statistics circulate without denominators, and in whether dating apps' matching does anything, where a product claim outran the one study that tested it. The difference here is the stakes: this is a page about a treatment, read by people deciding whether to spend several months and a good deal of money on it.
What follows from this if you are deciding whether to go
Five things, and none of them is a success rate.
The waiting argument is not a reason to hurry or a reason to despair. The six-year figure is an unreadable thesis on fewer than 20 couples, the best large-sample estimate is 2.68 years with most people inside two, and in the one trial that selected specifically for chronic serious distress, couples were still better off five years later. Severity at intake predicted larger gains, not smaller ones.
Expect a treatment with a relapse rate, not a repair. About half of couples are meaningfully better at the end, about 40% two years on, about a third at five years. The question to ask a therapist is what happens after termination, not what their success rate is.
Do not pick on the brand. Head-to-head comparisons rarely favour any model, no model currently meets the strictest classification standard, and the one trial suggesting a large long-term difference between two approaches is from the 1980s and has never been replicated. What the evidence keeps pointing at is common factors: the alliance, the therapist's warmth, both people's engagement, and a structured setting in which entrenched patterns can actually be addressed.
Treat finishing as the decision. About half of couples in real-world settings do not complete therapy, and the gap between couples who seek treatment and couples who receive treatment is where most of the effect lives.
And if the obstacle is cost, access or a partner who will not go in person, the online programmes are not a consolation prize — though what has been measured about the one-person version of one of them, and about pressure, refusal and who tends to be the reluctant one, is set out in what the research says when one partner won't go. An eight-hour structured programme with four short coaching calls produced a satisfaction effect of d = 0.69 against a waitlist in 300 nationally recruited couples, held its gains at one year in a low-income sample, and showed a lower deterioration rate than the in-person behavioural trials.
If what you want before any of that is a reading of where things currently stand rather than a forecast, the relationship satisfaction test reports satisfaction and commitment as separate scores, the conflict style test separates pressing for resolution from pulling away — the demand-withdraw pattern that most couple therapy spends its first sessions naming — and the attachment style test covers the individual difference that the therapy literature actually measures moving. All three say on their own pages what they cannot tell you, and none of them is a diagnosis or a substitute for the conversation this article is about.
What this evidence cannot tell you
The limits go here rather than in a footnote, because on a page about a treatment they are part of the answer.
It is almost entirely about opposite-sex couples. The authors of the largest meta-analysis name this as their own main limitation — a pooled sample of exclusively opposite-sex couples — and the 2026 systematic review ends by calling for large-scale studies with more diverse couples in order to establish generalisability. Nothing on this page is established for same-sex couples, and saying otherwise would be inventing it.
The samples are small, and that is the field's defining feature. The follow-up percentages above rest on 28, 58, 60 and 61 couples. The 2026 review found most trials running 15 or fewer couples per condition where 25 is the minimum standard and about 70 would be needed to compare two active treatments properly. Small samples both miss real effects and exaggerate the ones they report.
The best-evidenced figures describe termination, and the question is about durability. Follow-ups beyond two years exist for a handful of trials. Whether the components that modern therapies include specifically to prevent later deterioration actually prevent it is, in the words of the review that raises it, unknown.
Deterioration and dissolution are reported as one category. A relationship ending and a relationship getting worse are different outcomes, sometimes opposite ones, and the trials generally do not separate them.
Several key trials are paywalled. Where that is the case here, the figures quoted come from the published abstract or from an open review that reports them, and the source list says which is which. No number on this page is taken from another article's summary of a paper.
And none of it is about your relationship. Every percentage here is an average across a group of couples who were recruited, screened, assigned and measured on a schedule, in trials from which situations involving violence, infidelity and infertility were deliberately excluded. A 48% recovery rate is not a 48% chance for any particular pair of people, and the variables that actually predicted relationship quality in the largest study ever run on the question were not traits either person carried in — they were judgments about a relationship that already existed, which is worked through in what builds attraction over time and in what questions can and cannot reveal about compatibility.
Every empirical claim on this page carries a linked source, which is the promise made in our editorial policy; claims this site has had to retract are listed on the corrections page. The rest of this section is under Mindset.
Pulling it together
The short answer is that couple therapy works, moderately, for about half the people who complete it, and that the honest version has three conditions attached.
