Rewriting the night: nightmare disorder, its treatment, and a design for help at 3 a.m.

01Introduction

Most people know what a nightmare is. Fewer know how common recurrent nightmares are, how much they can shape a life, or that they are treatable. Nightmare disorder affects an estimated 4% of adults,^[1] and weekly nightmares are reported by roughly 4% to 10% of the population in the surveys reviewed by Levin and Nielsen.^[2] For those affected, the problem is not a single bad night. It is a pattern: waking in fear, dreading sleep, and carrying the night into the day.

Sleep medicine has a well-supported answer. Imagery rehearsal therapy (IRT), in which a person changes the storyline of a recurring nightmare while awake and then rehearses the new version, was recommended by the American Academy of Sleep Medicine (AASM) in 2010 and again in its 2018 position paper, where it was the only treatment recommended for both nightmare disorder and nightmares associated with posttraumatic stress disorder (PTSD).^{[3, 1]} Its landmark randomised trial was published in JAMA in 2001.^[4]

The treatment rarely reaches the people who need it. In two United States samples, only 37.8% and 11.1% of people with clinically significant nightmare symptoms had discussed their nightmares with a healthcare provider, and fewer than one in three believed that nightmares were treatable.^[5] A consensus paper on nightmare disorder likewise highlighted limited knowledge among healthcare providers and the poor delivery of evidence-based interventions.^[6] Clinical protocols assume several structured sessions with a trained therapist; one widely used description of IRT runs over four sessions of roughly two hours.^[7]

Our starting point is that this is partly a problem of packaging. The method was designed for the clinic and the daytime, but the need is most acute at night: a tired person, awake in the dark at 3 a.m., with a racing heart and no wish to read. Self-help and internet versions of imagery rehearsal show that the method can travel beyond the clinic,^{[8, 9]} but digital interventions face their own obstacle, which is that most people stop using them within weeks.^[10] The design question is therefore not whether a better method exists, but whether an existing method can be made usable in real life without overstating what it can do.

We focus on adults with recurrent nightmares, whether idiopathic or linked to trauma. Children’s nightmares, night terrors and REM sleep behaviour disorder are discussed mainly to mark the boundaries of what a self-help tool should attempt. Where findings come from narrow groups, such as veterans or survivors of sexual assault, we say so, because those groups differ from the people who are likely to find an app on their own.

This paper has four aims. First, to give a narrative review of what is known about nightmares and nightmare disorder: definitions, epidemiology, mechanisms, consequences and treatments. Second, to derive design principles from that evidence. Third, to show how each part of Baku, a calm, science-backed nightmare-relief app, maps to the evidence, using the evidence labels set out on our science page, and to state plainly where a component is a design hypothesis rather than an evidence-based treatment. Fourth, to describe how Baku will be evaluated.

Two boundaries apply throughout. Baku has not been evaluated, and nothing here is a claim that it works. And Baku is a wellness app built on evidence-based methods, not a medical device: it does not diagnose, treat, cure or prevent any condition (see our medical disclaimer). An overview of the product itself is on how it works.

02Definitions and classification

Nightmares and bad dreams

A nightmare is an extended, extremely dysphoric and well-remembered dream that usually involves threats to survival, security or physical integrity. Nightmares generally arise from REM sleep in the second half of the night, and on waking the person becomes alert and oriented quickly.^{[11, 12]} Researchers often use waking as the dividing line between nightmares and bad dreams, which are distressing dreams that do not wake the sleeper. In dream-log studies, nightmares were rated as markedly more intense than bad dreams, and about a third of nightmares had a primary emotion other than fear.^{[13, 14]} Physical aggression was the most frequent theme in nightmares, while interpersonal conflict predominated in bad dreams, suggesting that nightmares are a rarer and more severe expression of the same phenomenon.^[14]For design, this means a morning check-in should ask about a range of emotions, not only fear.

Nightmare disorder

Both main classification systems describe nightmare disorder in similar terms. In plain words, the Diagnostic and Statistical Manual of Mental Disorders (DSM-5-TR) requires repeated, extended, extremely distressing and well-remembered dreams; rapid orientation on waking; clinically significant distress or impairment; and that the nightmares are not attributable to a substance and not adequately explained by another mental or medical condition. Severity is graded by frequency, from less than weekly to nightly.^[11] The International Classification of Sleep Disorders (ICSD-3-TR) places nightmare disorder among the REM-related parasomnias and spells out the forms impairment can take, from fear of going to sleep and low mood to fatigue, daytime sleepiness and effects on work, study or relationships.^[12]

The emphasis on distress matters. Belicki showed that nightmare frequency and the suffering nightmares cause are related but distinct, and developed the Nightmare Distress Questionnaire to measure the latter.^[15] Someone with one nightmare a week may be more troubled than someone with three. An evaluation that counts nightmares but ignores distress would miss part of the problem.

What a nightmare is not

Two other parasomnias are often confused with nightmares and need a different response. Night terrors (sleep terrors) are sudden partial arousals from deep non-REM sleep, typically in the first part of the night, with screaming, intense fear and autonomic arousal, but with confusion and little or no recall of a dream; they are commoner in children.^[12] Because imagery rehearsal works on a remembered dream, it has nothing to work with in a night terror.

