Zeigarnik Effect: Why Unfinished Tasks Stick

Introduction

Almost every article ever written about the Zeigarnik effect repeats the same sentence. Unfinished tasks are remembered roughly twice as well as finished ones. It appears in psychology textbooks, in productivity books, in design blogs, in medical center health pages. It is one of the most quoted findings in popular psychology.

It is also, as a general claim about human memory, wrong.

In July 2025, two researchers at the University of Bern published the first meta-analysis ever conducted on the Zeigarnik effect and its lesser known sibling. Romain Ghibellini and Beat Meier screened 1,455 publications, narrowed them to 59 that met their criteria, and pooled the numbers [1]. The ratio of recall for interrupted versus completed tasks came out at 0.99. Interrupted tasks accounted for 49.16 percent of everything recalled, which is slightly under half. The overall effect size was a Cohen's d of 0.15, small enough that most researchers would treat it as noise.

Their conclusion was blunt. The Zeigarnik effect lacks universal validity.

So why does the idea feel so obviously true? Why does a half written message nag at you through an entire meeting? Something real is happening. It just is not the thing the famous headline describes. The pull to go back and finish is genuine and measurable. The memory boost is not. Untangling those two claims is the work of this article, and it turns out to be one of the more interesting detective stories in the history of memory research.

The Berlin Restaurant That Never Happened The Way You Heard It

Start with the origin story, because almost every version of it circulating online contains at least one invented detail.

The version most readers know goes like this. Bluma Zeigarnik sat in a café in Vienna and marveled at a waiter who could recite long orders from memory. She left, realized she had forgotten her jacket, went back, and the waiter no longer recognized her. Intrigued, she designed an experiment.

Colin MacLeod, a memory researcher at the University of Waterloo, went back through the primary sources in 2020 and reconstructed what actually happened [3]. The setting was Berlin, not Vienna. The person who noticed the waiter was Kurt Lewin, Zeigarnik's doctoral supervisor, not Zeigarnik herself. And the forgotten jacket does not appear anywhere in the documented record. It seems to have been added somewhere along the way and then copied endlessly.

Two credible accounts survive, and they disagree on a detail. Edwin Boring's history of experimental psychology records that the waiter could recall the running total of an unpaid table and lost it the moment the bill was settled. Donald MacKinnon, who was actually present at the table, told a different version through Alfred Marrow's biography of Lewin. In that account the waiter remembered what each person had ordered, and after payment simply said he no longer knew, because the bill had been paid.

Notice what both versions share. The trigger was payment. Not the food arriving. Closing the account, not delivering the plates.

That distinction matters more than it looks. Lewin had spent the mid 1920s building a theory of motivation on a physical metaphor. Forming an intention, he argued, sets up what he called a quasi need, a tension system inside the mind that behaves something like a charged battery [6]. The system stays charged while the goal is unmet. Reaching the goal discharges it. The waiter's memory was not a curiosity about waiters. It was a visible fingerprint of a tension system going flat.

Zeigarnik's job was to turn that metaphor into something a laboratory could count.

What Zeigarnik Actually Did In 1927

Her dissertation ran to 85 pages in the journal Psychologische Forschung, and the design is far more careful than the one sentence summary suggests [4].

The main experiment used 32 adults. Each of them worked through 22 short tasks, presented one after another in an individual session. The tasks were deliberately varied and physical. Winding thread. Folding paper. Modelling a figure from clay. A column of multiplication. Drawing a vase. Counting backwards. Stringing beads. Building something out of cardboard. Each was designed to take roughly three to five minutes, and participants were told to complete each one as quickly and correctly as possible.

About half were allowed to finish. The other half were interrupted, and the timing of the interruption was the crucial variable. Zeigarnik cut people off at the moment they appeared most absorbed, sliding the next task in front of them with a simple instruction to move on. The interruption was never announced as part of the design. It just happened, the way interruptions happen.

After the full set, she asked each person to list from memory the tasks they had worked on.

