The 72-Hour Rule in Dating: Is There Anything Behind It?

The rule was three friends arguing in a film

Start with where it came from, because the origin explains most of the problem.

The three-day rule entered dating culture through Swingers, the 1996 film written by Jon Favreau. Mike has a phone number and wants to know when to call. His friends work it out at the table: a day, then tomorrow, then two days, which Sue calls the industry standard. Trent counters that everyone in town already waits two days, so — in the film's phrasing — three days is kind of money. Sue disagrees on the spot: two is enough not to look anxious.

That is the canonical source. Not a study, not a clinician, not a survey of what works. A scene in which three characters invent a number in real time and do not agree on it, in a film whose joke is that the guy following the advice is miserable. The US matchmaking company Three Day Rule took its name from that film directly.

Since then the phrase has drifted, and it now names several different things that have nothing to do with each other. In its original sense it is a waiting period before contacting someone after a date. In a second sense that circulates widely, it is a rule about waiting seventy-two hours before reacting to something that upset you in a relationship. In a third, it is a no-contact window after a breakup — although the research on recovery measures something other than elapsed time, as the signs you have actually moved on sets out. These are separate pieces of advice with separate logic, and they share a number only because seventy-two hours sounds authoritative.

None of them was derived from evidence. That is not a criticism of anyone in particular — most of the dating advice on this site's own Dating shelf predates the research that would test it. It is just worth knowing that the number arrived first and the reasons were reverse-engineered afterwards. And it is the standing pattern in this genre rather than one bad film: the 3-6-9 rule and the 777 rule arrived on social video three decades later and acquired their reasons in exactly the same order.

Somebody finally ran the experiment, and the wait lost

For nearly thirty years the rule circulated without a direct test. In 2026 it got one.

Lars Teichmann, Hannes Petrowsky, Lea Boecker, Meikel Soliman and David Loschelder, at Leuphana University Lüneburg, published a preregistered experiment in the Journal of Social and Personal Relationships on exactly this question: how does the timing of the first text after a first date change what the person on the other end wants. They built the manipulation from a pre-study of 100 US and UK adults, who put the ideal moment at about six hours after the date (M = 5.86 hours), called anything under roughly twenty minutes too soon, and called anything past about forty hours too late.

The main experiment ran on 543 heterosexual participants from the US and UK, recruited through Prolific in April 2024 — 270 women and 273 men, mean age 40.3. The target of 508 came from an a priori power analysis, and the hypotheses, design and analyses were registered in advance. Participants imagined a neutral first date and then received the follow-up text at one of three moments: immediately after saying goodbye, the next morning, or after two days.

Relationship intentions differed by condition, F(2, 537) = 11.27, p < .001. The next morning came out highest (M = 6.15, SD = 1.26), the immediate text second (M = 5.80, SD = 1.33) and the two-day delay last (M = 5.50, SD = 1.59), with all three conditions differing significantly from each other. Fitted as a curve, the shape was an inverted U: the quadratic term was b = −0.403, p < .001.

The other two outcomes did not even give the delay that much. On perceived chemistry, F(2, 537) = 15.57, p < .001, the next morning (M = 6.49) and the immediate text (M = 6.41) were statistically indistinguishable from each other (p = .555), while two days sat well below both (M = 5.77, p < .001 against each). Motivation to actually do something about it behaved the same way: 6.86 next morning, 6.80 immediately, 6.12 after two days.

So the shape of the finding is a small bonus for a short delay on one measure, and a clear penalty for a long one on all three. The rule's core claim — that waiting longer buys you something — is the part that failed.

An inverted U also turns up one step earlier in the process, in the messaging that happens before two people have met at all, where a national survey of online daters found that both very little and a great deal of pre-meeting contact predicted a worse first meeting than something in between. That study, and the rest of what has been measured about whether these platforms deliver, is in whether dating apps actually work.

The longest wait anyone has tested is two days

Here is the caveat that matters most, and the researchers raise it themselves rather than leaving it for someone else to notice.

Their long condition was two days, chosen because their pre-study put the average "too late" threshold at 40.42 hours. They flag in their limitations that the standard deviation on that estimate was enormous (SD = 34.08), that people disagree wildly about when a message becomes late, and that their design therefore may not capture reactions to longer delays — naming the popular three-day rule specifically. Their expectation, stated as an expectation rather than a result, is that three days would have produced lower relationship intentions still.

Take that at face value and no further. The honest sentence is this: the longest delay ever tested was two days, it lost on every measure, and the researchers who tested it think three days would be worse. Nobody has run three days. Anyone telling you the seventy-two-hour rule is scientifically disproven is going past the data, and anyone telling you it is scientifically supported has no data at all.