REM sleep behaviour disorder (RBD) is characterised by loss of the normal muscle atonia of REM sleep, so that people act out their dreams, sometimes causing injury. Diagnosis requires video-polysomnography, and RBD is now recognised as a frequent early sign of neurodegenerative conditions such as Parkinson’s disease.^[16] If someone describes shouting, punching or leaving the bed during dreams, the right next step is a medical assessment, not a rewrite. Baku’s safety routing reflects this (section 11).

Idiopathic and trauma-related nightmares

Nightmares are described as idiopathic when no clear traumatic origin exists, and as post-traumatic when they replay or echo a trauma. Distressing dreams are listed among the intrusion symptoms of PTSD,^[11] and sleep disturbance has been described as a hallmark of the condition.^[17] The distinction is not only academic. Clinicians have argued that chronic nightmares can behave as a disorder in their own right, which justifies treating them directly rather than waiting for an underlying condition to resolve.^[7] At the same time, nightmares tied to trauma call for more caution: the work is best done with a professional, and a self-help tool should not press anyone to revisit a traumatic memory in detail.

03Epidemiology

Prevalence estimates depend on how nightmares are defined and counted: weekly versus “often”, nightmares versus nightmare disorder, retrospective questionnaires versus prospective logs. With that caveat, the picture is consistent. Weekly nightmares are reported by about 4% to 10% of adults,^[2] and the AASM estimates that nightmare disorder affects about 4%.^[1] A clinical review put the share of the general population with nightmare problems at roughly 5% to 8%.^[7]

Population surveys

In a two-phase community study of 8,558 adults in Hong Kong, 5.1% reported nightmares at least once a week. Frequent nightmares were associated with female sex, lower income, insomnia, symptoms of sleep-disordered breathing and daytime consequences, and people with frequent nightmares were 5.74 times more likely to have a psychiatric disorder.^[18] Eight Finnish population surveys between 1972 and 2007, covering 69,813 adults, found frequent nightmares in 3.5% of men and 4.8% of women; prevalence rose with age, particularly among men, occasional nightmares became more common over the 35 years, and war veterans reported more nightmares decades after their service.^[19] In a representative German sample the commonest nightmare themes were falling, being chased, being paralysed, being late, and the death of someone close.^[20]

Women and men, younger and older

A meta-analysis of 111 studies found that women report nightmares more often than men from adolescence onwards, with the gap narrowing with age and not detectable in children or older adults. The size of the difference depended on how questions were asked: binary items showed smaller differences than rating scales.^[21] Whether the difference reflects dream recall, rates of depression, insomnia and trauma, or reporting habits remains open.

A recent multinational estimate

The International COVID-19 Sleep Study collected data with a unified online survey across 16 countries between May and November 2021. In 15,854 adults, frequent (at least weekly) nightmares were reported by 11.0% for 2021 and by 6.9% when participants recalled 2019; frequent dream recall rose from 51.1% to 54.0%. Nightmare frequency showed a quadratic relation to sleep duration.^[22]The study is valuable because it used the same method in countries never surveyed before, but its limits are worth stating: the sample was recruited online and was 69.9% women, the 2019 figures depend on retrospective recall, and the survey took place during a pandemic. It suggests that population levels of nightmares can shift with collective stress, and that a single prevalence figure should not be treated as fixed.

Groups at higher risk

Nightmares are especially common after trauma and in PTSD, where they are among the most persistent symptoms.^[17] The Finnish data show the long shadow of war exposure.^[19] Occupational stress matters too: in a study of 460 nurses using the Nightmare Disorder Index, 7.4% met full criteria for probable nightmare disorder and a further 43.9% had subthreshold symptoms.^[23] Psychiatric disorders, especially mood disorders, are strongly associated with frequent nightmares in community samples.^[18] These groups are also those for whom self-help alone may be least appropriate, which is a recurring theme in the design sections.

What the numbers mean for design

Three implications follow. First, nightmares are common enough that a product built around them will reach people with very different needs, from occasional bad dreams to severe post-traumatic nightmares, so it must recognise when self-help is not enough. Second, because estimates vary with definitions and methods, change is best measured against each person’s own baseline rather than against population norms. Third, the strong links with insomnia and psychiatric disorder mean that fewer nightmares will not always mean better sleep, which is why sleep-focused methods sit alongside rewriting (section 6.3).

04Mechanisms

No single account explains all nightmares. What follows is a summary of the most influential models, ordered from well-established physiology to theories that remain contested. They matter for design because each suggests a different lever: the dream script, the level of arousal, or the way a person relates to their nightmares.

Sleep architecture and the timing of nightmares

Because nightmares arise mainly from REM sleep, and REM sleep is concentrated in the later part of the night, nightmares tend to occur in the second half of the main sleep period.^[11] This is one reason why the hardest moment for many people falls in the small hours. Waking at that time also brings sleep inertia, a transitional state of lowered arousal and reduced cognitive performance. Sleep inertia appears more intense when waking near the low point of core body temperature and, in the absence of major sleep deprivation, rarely lasts more than about 30 minutes.^[24] A person who has just woken from a nightmare is therefore frightened and cognitively blunted at the same time, which is a poor moment for reading or for complex choices.

