In this guide
In a 2026 trial of 495 working adults, bedtime stories from a mental health app did not do significantly better than a digital control at reducing self-reported sleep disturbance (Hedges' g 0.12, 95% confidence interval -0.13 to 0.37). Do not expect a story to fix poor sleep on its own.
That is our reading of the paper, and the authors say much the same. Here is what they tested, what the numbers mean, and what the trial leaves open.
What the trial tested
The paper is "USleep: efficacy of app-based audio interventions to improve sleep disturbance in working adults," published in SLEEP in 2026. It enrolled 495 working adults with self-reported sleep disturbance. They were recruited online, and their mean age was 32.7.
Participants were split into four arms for four weeks. Three arms used one category of a single app: Bedtime Stories, Sleep Sounds or Sleep Skills. The fourth was a digital control, a different set of daily content in the same app that was not designed for sleep.
So the control group was not "nothing." It got something to open each day, which matters for the numbers below.
The numbers, in plain words
Hedges' g is a way to compare two groups in one unit. It measures the gap between their average results in units of how spread out the scores were. Zero means no gap.
A 95% confidence interval is the range of true gaps that fit the data reasonably well. When the interval includes zero, the data are consistent with no gap at all, and the result is not statistically significant.
In the paper's Table 2, g for Bedtime Stories against the control was 0.12, with an interval of -0.13 to 0.37. That interval includes zero. It also reaches up to 0.37, so the trial could not rule out a modest benefit either. It could not show one.
The other arms looked alike. Sleep Sounds came to 0.14. Sleep Skills came to 0.07, with an interval of -0.18 to 0.32 in the paper's Table 2. Every arm improved over the four weeks, the control arm too. The authors say that improvement cannot be assumed to come from the sleep-specific content.
Their conclusion, in the paper's abstract: the audio tools did not show superiority over the control, and use as standalone treatments for sleep disturbance is not supported.
The significant result that favoured stories, and the per-protocol result that did not
Our view: report the result that cuts against the headline, because it is the fairer picture.
One secondary result significantly favoured stories. Self-reported sleep duration rose 53 minutes in the stories arm against 33 minutes for the control (g -0.26, 95% CI -0.51 to -0.01, p=.042). The interval stops just short of zero. It came from one self-report question, and the authors say it warrants further investigation.
The paper also reports a per-protocol analysis, which looks only at people who used their tool as intended. There, the stories showed a medium-sized effect favouring the control group. The main text does not state whether that result is significant, so we do not call it either way.
On safety, twelve participants across all four arms, including two in the control group, reported lasting bad effects. That was 3.58% of app downloaders, with no statistically significant difference between the sleep tools and the control (p=.618). One story-arm participant described relying on the app to get to sleep.
The limits the authors name, and the ones in the numbers
A limit is not a reason to dismiss a trial. It tells you how far to carry it.
- Self-report only. Sleep disturbance and duration were reported by participants in the trial. The authors call for objective sleep measures in future work.
- Four weeks. Nothing here speaks to longer use.
- Online recruits, one platform. Participants came from an online recruitment site and used one app's content.
- Missing follow-up. In the stories arm, 96 of 125 people completed follow-up, about three in four.
- A smaller sample than planned. Recruitment stopped short of its target.
- Low engagement. The per-protocol rate in the stories arm was 30.85%, so many people did not use the tool much. A weak dose could mean a weak test.
- The maker's involvement. The app's maker, Unmind, created the interventions and has ties to several authors.
What the trial did not test
Illustrative example: you have had three bad weeks, and you are deciding whether to pay for a story app. This trial answers one narrow question for you: does an app's story arm beat daily app content on a sleep-disturbance score after four weeks? Not significantly.
It did not test stories as part of a wind-down routine. It did not test whether the choice of story matters, or whether people who pick content they like do better. Objective measures such as a wrist sensor were not used either.
Here is where the web splits. Some pages say stories work. The Sleep Foundation's bedtime stories list, updated June 10, 2022, says none looked at adults. That predates the 2023 pilot and this trial, so it describes an older picture. "Stories work" and "there is no research" are both out of date.
What to do with this
Our position: a story app is a reasonable thing to buy because you enjoy it, and a poor reason to buy because you expect it to fix your nights.
The authors take a similar line. They make no specific recommendation, but they say use is not discouraged, particularly for people who find the tools enjoyable or use them in a wind-down routine. That is their suggestion, not a finding.
If your sleep has been poor for months, a doctor is the right call, and the authors say people with persistent problems should seek professional, evidence-based care. If a story is something you simply like, there is no evidence in this trial that you must stop.
Questions people ask
Does this trial prove sleep stories do not work?
No. Its main outcome was self-reported sleep disturbance, in one sample of working adults over four weeks, using one app. It found no statistically significant difference from the control. That is weaker than showing stories do nothing for anyone.
What does Hedges' g of 0.12 mean?
Hedges' g expresses the gap between two groups in units of the spread of scores, so 0 means no gap. A value of 0.12 is a very small gap, and its confidence interval runs from below zero to well above it.
Is it fine to keep listening to sleep stories I enjoy?
The trial's authors say use is not discouraged, particularly for people who enjoy stories or use them in a wind-down routine. That is their suggestion, not something the trial tested. Months of poor sleep are a reason to see a doctor.
Were there any bad effects in the trial?
Twelve participants across all four arms, including two in the control group, reported lasting bad effects. That was 3.58 percent of app downloaders, with no statistically significant difference between the sleep tools and the control. One story-arm participant described relying on the app to get to sleep.
Ho-Hum was not in this trial
Ho-Hum is coming to iPhone. It writes a boring bedtime story that begins with the dull parts of your day, and reads it aloud: an AI writer writes the opening and an AI voice reads it. The trial above tested a different app's stories, and nothing in it speaks to Ho-Hum.