The other limitations are worth stating too, because they cut against reading this as the last word. Participants imagined a date rather than having one. The content of the message was deliberately left blank so that timing could be isolated, which means the study cannot tell you how a well-judged three-day message compares with a clumsy same-night one. The date's quality was held neutral to avoid ceiling effects, so the study says nothing about what happens after an unusually good evening — or, for that matter, after one that went wrong in any of the ordinary ways. And the sample was US and UK adults on Prolific, which is not the world.

The wait fails for a reason the rule never anticipated

The interesting part of that study is not the ranking. It is the mechanism, because the rule's own premise was measured and mostly came out backwards.

The premise is that texting quickly makes you look needy, and looking needy costs you. The first half held up, strongly. Perceived neediness tracked timing almost perfectly, F(2, 537) = 72.80, with the largest effect size in the paper (ηp² = .213): initiators who texted immediately were rated most needy (M = 6.02), next morning less so (M = 5.05), two days least of all (M = 4.05), with every contrast significant. If your worry is that an eager text reads as eager, you are right.

The second half did not hold. Perceived neediness did not relate to relationship intentions. The thing the rule is built to avoid was reliably produced by fast texting and then turned out not to be charged for.

What was charged for was the delay, through two routes. The first is reciprocity — how much you conclude the other person likes you. F(2, 537) = 73.50, ηp² = .215: immediate texting read as most liking (M = 6.54), next morning slightly less (M = 6.10), two days far less (M = 4.86). The second is reliability, F(2, 537) = 24.32: an immediate or next-morning text read as reliable (5.48 and 5.56), two days did not (4.62). Both carried significant indirect effects in the mediation models — for women, reciprocity when comparing next morning against two days, b = −0.66, 95% CI [−0.915, −0.429]; for men the same route, b = −0.42, CI [−0.689, −0.210].

And the scarcity premise — that being harder to reach makes you look like a better catch — produced nothing at all. Relative mate value did not move with timing, F(2, 537) = 0.18, p = .837. Not a small effect. Not a marginal one. Nothing.

Read as a whole, the study says the rule correctly identified a real perception and then priced it wrong. You do look less needy after two days. You also look like someone who is not particularly interested and cannot be relied on, and those are the two things the other person is actually acting on. The same asymmetry runs through what evidence there is on texting habits generally: the variable people audit is rarely the variable that carries the weight. The rest of what this site has on messages is under Texting.

The hard-to-get literature opens with five failures

Behind the timing rule sits a bigger claim — that being harder to get makes you more wanted. It has been tested longer than the rule has existed, and the first serious attempt is a useful thing to know about.

In 1973, Elaine Walster, G. William Walster, Jane Allyn Piliavin and Lynn Schmidt published a paper in the Journal of Personality and Social Psychology whose abstract is unusually blunt. Five experiments were run to demonstrate that people value hard-to-get dates more than easy-to-get ones. All five failed. The sixth found something, and what it found was narrower than the hypothesis they had started with: what men preferred was a woman who was hard for other men to get but easy for them specifically — selectively hard to get. Uniformly hard to get, meaning unenthusiastic about dating anyone, did no better than uniformly easy to get. That experiment ran on 71 male summer-session students.

Two things follow. The first is about the finding: even in its successful version, the effect is not about withholding. It is about being seen as selective while being unambiguous with one person. Delaying a text does the opposite — it is ambiguity aimed at the one person, with no information conveyed about anybody else.

The second is about the literature. Fifty-odd years on, a 1973 paper with 71 male students is still routinely cited as the evidence base for hard-to-get. When the strongest classic result in an area is one experiment that worked after five that did not, the correct posture toward everything built on top of it is caution. That is the same caution this site applies to reading anyone's signals as though they were data.

Uncertainty is not the same thing as a schedule

The modern version of the argument does not rely on Walster. It relies on a 2011 experiment that gets cited whenever someone wants to justify keeping a person guessing, and it is worth reading closely rather than in summary.

Erin Whitchurch, Timothy Wilson and Daniel Gilbert told 47 female undergraduates at the University of Virginia that male students at other universities had viewed their Facebook profiles and rated them. Each participant then saw four fictitious male profiles, told either that these were the men who had liked her most, or the men who had rated her average, or — in the uncertain condition — that it might be either and neither she nor the experimenter knew which.

Attraction differed by condition, F(2, 44) = 15.06, p < .001. Being liked best beat being liked averagely, t(44) = 3.52, p = .001, which is the ordinary reciprocity effect. And the uncertain condition beat being liked best, t(44) = 2.07, p = .04. That is the headline: women were more drawn to men when there was a fifty per cent chance those men liked them best than when it was certain.