She ran four main experiments and several smaller ones. The second used 15 adults with a near duplicate set of tasks. The third tested 47 adults in a group setting with 18 tasks. The fourth tested 45 adolescents with a mean age of 14. Further experiments with 12 participants each tested alternative explanations, including whether the shock of interruption itself was doing the work, and whether telling people they would later resume changed anything.

Across the four main experiments, the ratio of interrupted to completed recall hovered close to 2.0 [3]. Roughly four in five participants recalled interrupted tasks better. She also noticed two patterns that later research would keep rediscovering. Interruption close to the finish line hit harder than interruption near the start. And participants she described as ambitious showed a stronger effect than those who did not seem invested.

There is one methodological detail worth knowing, because it explains a lot of what came later. Zeigarnik counted only the recall responses produced before a participant's first hesitation. Everything after the pause she treated as a different kind of remembering, not the spontaneous kind her tension hypothesis predicted. She reported that the restricted count did not differ meaningfully from the full count. Later reviewers were less sure, since interrupted tasks also tended to be named first, and being named first is its own advantage.

Her interpretation was never really about memory. It was about tension. The recall numbers were evidence for the tension, not the point of it.

The study landed at exactly the right moment. Gestalt psychology was ascendant, Lewin's laboratory was attracting an unusually talented group of graduate students, and here was a clean demonstration that an intention has momentum. Within a year, one of her labmates would extend it in a direction nobody expected.

The Number That Broke: A Century Of Failed Replications

The trouble started almost immediately.

In 1930, Walter Schlote published the first serious attempt to reproduce the finding and could not. Over the following decades the pattern repeated with unusual consistency. Some studies found the effect. Some found nothing. Some found the opposite. And crucially, which result you got seemed to depend less on memory than on the mood of the room.

Alfred Marrow made that dependency explicit in a pair of 1938 papers that remain among the most revealing in the whole literature [10]. He ran the same interrupted task procedure three times with different framing. In the first, participants got a neutral description of what was happening, and the memory advantage for interrupted tasks was minor. In the second, American students were told the study was replicating German work and that preliminary analysis showed the Americans performing better. The effect jumped sharply. In the third, the same students were told the Germans had done better, and the effect collapsed [11].

Nothing about the tasks changed. Only the story people were told about themselves.

Other researchers found the same fragility from other angles. Saul Rosenzweig showed that when failure threatened the ego, recall flipped toward completed tasks [13]. Alfred Glixman found the same reversal under stress [12]. Albert Caron and Michael Wallach went further and argued the difference was one of learning rather than retrieval [17]. William Prentice discovered that what happened after the tasks mattered too, with the advantage appearing mainly when the following activity was cognitively easy and reversing when it was demanding [14]. David Hays reported a related pattern with interpolated tasks of different complexity [16]. Thelma Alper found that personality moderated everything [15]. Donald Green found that volunteers showed a stronger effect than students required to take part [19], and Bernard Weiner reported that even the sex of a competitor shifted the result [20].

John Atkinson, who appears to have coined the phrase "Zeigarnik effect" in its modern form, showed in 1953 that the situation and the person interact [8]. Participants high in achievement motivation showed a strong effect in a competitive atmosphere. Participants low in achievement motivation showed the strongest effect in a relaxed one. Two groups, two opposite conditions, the same phenomenon.

By 1964, Earl Butterfield reviewed everything and delivered a verdict that has aged extremely well. The effect, he wrote, is far from being the invariable result in the interrupted task paradigm, and frequently more completed than uncompleted tasks are recalled [7].

Then came the most thorough attempt of all. Annie van Bergen spent her doctoral work at the University of Amsterdam trying to reproduce the original, running the paradigm repeatedly and varying conditions systematically. Her 1968 monograph reported a ratio of 0.88, meaning completed tasks were remembered slightly better, and fewer than a third of her replication attempts produced the effect at all [9]. She recommended the field stop treating it as a problem worth studying.

Meanwhile the textbooks kept printing the number two.