It works best measured at the moment it ends. The pooled effect on relationship satisfaction is large where untreated distressed couples do not move at all, and greater distress at intake predicted larger gains rather than smaller ones — which disposes of the idea that couples who come late are beyond help — and the companion finding, that most people thinking about ending a marriage never reach a professional at all, is in what a national panel found happened next. Roughly half are meaningfully better at termination, about 40% two years later, about a third at five years, and a real minority, between 4% and 18% depending on the trial, are worse off than when they started. In ordinary clinical practice, rather than in trials, the effect is about half the size and about half of couples never finish.
The branded method is the part to stop worrying about. No model currently meets the strictest classification standard, the methods perform about equally when compared with each other, and what the evidence keeps pointing to is the alliance, the therapist, and both people turning up. The single variable most under a couple's own control is the one the data most consistently rewards.
And the six years was never a finding. It was a sentence, passed along for a quarter of a century, that put a number on being too late. There isn't one.
Frequently asked questions
Does couples therapy actually work?
At the end of treatment, yes, and the effect is large. The biggest meta-analysis — Roddy, Walsh, Rothman, Hatch and Doss, 58 studies, 40 unique samples, 2,092 couples — reports a pre-to-post within-group change in relationship satisfaction of Hedges g = 1.12, while couples assigned to waitlists in the same studies did not significantly improve at g = 0.12. Significant effects also showed up on self-reported and observed communication, emotional intimacy and partner behaviours. Two things complicate it. In ordinary clinical practice the effect is roughly half that size, about d = 0.45 to 0.50 across nine effectiveness studies, and about half of couples do not complete therapy. And the gains fade: five years after behavioural couple therapy, 33% of couples had recovered, 13% had improved but were still distressed, and 54% had deteriorated, dissolved or not changed. The pooled sample is also exclusively opposite-sex couples, which the authors name as their main limitation.
What is the success rate of couples therapy?
It depends entirely on when you measure and what counts as success, which is why the figures in circulation vary so much. The clearest breakdown comes from Bradbury and Bodenmann's review in the Annual Review of Clinical Psychology. At the end of treatment, pooling behavioural and integrative behavioural couple therapy across 130 couples: 48% recovered, 18% improved but still distressed, 21% unchanged, 13% deteriorated. Two years out, behavioural couple therapy across 60 couples: 37% recovered, 23% improved, 15% unchanged, 25% deteriorated or dissolved. Five years out, 61 couples: 33% recovered, 13% improved but still distressed, 54% deteriorated, dissolved or unchanged. The widely quoted 70% figure has a real source — Spengler and colleagues' 2024 meta-analysis of Emotionally Focused Therapy, 20 studies and 332 couples, reporting about 70% symptom-free at the end of treatment. The separate claim that "70 to 80% of couples improve" is usually a garbled version of a different statement: that the average person receiving couple therapy is better off at termination than 70 to 80% of people not receiving treatment, which is an effect size expressed as a percentile, not a proportion of couples.
Is it true that couples wait six years before getting help?
No, and the figure has been traced. It comes from Buongiorno, J. (1992), Wait time until professional treatment in marital therapy, an unpublished master's thesis at the Catholic University of America. Doherty, Harris, Hall and Hubbard went looking for it and reported in the Journal of Marital and Family Therapy in 2021 that they could not obtain a copy from the university, and that the author no longer had one either — but described the study by email: non-divorced couples recruited through three to four private-practice therapists in Washington DC, and fewer than 20 couples. The figure appears to have been first cited in Gottman's The Marriage Clinic in 1999 and from there into books, peer-reviewed journals and thousands of practice websites, under three different author lines. When Doherty and colleagues measured it, in 270 people who had attended couple therapy, the mean interval from the onset of serious problems was 2.68 years, with 64% of them inside two years. A veteran sample found longer — 3.73 years for men and 7.39 for women — but with a standard deviation larger than the mean, which tells you an average is a poor description of this quantity.
Can couples therapy make things worse?