The neurocognitive model: affect load and affect distress

Levin and Nielsen proposed that dreaming normally supports emotional regulation, in particular fear extinction: elements of fear memories are recombined with new contexts, helping to form safety memories that compete with the original fear. Nightmares reflect a breakdown of this function.^[25] Two factors determine how often and how severely it breaks down: affect load, the day-to-day build-up of emotional pressure, and affect distress, a trait-like disposition to react to events with distressing, highly reactive emotions. The model links these to a network of regions involved in fear and its regulation, including the amygdala, medial prefrontal cortex, hippocampus and anterior cingulate cortex.^[2] A later consensus paper summarised the field in similar terms: chronic nightmares develop through the interaction of heightened arousal and impaired fear extinction, facilitated by trait affect distress arising from trauma, early adversity and individual susceptibility, and possibly by thought suppression and sleep-disordered breathing.^[6]

REM sleep and emotional memory

Walker and van der Helm proposed that REM sleep acts as a kind of overnight therapy: in a neurochemical environment low in noradrenaline, emotional memories are reprocessed so that their content is kept while their emotional charge is reduced. If this process fails, the emotional tone of memories may persist, contributing to anxiety disorders and PTSD.^[26] In support, a functional imaging study found that REM sleep physiology was associated with an overnight decrease in amygdala reactivity to earlier emotional experiences and with lower next-day emotional intensity.^[27] On this view, recurrent nightmares are a sign that the overnight processing of emotion is not doing its work.

Hyperarousal and PTSD

The idea that REM sleep disturbance is the hallmark of PTSD has prompted decades of research. Germain’s review concluded that polysomnographic findings are mixed and often modest, but that prospective and treatment studies give robust evidence that sleep disturbance, including nightmares, contributes to poor psychiatric outcomes and may be a modifiable risk factor. Chronic sleep disruption from nightmares may also reduce the effect of first-line PTSD treatments, while targeted sleep treatment may speed recovery.^[17] Heightened noradrenergic activity during sleep is one candidate mechanism, and it is the rationale for trials of the alpha-1 blocker prazosin (section 7).

Theories of what dreams are for

Two broader theories are often invoked. Threat simulation theory proposes that dreaming evolved to simulate threatening events, allowing threat perception and avoidance to be rehearsed safely.^[28] The continuity hypothesis holds that dream content reflects waking concerns and experiences, although not all waking activities are represented equally; focused activities such as reading and writing appear far less often than their share of waking life would predict.^[29] Neither theory is settled, and neither is needed to justify imagery rehearsal. They do shape how Baku talks about dreams: as reflections of waking life worth noticing, never as messages that predict the future.

How treatments might work

A systematic review of hypotheses about why psychological nightmare treatments work found that an increased sense of mastery was the explanation most often proposed, followed by emotional processing that modifies the fear structure, changes in beliefs, restored sleep function, reduced arousal and less avoidance.^[30] These map onto the three levers above: changing the script, lowering arousal, and changing a person’s relationship to their nightmares from something that happens to them into something they can shape.

05Consequences

Sleep and daytime functioning

Nightmares fragment sleep directly by waking people, and indirectly by making them reluctant to go back to sleep or to go to bed at all. Frequent nightmares travel with insomnia symptoms and daytime consequences in community samples.^[18] Fear of sleep, fatigue and daytime sleepiness are named in the ICSD-3-TR as forms of impairment.^[12] A widely used measure of sleep quality, the Pittsburgh Sleep Quality Index, includes bad dreams among the reasons for disturbed sleep.^[31]

Distress and mental health

Nightmare distress, more than frequency, appears to track waking emotional adjustment.^[15] Frequent nightmares are strongly associated with psychiatric disorders, mood disorders in particular, and with neuroticism even after psychiatric diagnoses are taken into account.^[18] The direction of these associations is likely to run both ways: distress and psychiatric illness may increase nightmares, and nightmares may worsen sleep and mood. In PTSD, nightmares and insomnia are not merely downstream symptoms; they may help to maintain the disorder.^[17]

Suicidal thoughts and behaviour

This section discusses suicide. If you are struggling to stay safe right now, please contact local emergency services or a crisis line; you can find options on our help page. Help is available, and people do get through this.

A meta-analysis of 39 studies with 147,753 participants found that sleep disturbances in general, and nightmares and insomnia individually, were associated with suicidal ideation, suicide attempts and suicide, and that depression did not explain the association.^[32] In a Finnish population cohort of 36,211 adults followed for up to 23 years, frequent nightmares at baseline were associated with roughly double the adjusted risk of later suicide compared with no nightmares.^[33] These are associations: most people with nightmares are not at risk, and nightmares do not cause suicide on their own. But nightmares are a modifiable and rarely asked-about signal, which is one reason why screening for them has been recommended.^[5] We follow the World Health Organization’s guidance on talking about suicide: no details of methods, an emphasis on help and recovery, and clear routes to support.^[34] More is in our article on nightmares and suicide risk.