What is striking about that century is how little the popular account changed while the evidence eroded underneath it. Willem Claeys tried to rescue the situation in 1969 by proposing two competing mechanisms, one pulling toward the unfinished and one pulling toward remembered success, with personality determining which won [22]. It was an elegant idea. It was also never followed up, though a later study found that fear of failure moderated the effect in the direction his account predicted [25]. Others tested whether the effect held in groups it had never been examined in, including adults with intellectual disability [27].

0.99: The Meta-Analysis That Settled It

Nobody had ever pooled the numbers. That is the strange thing. A finding taught in undergraduate psychology courses for ninety years had never been subjected to the standard tool for deciding whether an effect is real.

Ghibellini and Meier fixed that. Their search ran on 14 September 2023 across two databases, PsycInfo and PSYINDEX, using eight search terms covering both effects and the interrupted task paradigm. That returned 1,455 publications, which became 1,349 after removing duplicates. Title and abstract screening cut it to 124. Full text screening left 59, of which 38 studied the Zeigarnik effect, 20 studied resumption, and one studied both. Agreement between raters on a random sample of 60 studies was perfect [1].

Then they did something unusual, and it explains a great deal of the confusion in the older literature. They calculated three different measures, because the field had used three different ones without noticing how much they diverge.

The first is the measure Zeigarnik used herself. Compute the ratio of interrupted to completed recall separately for each participant, then average those ratios. It sounds sensible. It is also extremely sensitive to outliers, because one participant who recalls eight interrupted and one completed task produces a ratio of eight, which drags the average up regardless of what everyone else did. Pooled across seven publications, this measure gives 1.13. Drop Zeigarnik's own data and it falls to 1.09.

The second measure averages recall for each task type first, then takes the ratio. This is the version most later researchers used. Across 38 publications it comes out at 0.99. Remove Zeigarnik's original data and it stays at 0.99.

The third measure, introduced by Marrow back in 1938, asks what proportion of everything recalled was an interrupted task. If there is no effect, that number should be 50 percent. Across 14 publications it is 49.43 percent. Without Zeigarnik's data, 49.16 percent. Slightly under half. The authors recommend this measure as the least distorted of the three.

The pooled effect size, calculated from the eight publications that reported enough per participant detail, was a Cohen's d of 0.15.

Then they broke it down by condition, and the breakdown is where the real story lives.

Read that table twice. The only condition where the Zeigarnik effect appears at all is the one where the experimenters deliberately made the room feel casual. In a neutral setting it is absent. In a setting framed as a test of intelligence, the direction reverses and people remember what they finished. Achievement motivation, the personality variable that dominated this literature for four decades, produces no clear pattern of its own.

Here is the same data plotted against the line where no effect would sit.

Two bars tower over the rest. Both come from the 1920s and 1930s. Everything measured since then clusters within a hair of 1.0, and one of the most careful attempts lands below it.

The authors offer a historical explanation that is more interesting than a simple accusation of bad data. Zeigarnik reported that her participants worked conscientiously out of a sense of duty toward the experimenter, out of ambition, or out of interest in the tasks themselves. In 1920s European universities, the authority of a professor was extraordinary. The paper quotes a 1960 observation that professors were regarded not as human beings but as gods. Under those conditions, the situational pressure Marrow later had to manufacture with instructions was simply built into the room.

Modern laboratories have none of that. And the second ingredient, deep task involvement, has become harder to find for reasons the authors also document: notifications, email, and a habit of splitting attention across several things at once all erode the capacity to become absorbed in a three minute task [59] [60] [61].

What does this mean? The Zeigarnik effect may have been partly an artifact of its own experimental culture. Not fraud, not error, but a real phenomenon that required a kind of engagement and deference that has largely disappeared.

The Effect That Actually Survived

While Zeigarnik was measuring memory, a colleague in the same laboratory was measuring something else entirely, and hers held up.

Maria Ovsiankina published in 1928 [5]. Her question was not whether people remember interrupted tasks better. It was whether they go back and finish them. She interrupted participants, then engineered a low pressure gap where returning to the abandoned task was possible but nobody suggested it. They returned.