Now the parts that rarely survive the summary. The proposed mechanism was that uncertainty makes you think about the person more — and the overall effect of condition on how much participants thought about the men was not significant, F(2, 43) = 2.14, p = .13. The mediation bootstrap produced a 95% confidence interval of [−.03, .40], which includes zero, and the authors write in the paper that the evidence for mediation is tentative. The whole study is 47 people split across three conditions, with a key contrast at p = .04, and I can find no direct replication.

And even taken at full strength, it does not license the rule. What was manipulated was uncertainty about how much someone liked the participant, with no information given about that person's availability, choosiness, or schedule. A delayed text is not that. It is a specific behaviour with its own meanings attached, and reading it as a signal runs into the same trouble as reading any other quiet behaviour as a signal. When the Leuphana team measured the closest available analogue — how intensely people thought about their date while waiting — timing did not move it (p = .060).

Wanting and liking are not the same currency

The most careful work on hard-to-get splits the outcome in two, and the split is the whole story.

Xianchi Dai, Ping Dong and Jayson Jia, publishing in the Journal of Experimental Psychology: General in 2014, argued that playing hard to get pushes two responses in opposite directions: it can raise wanting, the motivational pull toward pursuing someone, while lowering liking, the plain affective judgement of whether you enjoy them. And they predicted the wanting half only appears if the person is already psychologically committed to pursuing you.

They ran a vignette study and a speed-dating experiment, both with male students at a Hong Kong university and a female confederate playing warm or aloof. In the published supplementary tables, liking is lower under the aloof strategy in every single cell. Feeling toward the partner: 3.60 versus 5.63 without commitment, 4.00 versus 5.40 with it. Enjoyment of the date: 3.20 versus 5.44, and 3.60 versus 5.47. There is no condition in which being harder to get made anyone better liked.

Wanting flipped only under commitment, and only just. Motivation to make a good impression ran 3.87 for hard-to-get against 5.56 for easy-to-get among participants with no commitment — a large penalty — and reversed to 5.40 against 4.53 among those who had chosen their partner. In the vignette, willingness to spend on a gift told the same story: HK$314 for the aloof partner among men who had chosen her, against HK$172 for the warm one, but HK$67 against HK$114 when she had been randomly assigned.

Two caveats before anyone builds a strategy on this. The samples are small and entirely male, at one university, and the effect is defined by an interaction rather than a main effect, which is the kind of result that most often fails to replicate. And the boundary condition is fatal for the rule in practice: whatever this buys, it only exists among people who are already committed to pursuing you. On a first date, by definition, nobody is. The rule prescribes for strangers a tactic whose only demonstrated effect requires prior investment — and charges the liking cost in every case.

Where the intermittent reinforcement argument actually comes from

The most confident version of the pro-waiting argument is neuroscientific in tone. Unpredictable rewards, it says, produce the strongest attachment; the slot machine is the model; make your availability intermittent and you become the slot machine.

The underlying science is real and old. C. B. Ferster and B. F. Skinner's Schedules of Reinforcement, published in 1957, is the systematic mapping of how different patterns of reward shape behaviour, and Abram Amsel's 1958 frustration theory in Psychological Bulletin set out why partially reinforced responses are harder to extinguish than continuously reinforced ones. That effect is robust. It has been demonstrated thousands of times.

What it says, though, is narrower than the dating version. It is a claim about how persistently a learned response continues once the reward stops — measured in pigeons pecking keys and rats running alleys. It is not a claim that unpredictable people are more likeable, more desirable, or better partners. Persistence and affection are different outcomes, and the schedules literature only ever measured the first.

Where intermittent reinforcement has actually been studied in human intimate relationships, the setting is not courtship. Donald Dutton and Susan Painter's test of traumatic bonding theory assessed 75 women who had recently left abusive relationships, 50 of which had involved physical violence, and found that the intermittency of maltreatment and the power differential between partners predicted how strongly a woman remained attached to a former partner — measured on leaving and again six months later. That is the human literature on alternating warmth and withdrawal: an explanation of why leaving is hard, not a technique.

Which is the point at which this stops being about texting. If you are reading advice that tells you to make your attention unpredictable so that someone becomes unable to stop thinking about you, the mechanism being described is the one above. Doing it deliberately to another person is not a dating tactic; it is the pattern that the coercive-control literature describes, and the same is true of the surveillance that jealousy escalates into. If it is being done to you — warmth and withdrawal on a schedule you cannot predict, a growing sense that you have to earn contact — that is worth naming as what it is, and the helplines in the sources at the foot of this page exist for exactly that conversation.