Physical health associations

Recent cohort studies have linked distressing dreams to later neurological outcomes. Using a single item from the Pittsburgh Sleep Quality Index, Otaiku found that among 605 cognitively normal middle-aged adults, those reporting weekly distressing dreams had a higher risk of subsequent cognitive decline (adjusted odds ratio 3.99, 95% CI 1.07 to 14.85), and among 2,600 older adults weekly distressing dreams were associated with a higher risk of dementia (adjusted odds ratio 2.21); the associations were statistically significant only in men.^[35] These findings are observational. Distressing dreams may be an early marker of a disease process rather than its cause, and the confidence intervals are wide.

In June 2025 a presentation at the European Academy of Neurology Congress reported that frequent nightmares were associated with faster biological ageing and earlier death. The congress press release described adults with weekly nightmares as more likely to die before the age of 70 than those who rarely or never had nightmares, with accelerated ageing accounting for part of the association.^[36] Other reports of the same work describe a different sample and a different age threshold, and as of October 2026 we have found no peer-reviewed publication. We therefore treat it as an unpublished, observational finding that needs confirmation, and we do not use it to motivate anyone through fear. Our article on the early-death headlines explains why.

06Treatment evidence: psychological treatments

6.1 Imagery rehearsal therapy

In imagery rehearsal, the person chooses a recurring nightmare, changes it in any way they wish, and rehearses the changed version in imagination each day while awake. Clinical descriptions embed this in education about nightmares and sleep and in practice with imagery, and, as usually described, the method does not ask the person to relive the whole nightmare in detail.^[7] A plain-language description is in our article on what imagery rehearsal is.

The landmark trial randomised 168 women with chronic nightmares, 97% of whom had experienced rape or other sexual assault and 95% of whom had moderate to severe PTSD, to three sessions of IRT or a waiting list. Treatment reduced nights with nightmares and the number of nightmares, improved sleep quality and reduced PTSD symptoms, with gains sustained at six months without further contact; an intention-to-treat analysis of all 168 participants confirmed the differences.^[4]

Meta-analyses support the overall picture. Across 13 studies of post-trauma nightmares, imagery rehearsal had large pre-to-post effects on nightmare frequency, sleep quality and PTSD symptoms, sustained at 6 to 12 months; adding CBT for insomnia improved sleep quality further, and effects were small in the one study with an active control.^[37] A meta-analysis of treatments using imagery rescripting and rehearsal or confrontation with nightmare content found large pre-post effects on nightmare frequency (Hedges’ g = 1.04), and longer exposure to nightmare content was associated with larger effects, leaving open which ingredient matters most.^[38] Across 19 randomised trials with 1,285 participants, psychological treatments had a moderate effect (d = 0.48), larger in individual than in self-help formats.^[39]

Not every trial is positive. In 124 male Vietnam veterans with chronic, severe PTSD, six sessions of group imagery rehearsal were not superior to a credible active comparison.^[40] In 108 Iraq and Afghanistan veterans, adding imagery rehearsal to CBT for insomnia conferred no overall advantage over CBT for insomnia alone.^[41] In 104 people with nightmare disorder, mostly idiopathic, three sessions of either imagery rescripting alone or imaginal exposure alone reduced nightmare frequency and distress compared with a waiting list, with effects sustained at six months.^[42] Taken together, the method is well supported against no treatment, its advantage over active alternatives is smaller and less consistent, and several routes, including rescripting and exposure, appear to help.

6.2 Exposure, relaxation and rescripting therapy

ERRT combines education, relaxation, writing about and exposure to the nightmare, rescripting, and attention to sleep habits. In a randomised trial with trauma-exposed adults, 84% of treated participants reported no nightmares in the week before the six-month assessment, although only 27 participants completed treatment.^[43] The AASM lists ERRT among treatments that may be used.^[1] A 2025 expert consensus statement on cognitive behavioural therapy for nightmares (CBT-N) set out the components of these multi-component treatments: relaxation, stimulus control, sleep efficiency training, sleep hygiene, nightmare exposure, rescripting and imagery rehearsal.^[44] Exposure to trauma content is clinical work. Baku uses the rescripting route, with only minimal re-telling of the dream.

6.3 CBT for insomnia and stimulus control

Insomnia often accompanies nightmares. Multi-component CBT-I is strongly recommended as the first-line treatment for chronic insomnia in adults by the AASM, and stimulus control is conditionally recommended as a single component.^[45] The European guideline likewise recommends CBT-I as first-line treatment.^[46] For nightmares, the AASM lists CBT-I as an option for PTSD-associated nightmares,^[1] combined treatment improved sleep quality more than imagery rehearsal alone,^[37]and in veterans adding imagery rehearsal to CBT-I gave no overall advantage over CBT-I alone.^[41] Stimulus control asks people to use the bed only for sleep and to get up when they cannot sleep, returning when sleepy. It is the basis of the “twenty-minute rule” in Baku.