The meta-analysis pooled 21 publications on this question. The resumption rate came out at 67 percent, against a chance baseline of 50 percent. Removing Ovsiankina's own data barely moves it, to 66.79 percent [1].

Sixty seven percent is not a subtle finding. It is the kind of number that shows up across wildly different populations and methods. Wera Mahler found that people resumed even when offered an appealing alternative activity [63]. Käte Lissner found the pull weakened only as the substitute task grew more similar to the interrupted one, which is exactly what a goal based account predicts [64]. In 2020, a computerized version told participants a network problem required a sixty second wait, then quietly allowed them to dismiss the prompt and continue. They dismissed it and continued [29].

Two findings complicate the picture in useful ways. Kenneth McGraw and Jirina Fiala discovered that paying participants for their time reduced resumption rather than increasing it [23]. Their reading was that money replaced the internal goal of finishing with an external goal of participating, and once you have been paid, the task no longer needs closing. Johnmarshall Reeve and colleagues then showed that resumption and intrinsic motivation, while related, are not the same thing, since participants kept behaving in intrinsically motivated ways even after finishing [24].

Framing shifts it too. Nira Liberman and colleagues found that people asked to avoid losses resumed interrupted tasks more often than people asked to pursue gains [26].

Here is the comparison that almost no popular article draws.

This is the error running through nearly every treatment of the topic online. Writers reach for the Zeigarnik effect to explain progress bars, cliffhangers, streaks, and abandoned shopping carts. But none of those things depend on you remembering better. They depend on you coming back. That is Ovsiankina, and Ovsiankina replicates.

What Your Mind Is Actually Doing With An Open Goal

If the memory advantage is unreliable but the pull to return is not, what exactly is an unfinished task doing inside a head?

Lewin described tension systems with no way of looking inside. The modern answer replaces tension with activation, and the cleanest evidence came from Thomas Goschke and Julius Kuhl in 1993 [30]. Their design is beautifully simple. Participants memorized two short scripts describing everyday activities, something like setting a dinner table and clearing a messy desk. Then they were told they would later have to actually perform one of them and merely remember the other. Before anyone did anything, the researchers measured how quickly people could recognize words from each script.

Words belonging to the script they intended to perform were recognized faster. The intention had given that material extra activation. And once the action was carried out, the speed advantage vanished.

That is Lewin's charged and discharged battery, observed as a difference in how accessible information is. An intention you still plan to act on keeps its content primed. Complete it and the priming fades. The finding became known as the intention superiority effect, and it has been extended and argued over ever since. It sits close to the broader question of how attention shapes memory, since what stays activated is largely what keeps getting noticed.

The idea connects to a broader theme in motivation research. Eric Klinger described the unfinished goals a person is committed to as current concerns, background filters that bias what gets noticed and what the mind drifts toward when it is not fully occupied [32]. Mind wandering research has since documented how much of ordinary thought is occupied by these unresolved commitments [57].

The most practically interesting result came from E. J. Masicampo and Roy Baumeister in 2011 [31]. Across a series of studies they showed that an activated unfulfilled goal produced intrusive thoughts during an unrelated reading task, made goal related words more accessible, and hurt anagram performance. Then they added a step. Some participants wrote a specific plan for the unfinished goal. Not completing it. Just writing down when and where and how it would get done.

The interference disappeared.

The goal was still unfinished. But the mind seemed to treat a credible plan almost the way it treats progress, releasing the resources the open loop had been consuming. Worth noting that this work sits close to a literature that has faced serious replication scrutiny in the years since, so it is best read as one influential set of studies rather than settled fact.

A few neighboring ideas are worth separating, because they get blurred together constantly.

George Loewenstein's information gap theory describes the discomfort of a gap between what you know and what you want to know, and the drive to close it [33]. That is an open loop about information specifically. A withheld answer, a cliffhanger, a question left hanging. Later work found that states of curiosity improve learning for the material you were curious about and even for incidental material encountered at the same time, apparently through dopaminergic circuits [34].