Yes, and there is a published rate. At the end of treatment, between 4% and 18% of couples in these trials are worse off than when they started: 13% pooling two behavioural models across 130 couples, 18% in the behavioural arm specifically, 4% in a small emotionally focused sample, 7% in the online OurRelationship trial. Two years out the figure is 25% to 32%, though that category merges deterioration with relationships that ended, which are not the same outcome. The worrying part is the stickiness: of couples who had deteriorated by the end of treatment, 87% were still in that category five years later. Three caveats. These samples were selected for serious chronic distress, in which the untreated base rate of getting worse is not zero — waitlisted couples in the meta-analysis did not improve. The trials cannot separate getting worse because of therapy from getting worse despite it. And the 2026 systematic review notes in passing that couple therapy can provoke difficult emotions and conflict, which is the active ingredient and the risk at the same time. None of the twelve pages ranking for this question mentions deterioration at all.
Which type of couples therapy is most effective?
On the evidence, probably none of them in particular. Rathgeber and colleagues meta-analysed randomised controlled trials only — 33 studies, 2,730 participants — and found g = 0.53 for behavioural couple therapy and g = 0.73 for emotionally focused at post-test, with no significant difference between the two methods, smaller effects at six months, and g = 0.06 in the twelve-month data available. The 2026 systematic review found couple therapies consistently superior to inactive conditions but mostly equivalent to each other, and names the dodo bird verdict for the pattern. The one exception is old and unreplicated: Snyder and Wills randomised 79 distressed couples to behavioural or insight-oriented therapy, and at four years 38% of the behavioural couples had divorced against 3% of the insight-oriented ones — a result from the 1980s with no new efficacy study of insight-oriented couple therapy published since. What anyone claiming a single evidence-based method is reporting is a classification scheme rather than a finding, and there is more than one scheme: the same trials produce three well-established models under one framework and none under another, the difference being that the first has no sample-size requirement.
Does online couples therapy work?
The online programmes are, by an accident of research funding, better measured than much of the in-person field. Doss and colleagues tested OurRelationship — an eight-hour online programme with four fifteen-minute coaching calls — in 300 heterosexual couples broadly representative of the United States, randomised against a two-month waitlist. Relationship satisfaction came out at d = 0.69, relationship confidence 0.47, negative relationship quality 0.57, plus improvements in depressive and anxious symptoms, perceived health, work functioning and quality of life, concentrated among people who began with a difficulty in that domain. The breakdown was 57% recovered or improved, 36% unchanged, 7% deteriorated — a lower deterioration rate than the 13% to 18% in the in-person behavioural trials. A one-year follow-up of 668 enrollees in a low-income sample found gains maintained with no significant decline and no difference between two different programmes. Two caveats worth keeping: the trials come from the group that developed the programme, and the comparison is against a waitlist rather than against a therapist, so nothing here shows that online is as good as in-person — only that it does considerably more than nothing, cheaply, for a population the in-person system reaches badly.
Is it too late for couples therapy if things are already bad?
The evidence runs against that idea rather than for it. Christensen, Atkins, Baucom and Yi followed 134 chronically and seriously distressed married couples for five years after roughly eight months of treatment — a sample selected for exactly the condition the "too late" story describes. Pre-to-post effect sizes were d = 0.90 and d = 0.71 for the two models; at five years, relative to pre-treatment, d = 1.03 and d = 0.92, with 50.0% and 45.9% showing clinically significant improvement, and 25.7% and 27.9% separated or divorced. Separately, the largest meta-analysis tested whether baseline distress moderated how much couples gained and found that greater distress at intake went with larger gains. The authors of the research note that traced the six-year figure make this their central point: the number's real harm is the unspoken inference that couples who waited cannot be helped, and they worry it deters people from seeking help at all. What the data cannot do is say anything about a particular couple, and the category that gets conflated with lateness — a relationship involving control, fear or violence — is a different question with a different answer.
How many sessions does couples therapy take?
There is no established dose, and anybody quoting a specific number of sessions as the research-backed answer is supplying something the literature does not. What exists are descriptions of what people actually did. In Doherty and colleagues' sample of 371 people who had been in therapy for relationship problems, participants had seen an average of two therapists and attended an average of eight sessions. The trials are much longer: the Christensen comparison ran about eight months, and Snyder and Wills's trial about 20 sessions. The online OurRelationship programme is eight hours of content plus four fifteen-minute calls, and produced a satisfaction effect of d = 0.69 against a waitlist — which at least establishes that substantially less than eight months of conjoint therapy does something measurable. The more useful question than how many sessions is what happens after the last one: gains fade for about half the couples who get them, 54% of couples were deteriorated, dissolved or unchanged five years after behavioural couple therapy, and whether the components therapies include to prevent that actually prevent it is, in the words of the review that raises the point, unknown.