6.4 Lucid dreaming therapy

Lucid dreaming therapy teaches people to recognise that they are dreaming and to change the dream from within. The AASM lists it as an option that may be used for nightmare disorder.^[1] In a pilot trial with 23 participants, a single session reduced nightmare frequency, but lucidity itself was not necessary for the improvement.^[47] A review concluded that results are promising but that studies are few, small and methodologically limited.^[48] We label it Emerging and do not make it a default path.

6.5 Targeted memory reactivation

In a randomised study, 36 people with nightmare disorder performed imagery rehearsal; for half, a sound was paired with the positive new version during the initial session. Over two weeks all participants rehearsed at home and heard the sound during REM sleep, delivered by a headband that detected sleep stages. The group with the paired sound had fewer nightmares and more positive dream emotions after two weeks, and a sustained reduction at three months.^[49] This is an elegant result from one small trial that relied on EEG-based detection of REM sleep, which a phone cannot do reliably. We discuss it in a sound that helps a new ending stick; any night cue in Baku would be opt-in, labelled experimental and measured before being promoted.

07Treatment evidence: delivery, medication and the guideline picture

7.1 Self-help and digital delivery

Self-delivered versions of nightmare treatment have a long history. In an early study of 20 people with chronic nightmares, both a single session of imagery rehearsal and simply recording nightmares for a month reduced nightmare frequency, with no significant difference between the groups.^[50] In a trial of 399 participants recruited through a Dutch nightmare website, six-week self-help imagery rehearsal and self-help exposure both improved nightmare frequency, distress and sleep quality compared with a waiting list; recording nightmares also outperformed the waiting list, and imagery rehearsal produced the fastest reduction in diary-recorded nightmares.^[8] In a trial of internet-based imagery rehearsal with 127 people with mainly idiopathic nightmares, the intervention outperformed a frequency-monitoring control on nightmare frequency and distress, guidance by a coach made no difference to outcomes, and writing the nightmare narrative in detail also reduced frequency.^[9]

Two cautions follow. Self-help formats have produced smaller effects than individual therapy,^[39] and simple monitoring itself reduces nightmares, which any evaluation of an app must control for. Engagement is the larger problem: an analysis of 93 popular mental health apps found median retention of 3.9% at 15 days and 3.3% at 30 days,^[10] and a meta-analysis of app trials for depressive symptoms found pooled dropout of 26.2%, rising to 47.8% after adjustment for publication bias, with lower dropout where human feedback or in-app mood monitoring was offered.^[51] A digital version of imagery rehearsal is plausible; the bottleneck is whether people keep using it long enough for rehearsal to work.

7.2 Pharmacotherapy

The story of prazosin shows how evidence can change. The 2010 AASM guide recommended it for PTSD-associated nightmares at its highest level.^[3] In 2018, a trial of 304 veterans across 13 Veterans Affairs centres found no difference from placebo in distressing dreams, sleep quality or global clinical change.^[52] The AASM then moved prazosin to “may be used”, and listed clonazepam and venlafaxine as not recommended for nightmare disorder.^[1] A later meta-analysis of eight imagery rehearsal and seven prazosin trials found moderate effects for both on nightmare frequency (g = 0.51 and g = 0.61) with no significant difference between them, and argued that the downgrade was premature.^[53] Baku provides information only. Decisions about medication belong with a doctor; we summarise the debate in prazosin and nightmares.

Devices have also entered the field. In 2020 the United States Food and Drug Administration granted De Novo authorisation to NightWare, prescription software for an Apple Watch that uses heart rate and movement during sleep to deliver vibrations intended to reduce sleep disturbance from nightmares in adults aged 22 or older; it is not intended as a standalone PTSD therapy.^[54]This is medical-device territory and not part of Baku.

7.3 The AASM position paper as a whole

The 2018 AASM position paper, based on literature from 2009 to 2017, used three categories: recommended, may be used, and not recommended. Only imagery rehearsal was recommended for both nightmare disorder and PTSD-associated nightmares. A long list of psychological and pharmacological options may be used, reflecting thin or mixed evidence rather than demonstrated equivalence, and the paper stresses that clinical judgement should decide care.^[1] Table 1 summarises the evidence in Baku’s labels.

Table 1. Approaches to nightmares and their evidence, in the labels Baku uses. Strong: recommended in guidelines and supported by several randomised trials. Good: supported by trials or guidelines, with limits. Emerging: promising early studies, often small. Exploratory: laboratory findings, surveys or design reasoning. Medical device territory: regulated products outside Baku.

08From evidence to design: principles

The review points to a method that works, a population that rarely receives it, and a moment of need, the night-time awakening, that no clinical protocol addresses directly. We derived eleven principles from it. Each states what we do, why, and how strong the evidence behind the choice is.