The goal gradient hypothesis explains why the pull intensifies near the finish. Ran Kivetz and colleagues resurrected it with data from a real coffee loyalty program, showing that customers bought more frequently the closer they got to a free drink [35]. Zeigarnik's own observation that interruption near completion produced stronger recall points in the same direction.

And the need for closure, formalized by Arie Kruglanski and Donna Webster, describes an individual difference in how uncomfortable ambiguity feels and how quickly a person seizes on an answer and freezes on it [54]. MacLeod argues that this construct is where Zeigarnik's core idea genuinely lives on.

Different mechanisms. Different evidence bases. Lumping them under one name is how a fragile finding acquired a reputation it could not support.

Unfinishedness Reaches All The Way Down To Vision

In January 2026, a result appeared that nobody in this literature seems to have predicted.

Joan Ongchoco, now at the University of British Columbia, along with Kimberly Wong and Brian Scholl at Yale, asked a question that sounds almost too simple. Does the visual system itself notice when something is unfinished, even when finishing has nothing to do with the task at hand [2]?

The design involved four experiments and 120 observers total, thirty per experiment. Participants watched a path unfold through a two dimensional maze, from a starting point toward an endpoint. On some trials the path reached the end. On others it stopped just short. Four small probes flashed briefly along the path, and afterwards, with the maze and probes gone, observers had to reproduce where those probes had been. Error was measured in pixels. Smaller error meant more precise memory for spatial position.

The finished or unfinished manipulation was completely irrelevant to what participants were asked to do.

It changed performance anyway. In the second experiment, which equated path length, distance travelled and duration, reproduction error on unfinished trials averaged 63.54 pixels against 73.61 pixels on complete trials, a difference of medium size at d equal to 0.62. The third experiment stripped away the drawn line entirely, using a moving dot that left no trace, ruling out any explanation based on visual connectedness. The pattern held at 59.91 against 65.89 pixels, d equal to 0.38.

The fourth experiment is the one that changes the interpretation. It repeated the traceless version and added a neutral baseline in which no endpoint was shown at all. Unfinished trials beat complete trials again, 57.29 against 66.62 pixels, d equal to 0.52. Unfinished also beat the baseline, 57.29 against 67.32 pixels, d equal to 0.50. But complete trials and baseline trials did not differ in the slightest, with a probability value of 0.880 and a d of 0.03.

Think about what that means. This is not a penalty for finishing. It is a bonus for not finishing. Incompleteness itself grabs something.

The authors argue that unfinishedness, usually treated as a property of goals and obligations and higher thought, is extracted by perception at a surprisingly deep level. Their framing is that vision pulls out an unexpectedly rich property from a display, and that the incompleteness of an event holds a privileged position in the mind.

One important caution. This is a single paper. It has not been independently replicated. The sample sizes are modest by modern standards, thirty per experiment, chosen from a pilot effect size. It is a genuinely new finding rather than a settled one, and it deserves to be described that way rather than folded immediately into the popular narrative it happens to rhyme with.

Four Numbers That Should Not Be Repeated

Anyone reading widely on this topic will meet several figures that do not survive contact with the primary sources. They are worth naming, because they propagate through hundreds of pages that copy from each other.

The first is ninety percent. Several popular treatments state that Zeigarnik found interrupted tasks recalled ninety percent better. No primary source supports that figure. Her reported ratios cluster around 1.9 to 2.1, which is roughly twice as well, not ninety percent better in the sense usually implied, and even that number belongs to her data alone and disappears when pooled with everything since [1].

The second is 138 children. This appears on several widely copied pages describing her experiments. The primary record documents an adolescent group of 45 with a mean age of 14, not 138 [3]. Where the larger figure came from is unclear.