When is couples therapy not the right thing?
There is one category that is not a relationship problem with two authors, and the research on this page is structurally silent about it. Being told who you may see or message; having your phone checked or your location demanded when it is not mutual and freely chosen; being made to account for ordinary hours; money controlled or withheld; anger, punishment or silence arriving as a sanction when you say no; being separated from friends and family a little at a time; any fear of your partner or any physical harm. UK statutory guidance on controlling or coercive behaviour names isolation from support, monitoring of time and movements, control of money and deprivation of independence as recognised markers. The 2026 systematic review states plainly that studies of couples presenting with intimate partner violence were excluded from its evaluation, which means every effect size quoted on this page was measured in couples from whom such situations had been screened out. There is a specialist literature on conjoint treatment where violence is present, but it is a separate question requiring separate screening, and it is not what a general therapist's advertised success rate refers to. Find A Helpline 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.
Sources
What this article draws on. For the broader bibliography see our sources page.
- Doherty, W. J., Harris, S. M., Hall, E. L. & Hubbard, A. K. (2021) — How long do people wait before seeking couples therapy? A research note. Journal of Marital and Family Therapy, 47(4), 882-890 — the paper that traced the six-year figure to an unpublished master's thesis and measured it properly: N = 371 (270 couple therapy, 101 individual therapy for relationship problems), mean interval 2.68 years (SD 2.09), 64% within two years. Paywalled at Wiley.
- Doherty, Harris, Hall & Hubbard (2021) — full text, Dutch translation, open access — Hoe lang wachten mensen voordat zij hulp zoeken voor relatieproblemen? Een onderzoeksnotitie. Gezinstherapie Wereldwijd, 34, 65-75 (2023) — the same research note, translated and published open access under a Creative Commons licence. Every detail quoted on this page about the Buongiorno thesis, the author's 2020 email, the citation forms, the sample characteristics and the full distribution of waiting times was read here rather than from a secondary account.
- Buongiorno, J. (1992) — Wait time until professional treatment in marital therapy. Catholic University of America. Unpublished master's thesis — the entire provenance of the "average couple waits six years" claim. No copy could be obtained by the researchers who went looking, including from the author; the title returns zero records in Europe PMC, searched 8 November 2026. Listed here with no URL because there is nothing to link, which is the point.
- Jarnecke, A. M., Ridings, L. E., Teves, J. B., Petty, K., Bhatia, V. & Libet, J. (2020) — The path to couples therapy: A descriptive analysis on a Veteran sample. Couple and Family Psychology: Research and Practice, 9(2), 73-89 — 97 opposite-sex couples at a VA medical centre; men reported waiting 3.73 years (SD 4.36) and women 7.39 years (SD 11.64), t(29) = -1.83, p = 0.079; longer waits went with more optimistic expectations. Also the paper whose reference list shows the thesis cited as "Notarius & Buongiorno, 1992 … Unpublished paper", at second hand via Gottman. Free full text in PubMed Central, PMC7351137.
- Hubbard, A. & Harris, S. (2020) — A critical review of help-seeking for couples therapy. Contemporary Family Therapy, 42(2), 152-162 — cited here only for its reference list, which lists the thesis as "Buongiorno, J., & Notarius, C. (1992)", one of the two incorrect author lines. Paywalled; the reference list is readable on the publisher's page.
- Roddy, M. K., Walsh, L. M., Rothman, K., Hatch, S. G. & Doss, B. D. (2020) — Meta-analysis of couple therapy: Effects across outcomes, designs, timeframes, and other moderators. Journal of Consulting and Clinical Psychology, 88(7), 583-596 — 58 studies, 40 unique samples, 2,092 couples; pre-to-post satisfaction Hedges g = 1.12 CI [0.92, 1.31] against g = 0.12 CI [-0.04, 0.29] for waitlists; greater baseline distress associated with larger gains; limitation stated by the authors is a sample of exclusively opposite-sex couples. Paywalled; figures quoted from the published abstract.