1. Value before effort. Because most app users leave within weeks,^{[10, 51]} the first minutes must help. Everything before sign-in works offline, and the night-time mode never asks anyone to sign in or pay. This principle rests on engagement data, not on nightmare trials.
2. Night rules. After lights out the interface is ember on black, offers at most three choices, uses no more than two lines of text, does not scroll and fades to black after ten seconds without input. The human circadian system is most sensitive to short-wavelength light,^[56] and a person waking from a nightmare is affected by sleep inertia,^[24] so the interface should ask for almost nothing. We acknowledge a tension: stimulus control discourages lying awake in bed, and any phone use at night carries some cost.^[45] Nightlight is designed to be brief and to end in sleep or in getting up.
3. Milder dream first. Clinical descriptions of imagery rehearsal favour learning the technique on a manageable nightmare and avoid dwelling on trauma content.^[7] The Rewrite studio suggests a milder dream first and asks only for the turning point, not a full retelling.
4. The person writes the ending. Imagery rehearsal lets the dreamer change the nightmare however they wish, and mastery is the most often proposed reason that nightmare treatments work.^{[7, 30]} Baku offers ways in (a new ending, a helper, a new place, a power, something absurd), and its AI Guide may ask for sensory detail, but the words are the person’s own.
5. Scheduled daytime rehearsal. Rehearsal is the step that turns a rewrite into a treatment. If-then plans that link an intended action to a specific time and place reliably improve follow-through; a meta-analysis of 94 tests found a medium-to-large effect.^[57] Each new ending gets seven days of rehearsal at a time the person chooses, during the day, followed by a check-in. Seven days is a design choice, not a dose established by trials.
6. Arousal reduction at 3 a.m. In a randomised study, five minutes a day of slow breathing with long exhalations, practised in the daytime, improved mood and reduced respiratory rate more than mindfulness meditation.^[55] Nightlight pairs breathing (in for four seconds, out for six) with a haptic heartbeat that slows from 60 to 50 beats a minute, an idea drawn from a study in which wrist vibrations suggesting a slow heart rate helped people regulate anxiety.^[58] The breathing has Good support; the haptic heartbeat is Exploratory.
7. Self-monitoring without sleep scores. Recording nightmares alone reduces them modestly,^{[50, 8]}and progress is best judged against a person’s own baseline. But tracking can backfire: some people become anxious in pursuit of perfect sleep data.^[59] Baku’s moon calendar shows nights, not scores, compares weeks with the person’s baseline, keeps effort (practice) separate from outcome (calm nights), and never lets a nightmare “break” a streak.
8. Partner support. Most adults sleep with a partner, and relationship quality and sleep appear to influence each other.^[60] Partner mode lets a loved one leave a recorded voice for the night and bedtime notes, without ever seeing dream content. This is a design hypothesis.
9. Evidence labels. Fewer than one in three people with significant nightmares believe they are treatable.^[5]Telling people clearly what is known, and how well, is part of the help. Following the logic of GRADE, which separates the quality of evidence from the strength of a recommendation,^[61] every method in Baku carries one of five labels and a source one tap away.
10. Safety routing. Nightmares are linked to suicidal thoughts,^[32] trauma-related nightmares are best treated with a professional, and dream enactment needs medical assessment.^[16] Baku routes these situations to human help (section 11), using safe-messaging practice.^[34]
11. Privacy as a precondition for honesty. People edit what they report on sensitive topics to avoid embarrassment or repercussions, and how much they disclose depends on the setting.^[62] Many mental health apps share data with third parties, often without disclosure.^[63] Accurate dream and nightmare logs depend on trust, so Baku has no advertising, sells no data, locks the journal behind device authentication and lets any entry be kept private from the AI.

09From evidence to design: how Baku maps to the evidence

Table 2 maps each Baku component to its rationale, evidence label and sources, and states whether it delivers an evidence-based method or tests a design hypothesis. The labels refer to the underlying method, not to Baku: no component of Baku has itself been evaluated. Where a component combines elements with different levels of support, we give each.

Table 2. Baku components, their rationale and evidence labels. Labels describe the underlying method, not Baku itself.

How labels are assigned, and what would change them

Labels are assigned by the team from the sources cited, using the definitions under Table 1, and will be reviewed by our clinical advisor before launch. They are judgements informed by the logic of GRADE, not formal GRADE ratings.^[61] A label can move in either direction: a replicated trial of sound cues during sleep could raise targeted memory reactivation above Emerging, while evidence that a component distracts from rehearsal or increases distress would lead us to change or remove it. Results of Baku’s own evaluation will be reported separately from these method labels, so that evidence for a method and evidence for the app are never confused.

Three points deserve emphasis. First, only the rewrite and rehearsal core and the evening relaxation elements rest on treatments with guideline support; the rest are reasoned design choices. Second, an evidence-based method delivered through a new format is not automatically evidence-based in that format: the gap between clinic-delivered and self-help effects shows why.^[39] Third, some things Baku deliberately does not do. It does not deliver exposure to trauma memories, interpret dreams as predictions, give medication advice, detect nightmares through sensors, or play sound cues during sleep at launch. The Atlas also presents dream traditions from many cultures, opt-in and clearly separated from treatment claims; see traditions and what imagery rehearsal is.

10Evaluation plan

Baku makes no claims about its own effectiveness, and will make none until they have been measured. This section describes how we intend to measure them. Details will be fixed in a protocol before any controlled study begins.