The third is a neurological claim. Several health and productivity pages state that the anterior cingulate cortex, a region sitting deep in the front middle of the brain, keeps unfinished tasks active in working memory. There is a genuine and substantial literature on this region, covering conflict monitoring [55] and the calculation of how much mental effort a task is worth [56]. What does not exist is a peer reviewed neuroimaging study tying that region specifically to Zeigarnik style unfinished task memory. The claim is an inference drawn from adjacent research, and it is usually presented without a citation of any kind.

The fourth is twenty three minutes and fifteen seconds, the figure endlessly quoted as the time needed to recover focus after an interruption. The number is real, but it comes from a 2005 workplace study by Gloria Mark, Victor Gonzalez and Justin Harris, and it measures how long people took to return to an interrupted task, not how long focus took to recover [38]. It is routinely attributed to a different 2008 paper by Mark and colleagues, which actually found that interrupted work was completed faster but at the cost of higher stress, frustration and effort [37].

Four numbers. Four different failure modes. One shared cause, which is that a striking figure travels faster than a careful one.

The Study Advice That Does Not Follow From The Evidence

Now to the claim that matters most for anyone trying to learn something.

A version of this sentence appears in encyclopedia entries and study guides: because interrupted tasks are remembered better, students who break up their study sessions and do something unrelated in between will remember the material better than students who study straight through.

The claim traces back partly to a 1935 study of retention after interrupted learning activities [65]. And there is a genuine, well evidenced phenomenon in the neighborhood. It just is not this one.

Breaking study into separated sessions does improve retention. The evidence is overwhelming. A synthesis of 839 assessments across 317 experiments established the effect of distributed practice as one of the most reliable findings in learning research [48]. A later study mapped the optimal gap between sessions as a function of how long you need to remember something [49]. That is the spacing effect, and its mechanism has nothing to do with unfinished business. It works because forgetting between sessions makes the next retrieval harder, and harder retrieval strengthens memory more.

The same goes for testing yourself. Henry Roediger and Jeffrey Karpicke ran an experiment in 2006 that has become a landmark. Students who repeatedly tested themselves on a passage recalled 61 percent of it after a week. Students who repeatedly reread it recalled 40 percent. On an immediate test five minutes later, rereading actually looked better, 83 percent against 71 percent, which is exactly why so many students trust a method that fails them [45]. A later study found retrieval practice beat elaborative concept mapping as well [46]. A meta-analysis of the testing effect put the pooled effect at a g of 0.50 [47], and a separate meta-analysis across educational settings reached similar conclusions [66].

Set those numbers side by side. Retrieval practice sits around 0.50. Distributed practice is one of the best replicated findings in the field. The Zeigarnik effect sits at 0.15 and vanishes when the room is not casual.

Robert and Elizabeth Bjork gave the underlying principle a name: desirable difficulties, conditions that make learning feel harder in the moment and produce better retention later [50]. Spacing qualifies. So does retrieval practice. Mixing topics rather than blocking them qualifies under specific conditions, with a meta-analysis showing the benefit depends heavily on how similar the material is [51].

Interrupting yourself does not qualify. Nothing in the evidence supports it as a memory technique, and the interruption literature suggests the opposite is at least as likely [36].

What does this mean in practice? If someone tells you to deliberately stop mid page so the material sticks better, that specific instruction has no reliable support. If someone tells you to space your sessions and quiz yourself instead of rereading, that instruction is backed by hundreds of experiments.

There is one more piece of folk psychology worth examining here, because it is repeated on nearly every page about this topic. The claim goes that students forget exam material immediately afterwards because the brain treats the exam as a completed task and deletes what it no longer needs.

It is a satisfying story. There is no direct empirical support for it.

What the evidence does explain is why cramming fails, and the explanation is less exotic. Material learned in a single massed session shows steep forgetting because it was never distributed, never retrieved, and often encoded shallowly. Attributing it to a Zeigarnik style deletion mechanism adds a story to a phenomenon that already has a better one.