- Jackson, J. B., Codecá, L., Miller, R. B., Simpson, J. E., O'Leary, A. M. & Renner, S. M. (2026) — Empirically supported couple therapy models for couple relationship distress: A replication and update. Journal of Marital and Family Therapy, 52(4), e70162 — 40 eligible studies, 12 models, Chambless criteria: no model met the standard for strong empirical support; seven modest, five weak, three with insufficient research. Source for the 25-couples-per-condition requirement, the finding that most trials ran 15 or fewer, the ~70 needed to compare two active treatments, the note that the APA Society of Clinical Psychology site still lists EFT as strongly supported on 19 mostly off-population studies, the exclusion of intimate partner violence samples, and the dodo bird discussion. Free full text in PubMed Central, PMC13474905.
- Bradbury, T. N. & Bodenmann, G. (2020) — Interventions for couples. Annual Review of Clinical Psychology, 16, 99-123 — the source for the four-way outcome breakdowns at termination, two years and five years; for the effectiveness-study effect size of d = 0.45-0.50; for the finding that about half of couples in real-world settings do not complete therapy and 30-60% of those who improve relapse; for the 87% of deteriorated couples still deteriorated at five years; and for the table of EFT and BCT meta-analytic effect sizes by number of studies, including Johnson et al. (1999) at d = 1.31 from four studies. Author copy readable in full.
- Lebow, J. & Snyder, D. K. (2022) — Couple therapy in the 2020s: Current status and emerging developments. Family Process, 61(4), 1359-1385 — open access. Source for the statement that the average person receiving couple therapy is better off at termination than 70-80% of individuals not receiving treatment; for deterioration or divorce in roughly 35-50% of couples at 4-5 years, with Snyder's 20% exception unreplicated; for the classification of CBCT, IBCT and EFCT as well-established, which the 2026 Chambless-criteria review contradicts; and for the note that some couple treatments have evidence for intimate partner violence as a separate specialist question.
- Rathgeber, M., Bürkner, P.-C., Schiller, E.-M. & Holling, H. (2019) — The efficacy of emotionally focused couples therapy and behavioral couples therapy: A meta-analysis. Journal of Marital and Family Therapy, 45(3), 447-463 — randomised controlled trials only; 33 studies, 2,730 participants; post-test g = 0.60 overall, 0.53 BCT, 0.73 EFCT; six months 0.44, 0.35, 0.66; twelve months (BCT only) g = 0.06; no significant difference between the two methods; authors flag potential publication bias.
- Spengler, P. M., Lee, N. A., Wiebe, S. A. & Wittenborn, A. K. (2024) — A comprehensive meta-analysis on the efficacy of emotionally focused couple therapy. Couple and Family Psychology: Research and Practice, 13(2), 81-99 — 20 studies and 332 couples; pre-to-post d = 0.93 CI [0.75, 1.12], EFT versus viable alternative couple interventions d = 0.44 CI [0.03, 0.85], pre-to-follow-up d = 0.86 CI [0.56, 1.15]; the source of the figure that about 70% of couples are symptom free at the end of treatment. Paywalled; quoted from the published abstract.
- Johnson, S. M., Hunsley, J., Greenberg, L. & Schindler, D. (1999) — Emotionally focused couples therapy: Status and challenges. Clinical Psychology: Science and Practice, 6(1), 67-79 — the origin of the largest effect size anyone reports for EFT, d = 1.31, pooled from four studies, and the ancestor of the 70-73% and 86-90% figures that circulate without a denominator. Paywalled, and its published abstract reports no percentage; the four-study effect size quoted here is from Bradbury and Bodenmann's table rather than from the paper.
- Christensen, A., Atkins, D. C., Baucom, B. & Yi, J. (2010) — Marital status and satisfaction five years following a randomized clinical trial comparing traditional versus integrative behavioral couple therapy. Journal of Consulting and Clinical Psychology, 78(2), 225-235 — 134 chronically and seriously distressed married couples; pre-to-post d = 0.90 (IBCT) and 0.71 (TBCT); at five years d = 1.03 and 0.92, with 50.0% and 45.9% clinically significantly improved and 25.7% and 27.9% separated or divorced; IBCT's advantage significant through two years then converging. Paywalled; quoted from the published abstract.