Who pays for the wait, and what the evidence does and does not say

A three-day silence is not experienced equally. That much is obvious. What the research supports about it is narrower than the popular version.

What is well documented is that people differ in how they interpret ambiguous behaviour from a partner. Nancy Collins asked 135 undergraduates — 82 women and 53 men — to imagine a dating partner behaving in ways that leave the reason open, including not responding when they reached out. People higher on anxiety about being unloved gave explanations that read the behaviour as rejection and as a sign the relationship was in trouble, and rated conflict as more likely to follow; the correlations with the negative explanation content ran from about −.18 to −.34. The two-dimension framework those measures sit in — attachment anxiety and avoidance as continuous dimensions rather than types — comes from Brennan, Clark and Shaver's 1998 work, which is also the basis of our own attachment style test.

Here is the honest complication. The Leuphana experiment measured trait attachment as a moderator of the text-timing effect and found nothing: all ps ≥ .634. Uncertainty avoidance did not moderate it either. So the defensible claim is not that anxious people react worse to a delayed text — the one study that tested that directly came back null, and the authors suggest priming rather than measuring attachment as a next step. The defensible claim is the more general one: how a person explains an ambiguous silence varies systematically with what they already believe about being wanted, and the explanation is where the damage happens.

Which is worth sitting with if you are on the receiving end of a long gap and building a case out of it. The gap is one observation. What you are doing with it is a habit, and the difference between caring about someone and being activated by them is mostly visible in what you do with exactly this kind of ambiguity.

Silence is not the absence of a message

There is a small research literature on what pauses communicate in written media, and its central point is the one the rule gets wrong.

Yoram Kalman and Sheizaf Rafaeli studied response latency as a chronemic cue — timing treated as nonverbal communication. In their experiment, managers evaluated job candidates whose replies to an email came after one day, after two weeks, or never within a month, and the effect of the delay on how the candidate was rated depended on what the evaluator had expected of that candidate in the first place. It is a hiring study, not a dating one, and it should be read as a demonstration that latency carries social information rather than as a finding about romance. But the structure transfers: a delay is not neutral airtime. It is a message, and its content is supplied by the receiver's prior expectations.

Modern messaging makes that worse in a specific way. When the platform tells the other person that you read something and did not answer, the delay stops being ambiguous about whether it was a decision. Shinichi Kato and colleagues found that negative emotions — sadness, anxiety, anger, guilt — arose after markedly shorter waits when the sender could see the message had been read than when it still showed as unread. The 1996 version of the three-day rule was played on a landline with no receipts. That is not the game anyone is playing now.

Which returns to the practical point. In 1996 a three-day gap was genuinely uninformative — you might have been out, the message might not have reached you. Today the same gap is legible as a choice, and the evidence on what it gets read as is the mediation data from the Leuphana study: not interested, and not reliable. Our piece on what phone behaviour on a date actually signals covers a close relative of this problem, where the thing that predicted anything was not the behaviour but the conclusion drawn from it.

The version of waiting that survives all this

None of the above is an argument for replying instantly to everything, and it is not an argument that the impulse behind the rule is stupid. Two versions of waiting come out of this intact, and it is worth being precise about why.

The first is deciding. If a date was ambiguous — pleasant, not compelling, you genuinely do not know — then a pause before writing is not a tactic, it is the time it takes to work out what you actually want to say. A same-night message sent to be polite can commit you to a warmth you do not feel, and unwinding that later costs more than the delay would have. There is no study on this. I am not going to pretend there is. It is a reasonable thing to do that nobody has measured.

The second is regulation. If you would be writing from anger, exhaustion or a spike of anxiety, waiting until that has passed is not impression management; it is not sending a message you would want back. The research on repair after conflict is about a different situation, but the same distinction applies: a pause in service of a better message is a different object from a pause in service of a manufactured impression.

What does not survive is the third version, the one the rule is actually about — waiting a fixed period, when you already know what you want to say, in order to appear less interested than you are. That is the specific purchase the one experiment on the question says is not available. You will succeed in looking less needy, and nobody was charging you for that. You will also be read as less interested and less reliable, which is what people were acting on.

And if the delay is being used to find out whether someone can be made to want you more by being made uncomfortable, that is worth naming plainly. It is the difference between having an effect on someone and doing something to them, and it will not tell you what you were hoping to learn. What it will tell you is how a person behaves when they are unsure whether you like them, which is a fact about their history rather than about their feelings for you.

The wider version of the same folklore — that absence itself makes somebody want you more — has its own body of evidence, and it splits the same way: separation genuinely raises how positively people rate a partner, and it does so by withholding the information that would revise the rating. That is the subject of what distance actually does to attraction.