A related area does connect properly. Prospective memory, the ability to remember to do something in the future, is a genuine descendant of this research tradition. Lia Kvavilashvili argued in 1987 that remembering an intention is a distinct form of memory rather than a variant of ordinary recall [52]. Later work established that intentions are held through a mix of automatic and effortful processes depending on context [53]. And research on flight crews and other demanding environments showed that intentions decay surprisingly fast when attention is loaded [28].

That last finding sits awkwardly beside the popular claim. If unfinished intentions were reliably privileged in memory, they should not evaporate so quickly under load. The one study that tested the idea on actual exam material found the same conditionality, with recall of unanswered questions depending on the student's achievement motive rather than on incompleteness alone [21].

Hemingway's Trick And Why It Comes With A Condition

Ernest Hemingway described a working habit that has become a favorite piece of writing advice. Stop each day in the middle, while you still know what happens next, so the unfinished sentence pulls you back to the desk the following morning.

In 2018, three researchers actually tested it. Yoshinori Oyama, Emmanuel Manalo and Yoshihide Nakatani ran two studies with Japanese undergraduates and named the result the Hemingway effect [44].

In the first study, participants worked on a writing task and were stopped at different points. Those who had relatively little left to write, twenty four characters or fewer, reported significantly higher motivation to continue than those who had more remaining or who had finished. The pull was strongest close to the end, which lines up with both Zeigarnik's original observation about interruption timing and the goal gradient literature.

The second study, with 131 participants, found the condition that makes it work. The effect appeared only among participants given a structured writing task. When the task had no clear structure, so people could not estimate how much was left, stopping partway produced no motivational benefit at all.

That qualifier is the whole finding, and it is almost always dropped when the advice gets repeated. The pull does not come from stopping. It comes from stopping somewhere you can see the finish line from.

Notice also what was measured. Motivation to return. Not memory for the content. The Hemingway effect belongs on the resumption side of the ledger, alongside Ovsiankina, not on the recall side alongside the claim that made Zeigarnik famous.

The Cost Side: Interrupted Work, Residue And Sleep

Everything so far has treated the open loop as an interesting quirk. A separate body of research treats it as a cost, and that research has been considerably more consistent.

Sophie Leroy introduced the idea of attention residue in 2009 [36]. Across two experiments she showed that when people switch from one task to another, part of their attention stays behind on the first, and performance on the second suffers. The residue was worst when the first task had been left unfinished under time pressure. The mechanism is the same open loop, viewed from the other side. What keeps a task accessible also keeps it in the way.

The workplace interruption literature agrees. The 2008 study by Gloria Mark and colleagues found that interrupted tasks were completed faster, apparently through compensatory effort, but at measurably higher levels of stress, frustration, time pressure and mental workload [37]. Erik Altmann and Greg Trafton built a computational model explaining why resuming a task carries a cost, since the goal has to be reactivated from memory against interference [39]. Anyone interested in the mechanics of this in learning specifically will find it covered in more depth in work on fragmented attention.

Then there is sleep.

In 2017, Christine Syrek, Oliver Weigelt, Corinna Peifer and Conny Antoni ran a diary study that has become one of the most cited applications of this idea [40]. They matched 357 Friday and Monday observations across a twelve week period from 59 employees. Employees who ended the working week with more unfinished tasks reported worse sleep over the weekend. The relationship ran through affective rumination, meaning the emotionally charged repetitive thinking about work, rather than through problem solving reflection, which showed no such link. An earlier study by Syrek and Antoni had found a similar pattern and reported that it was stronger when leaders held high performance expectations [41].

The clinical literature made a related argument much earlier. Ian James and Katherine Kendell proposed in 1997 that intrusions and repetitive thoughts seen in emotional disorders might partly reflect unfinished processing, and they framed the Zeigarnik effect as one factor among several that could maintain such intrusions [42]. Their paper is a theoretical proposal rather than an experimental demonstration, and it should be read as such. Separately, Kenneth Savitsky and colleagues examined how unresolved actions and inactions relate to the availability of regret [43]. Earlier still, Maria Rickers-Ovsiankina had reported reduced spontaneous resumption among participants with schizophrenia, and argued this reflected difficulty maintaining goal directed tension systems [62].