Outcomes

• Weekly nightmare frequency, recorded prospectively each morning (nights with nightmares and number of nightmares), as in the landmark trial.^[4]
• Nightmare distress, measured with the Nightmare Distress Questionnaire,^[15] because frequency alone misses suffering.
• Nightmare disorder symptoms, measured with the Nightmare Disorder Index, a brief self-report measure aligned with DSM-5 criteria.^[23]
• Sleep quality and insomnia, measured with the Pittsburgh Sleep Quality Index and the Insomnia Severity Index.^{[31, 65]}
• Safety, including crisis handoffs and any reports of increased distress after rewriting or rehearsal.
• Engagement, including rehearsal days per week and retention, because adherence determines whether rehearsal can work.

Questionnaires will be administered at baseline and at weeks 2, 4 and 8, subject to licensing terms for each instrument.

Measurement inside the app

The morning check-in asks the person to label the night as calm, bad dream, nightmare or no memory, which creates a prospective log without extra questionnaires. Baseline is the average weekly count over the first week of use, reported alongside the retrospective estimate given during onboarding. A missed morning is treated as missing, never as a calm night, and analyses will report how much data are missing and how they were handled. Because morning logging is itself a form of monitoring, its effect is part of what Baku delivers and has to be separated from the rest of the app in a controlled design.

The north-star metric and its limits

Our operational north-star metric is the share of active people who have fewer nightmares in week 4 than at their own baseline. It keeps the team focused on outcomes rather than screen time. It is not evidence of efficacy. People tend to start an app when nightmares are at their worst, so many would improve anyway (regression to the mean); nightmares fluctuate naturally; counting only active people excludes those who left, perhaps because it did not help; and self-reports can be shaped by expectation. Monitoring alone reduces nightmares,^{[50, 8]} so an uncontrolled before-and-after comparison cannot separate Baku’s contribution from these effects.

Stages

1. Private beta. A small group of adults will use Baku with informed consent. The aims are feasibility, safety, acceptability and the quality of measurement, supported by interviews. Beta results will be described as such, never as efficacy. If distress reports or crisis handoffs rise in connection with a feature, that feature will be paused and reviewed with the clinical advisor.
2. Pre-registration. Before a controlled study, we will publish a protocol and analysis plan, with primary outcome, sample size reasoning and handling of missing data, on a public registry. Pre-registration separates planned tests from exploratory findings.^[66]
3. Independent randomised evaluation. We intend to fund or support a randomised trial run and analysed by academic partners without a financial stake in the result. Because monitoring alone helps, the comparison should include an active control, such as nightmare diaries, as well as usual care or a waiting list. Analysis will be by intention to treat, and reporting will follow CONSORT-EHEALTH.^[67]

What we will publish, and when we will not

We will publish the protocol, all pre-specified outcomes whatever their direction, attrition and adverse events, and any changes to the plan with reasons. Null or negative results will be published with the same prominence as positive ones. We will not present beta or uncontrolled data as proof of efficacy, will not report subgroups chosen after seeing the data unless labelled exploratory, will not use testimonials as evidence, and will never publish data that could identify a person. Research use of anyone’s data will require separate, explicit consent.

11Safety, ethics and data protection

Wellness positioning

Under the European Medical Device Regulation, software is a medical device when its manufacturer intends it for a medical purpose such as the diagnosis, monitoring or treatment of disease.^[68] Baku is positioned and described as a wellness app built on evidence-based methods. It does not diagnose, treat, cure or prevent any condition, and its language is controlled accordingly: “research suggests”, “may help”, never “cures”. A medical-device route would require clinical trials and is not part of the current plan.

Crisis detection and human handoff

Guide conversations and journal entries sent to the Guide pass a risk check. Signs of crisis or self-harm lead to a calm screen with local, human help by country, written by people rather than generated. Baku is not an emergency service and says so. Wording follows safe-messaging guidance: no method details, an emphasis on help, and hope without minimising.^[34] Crisis resources are on our help page.

Trauma

Baku never asks anyone to describe a traumatic nightmare in detail. It suggests starting with a milder dream, asks only for the turning point, and, when trauma comes up, recommends doing this work with a professional as well. Baku can sit alongside therapy, and a person can export their entries to share with a therapist; it is never a substitute for trauma-focused care.

Age

The GDPR sets 16 as the default age from which a child can give consent alone to online services, and allows member states to lower it to no less than 13.^[69] Baku is available from 16 in Switzerland, the European Union and the United Kingdom, and from 13 elsewhere. Children’s nightmares are common and deserve their own approach, which we plan to address later with guidance for parents.

Boundaries for the AI Guide

The World Health Organization’s guidance on large multi-modal models in health highlights risks such as inaccurate or biased outputs, automation bias and cybersecurity weaknesses.^[64] The Guide is bounded accordingly. It gives no diagnoses, no medication advice beyond suggesting a conversation with a doctor, no dream prophecy and no fear-based symbol meanings. It never pushes trauma disclosure. It proposes changes as cards that do nothing until the person confirms them. Its memory is visible and can be edited or deleted. Entries marked “keep private from AI” are never sent, identifying details are removed where possible, and prompt or model changes are tested against a set of normal, distressing, trauma-related and crisis messages before release.