- Snyder, D. K. (research summary, Texas A&M University) — Research on couple therapy — the author's own account of the trial that randomised 79 distressed couples to behavioural therapy, insight-oriented therapy or a waitlist: both treatments produced statistically and clinically significant gains at termination after about 20 sessions, and at four years 38% of the behavioural couples had divorced against 3% of the insight-oriented couples. The primary papers are Snyder & Wills (1989) and Snyder, Wills & Grady-Fletcher (1991), both in the Journal of Consulting and Clinical Psychology and both paywalled.
- Roesler, C. (2020) — Effectiveness of couple therapy in practice settings and identification of potential predictors for different outcomes: Results of a German nationwide naturalistic study. Family Process, 59(2), 390-408 — 554 couples, using the same methodology as two earlier German studies and finding remarkable consistency across all three: improvements in practice settings are about half the effect sizes reached in efficacy trials, and there is no improvement in up to 50% of couples. Paywalled; quoted from the published abstract.
- Doss, B. D., Cicila, L. N., Georgia, E. J., Roddy, M. K., Nowlan, K. M., Benson, L. A. & Christensen, A. (2016) — A randomized controlled trial of the web-based OurRelationship program: Effects on relationship and individual functioning. Journal of Consulting and Clinical Psychology, 84(4), 285-296 — 300 heterosexual couples (N = 600), broadly representative of the US on race, ethnicity and education, randomised against a two-month waitlist; satisfaction d = 0.69, confidence 0.47, negative quality 0.57; depressive symptoms 0.71 and anxious symptoms 0.94, especially among those beginning with difficulty in that domain. Free full text in PubMed Central, PMC4804631.
- Roddy, M. K., Knopp, K., Georgia Salivar, E. & Doss, B. D. (2021) — Maintenance of relationship and individual functioning gains following online relationship programs for low-income couples. Family Process, 60(1), 102-118 — 668 individuals contacted one year after completing either OurRelationship or a behavioural online programme; relationship gains maintained with no significant linear or quadratic change, medium-to-large within-group effects from pre-treatment to one year, 68.3% still with the same partner, and no significant differences between the two programmes. Paywalled; quoted from the published abstract.
- Literature and page-one counts, run on 8 November 2026 — Written out so they can be re-run rather than taken on trust. PubMed, title-and-abstract: "couple therapy" OR "couples therapy", 1,083 records, of which 119 carry the randomized-controlled-trial publication type and 79 also mention relationship distress; Europe PMC, same phrases in title or abstract, 993. Europe PMC for the phrase "Wait time until professional treatment": zero records. Twelve pages returned for "does couples therapy work" were downloaded as plain text and searched: zero link a DOI, zero give the sample size of a study they describe, zero contain the word deterioration in any form, one contains "et al", nine mention Gottman, eight mention Emotionally Focused Therapy, seven quote a success figure in the seventies, and four repeat the six-year waiting claim. Google returned its unusual-traffic challenge for the People also ask block on the same day, so that block is not reported here.
- Home Office (UK) — Controlling or coercive behaviour in an intimate or family relationship: statutory guidance framework — the source for the markers listed in the section on when couple therapy is not the room.
- Find A Helpline — Free, confidential support services listed by country — the referral at the top and near the end of this article, which covers the situation that nothing else on this page addresses.
Keep reading
- MindsetDo the Five Love Languages Work? What the Studies Actually MeasuredIn 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.
- MindsetDoes the No Contact Rule Work? What Has Actually Been Measured About Cutting ContactThe phrase returns nothing in PubMed. The one diary study the pages half-cite is 58 people, and nearly half of exes reconcile with no technique at all.
- MindsetWhen One Partner Won't Go to Couples Therapy: Who Refuses, and What Working on It Alone DoesIn a clinical sample of 913 couples, 40% arrived with mismatched agendas. The one trial of doing the work alone moved the person, not the relationship.
- DatingHow Many Dates Before You Know? What the Research Can Actually TimeNobody has counted. In the one study that plotted whole relationships, the ones that lasted and the ones that did not looked identical through the first kiss.
- DatingWhat Is the Ick? One Study, Three Different Sample SizesExactly one empirical study of the dating ick exists. Its abstract says 138 people, every report of it says 125, and it missed its own power target.
- DatingDo Opposites Attract? The Largest Study of Couples, and the Number It Actually ReportsThe study everyone quotes measured 133 traits in up to 79,074 couples. Its mean partner correlation is 0.19, and not one page on the first screen reports it.