These are research findings about groups, measured under specific conditions. They describe statistical relationships, not individual diagnoses or prescriptions.

What Is Still Unresolved

An honest account has to end with the arguments that are still live, and there are several.

The first concerns what the 2025 meta-analysis actually establishes. It shows that when situational influences and individual differences are averaged over, no memory advantage remains. It does not show that no such advantage can ever be produced. The relaxed condition ratio of 1.07 is a small positive result, and Atkinson's high achievement participants in a competitive setting reached 1.24 [8]. A conditional effect is not the same as no effect. It is a much weaker claim than the textbooks make, and a stronger claim than pure noise.

The second concerns the headline number itself. Removing Zeigarnik's own 1927 data changes the pooled ratio of means not at all, from 0.99 to 0.99, but it does move the outlier sensitive measure from 1.13 to 1.09. Which measure you consider primary shapes how dramatic the correction looks. The authors argue for Marrow's proportion measure, and by that measure the number sits just under half.

Third, the mechanism is contested. The classical account says persisting tension. The activation account says a goal representation held in an elevated state [30]. But the same two researchers who ran the meta-analysis published a separate study using anagrams that points somewhere else again [58]. Participants saw twelve anagrams with sixty seconds each, and when they failed, the solution was shown. Unsolved solutions were recalled better overall, but only among people high in what the achievement literature calls hope of success. Their reading favors a discrepancy account, where the gap between expected success and actual failure drives the memory, rather than an unreleased tension.

Alan Baddeley's 1963 result sits in the same awkward space [18]. He found a large apparent effect with anagrams, but because participants were shown the answer, the intention was terminated even though the task had been interrupted. He labelled his own result Zeigarnik like rather than Zeigarnik, which was scrupulous of him and largely ignored afterwards.

Fourth, the 2026 visual finding is one paper. It is elegant, well controlled and genuinely surprising. It is also unreplicated, with thirty observers per experiment. Whether unfinishedness turns out to be a basic perceptual property or a result specific to that maze paradigm is an open question that will take years to settle.

Fifth, there is the question of whether we can even run the original experiment anymore. If the effect really did depend on experimenter authority and deep task involvement, and if both have eroded, then the phenomenon may be historically situated in a way that psychology rarely acknowledges. That possibility is uncomfortable, because it suggests some findings are not timeless facts about minds but facts about minds in particular conditions.

None of this is a scandal. It is what a century of ordinary science looks like when someone finally adds up the numbers.

Conclusion

The Zeigarnik effect has had a strange career. Its most famous claim, that unfinished tasks lodge in memory roughly twice as firmly as finished ones, turns out to be among the weaker findings in the history of memory research. Ninety eight years of replication attempts produced a pooled ratio of 0.99 and an effect size of 0.15. The one condition where it reliably appears is a laboratory atmosphere deliberately made casual.

And yet almost nobody who reads about it feels that it is false. That intuition deserves an explanation rather than a dismissal.

The explanation is that the intuition is tracking something real that got attached to the wrong measurement. An open goal does keep pulling. People resume interrupted tasks about two thirds of the time, well above chance, across children and adults and across ninety years of methods. Intentions you still plan to act on stay primed until you act. Writing a specific plan releases the interference without finishing anything. The pull toward closure is not in doubt. It was only ever the memory claim that was.

That distinction is not a technicality. It changes what follows from the science. If the effect were about memory, deliberately interrupting your studying would be a technique. It is not, and the evidence for what actually improves retention points firmly elsewhere, toward spacing sessions apart and toward retrieving rather than rereading. If the effect is about goal pursuit, then the useful lever is different: not stopping in the middle, but knowing how far you are from the end.

There is something fitting in how this story resolves. A finding about unfinished business stayed open in the literature for nearly a century, cited freely, rarely questioned, never quite closed. Someone finally counted. The answer was not the one everyone expected, which is usually a sign the counting was worth doing.