Health data

Dreams and sleep records are health data. Under the GDPR they belong to the special categories of personal data in Article 9, and Baku processes them on the basis of explicit consent.^[69] Under the Swiss Federal Act on Data Protection they are sensitive personal data.^[70] Baku stores entries locally first, encrypted at rest, behind device authentication; hosts its servers in Europe; completes a data protection impact assessment before launch; collects analytics without content; and offers export and deletion at any time. Baku has no advertising and sells no data, in a market where third-party data sharing by mental health apps is common.^[63] Details are in our privacy policy.

12Limitations

A narrative, not a systematic, review. We did not register a protocol, run an exhaustive search, screen studies in duplicate or assess risk of bias formally. We prioritised guidelines and meta-analyses and used key trials and observational studies to illustrate them. Relevant studies, especially those not published in English, may have been missed, and our selection may be biased towards findings that support our design.

The quality of the underlying evidence. Many nightmare trials are small, use waiting-list controls and rely on retrospective self-report. Effects shrink against active comparators.^{[37, 40]} Samples are dominated by particular groups, such as sexual assault survivors, veterans and students, so results may not transfer to people who download an app on their own initiative. Prevalence figures depend on definitions, and pandemic-era estimates and unpublished congress findings are provisional.

Digital adherence. Self-help effects are smaller than those of individual therapy,^[39] and most app users stop within weeks.^[10] Baku’s design addresses this, but whether it succeeds is an empirical question.

Who the evidence covers. Most trials enrolled adults, while Baku will be available from 16 in Switzerland, the European Union and the United Kingdom, and from 13 elsewhere, so the evidence for adolescents is thinner. Most research comes from North America and Europe, with notable exceptions such as the Hong Kong community study and the 16-country survey.^{[18, 22]}

Our own labels and measures. The evidence labels are the team’s judgements from the cited sources, not ratings by an independent panel. The in-app nightmare log is a self-report that may itself change what it measures, and some validated questionnaires have licensing terms that could limit how they are used inside an app.

Possible harms. Thinking about a nightmare in order to rewrite it may briefly increase distress for some people, and a phone at the bedside can itself disturb sleep. We have tried to minimise both, and our evaluation will look for them, but we cannot rule them out.

Untested components. The haptic heartbeat, partner mode, the AI Guide and the visual moon calendar are design hypotheses. Their combination with established methods could help, do nothing, or in some cases distract from rehearsal.

No outcome data and a conflict of interest. Baku has not been evaluated. This paper was written by the team that builds and sells Baku, which is a conflict of interest; it has not been peer reviewed, and review by a clinical advisor is pending before launch. These are the reasons we commit to pre-registration and independent evaluation.

A moving field. Several recent findings, including pandemic-era prevalence estimates and the congress report on mortality, have not yet been replicated or fully published. This paper will date; we will revise it in numbered versions and say what changed.

13Conclusion

Nightmares are common, distressing and linked to poorer sleep and mental health. Weekly nightmares are reported by roughly 4% to 10% of adults, nightmare disorder affects about 4%, and recent multinational data suggest that population levels can rise in times of collective stress.^{[2, 1, 22]} Their associations with insomnia, psychiatric disorder and suicidal thoughts are well documented, while the reported links with dementia and mortality are observational and, in part, unpublished.

Nightmares are also treatable. Imagery rehearsal, a brief and drug-free method, is the only treatment the AASM recommends for both nightmare disorder and PTSD-associated nightmares, and it has helped in clinics, in self-help formats and online.^{[1, 4, 8]}Its benefits are clearest against no treatment and smaller against active alternatives. Exposure-based and multi-component cognitive behavioural treatments, insomnia treatment and relaxation add further options; lucid dreaming therapy and sound cues during sleep are promising but rest on small studies; and the evidence for prazosin is mixed.

The gap is one of access. Most people never mention their nightmares to a clinician, many do not know that help exists, and few clinicians are trained to provide it.^{[5, 6]} The moment of greatest need, a night-time awakening, falls outside every clinical protocol.

Baku is our attempt to close part of that gap by packaging an established method for real life: rewriting by day, rehearsing on a schedule, keeping a calm record of nights, and being one quiet tap away at 3 a.m. The design principles we derived are modest and testable. Ask little of a tired mind. Let the person write their own ending. Make rehearsal easy to keep. Lower arousal rather than add information. Measure progress against the person’s own baseline without turning sleep into a score. Route risk to humans. Earn honesty through privacy.

We have tried to show our reasoning in full: which parts rest on strong evidence, which on good evidence with limits, and which are educated guesses. Baku has not been evaluated, and this paper is not evidence that it works. It is a statement of what we believe, why we believe it, and how we will find out whether we are right: a private beta, a pre-registered plan, an independent randomised evaluation, and publication of the results whatever they show. If the evidence says that parts of Baku do not help, we will change them, and we will say so.