Search for help with snoring and you will meet a family of exercises: press the tongue to the roof of the mouth, hold, repeat; purse the lips against resistance; hum vowels; chew with exaggerated care. The pitch is appealing — no machine, no mask, a few minutes a day to train the muscles that hold the airway open at night.

The largest randomised test of that idea so far has now reported, and the result is unusually flat. Twelve weeks of a standardised orofacial myofunctional therapy programme did not reduce the number of times breathing stopped or slowed per hour of sleep, and did not make participants measurably less sleepy during the day.

The trial appears in the Journal of Clinical Sleep Medicine (issue dated 10 November 2026) and was run from two centres, Akershus University Hospital in Norway and a private clinic in Tallinn, Estonia. Of 283 patients screened, 104 adults with newly diagnosed mild-to-moderate obstructive sleep apnea were randomised, half to the exercise programme and half to a waiting list.

What the exercises were, and how the result was measured

The protocol was not vague. It comprised ten isotonic and isometric exercises for the face, tongue and soft palate, taught in a single 60-minute session and then performed at home three times a day for 40–60 minutes in total, with optional video check-ins with a therapist every two weeks. Adherence was logged daily in a phone app.

Measuring the outcome was unusually careful for this literature. Rather than one night in a laboratory, participants used self-applied home polysomnography on three consecutive nights before and after the intervention, scored blindly by four technologists — a design meant to blunt the night-to-night variability that can make a single sleep study misleading.

The primary measure was the apnea–hypopnea index, or AHI: apneas and hypopneas per hour of actual sleep. The paper sets out the usual bands — 5 or more with symptoms for mild, 15 or more for moderate, 30 or more for severe — and this group sat in the middle, with an average baseline AHI of about 17 to 18.

No difference in breathing events, and none in sleepiness either

At the end of the 12 weeks the two groups were effectively indistinguishable. Mean AHI was 19.1 in the exercise group and 19.2 in the waiting list, a between-group difference in change of 0.6 events per hour (95% confidence interval −3.3 to 4.5; P = .774). An interval that wide cannot exclude small effects in either direction, but it rules out anything resembling the halving of AHI that the promotional version of this therapy implies.

Follow-up analyses did not rescue the result. Restricting the comparison to people who completed the protocol properly — 21 in the exercise arm, 37 in the waiting arm, with average adherence of 92.4% — gave a difference of −1.3 events per hour (95% confidence interval −4.9 to 2.2; P = .460). Limiting the analysis to breathing events that happened while participants lay on their backs gave −4.7 (confidence interval −13.3 to 3.9). Neither reached significance.

The predefined secondary outcomes were equally still. Daytime sleepiness on the Epworth scale moved by −0.1 points between groups (95% confidence interval −1.8 to 1.5), and neither desaturation measure changed — nor the oxygen desaturation index, average overnight saturation or time spent below 90%.

Adherence, in other words, was not the explanation. Participants managed all three daily sets on 64.6% of days on average, and at least one full set on 73.3% of days. More adherence did not buy more improvement: the association between adherence and AHI change was essentially zero (β = −0.1, 95% confidence interval −0.2 to 0.1).

Why the earlier reviews promised so much more

This is where the study becomes interesting rather than merely negative. Reviews of this therapy have generally been encouraging. A systematic review and meta-analysis published in Sleep in 2015 pooled nine adult studies covering 120 patients and reported the AHI falling from a mean of 24.5 to 12.3 events per hour, a mean difference of −14.26 (95% confidence interval −20.98 to −7.54). The authors summarised the field as showing reductions of roughly 50% in adults and 62% in children, with gains in lowest oxygen saturation, snoring and sleepiness.

Read the method, though, and the shape of that evidence changes. The pooled figures were before-and-after comparisons within the same patients, not head-to-head results against a control group, and heterogeneity was severe (I² = 91%). Removing two studies brought the mean difference down to −10.49 (confidence interval −12.67 to −8.31) with heterogeneity close to zero, and almost every study ran only two to six months.

The new trial’s authors concede that reviews suggest benefit, but note that “several trials included in these reviews reported no significant effect on AHI”. Their explanation is a distinction between muscle activation and task-specific retraining: the trials reporting large reductions bundled strengthening with retraining of habitual patterns — nasal breathing, resting tongue posture, swallowing, chewing — while isolated strengthening of individual muscles has repeatedly failed. Training the genioglossus for strength, on this reading, is not the same as retraining how the airway is used.

The one number that moved, and why it is a footnote

There was a single statistically significant finding, and it came from the per-protocol sample: the proportion of total sleep time spent snoring fell by 5.7 percentage points more in the exercise group (−11.1 to −0.3; P = .041). The authors themselves treat this cautiously. The analysis was exploratory, no adjustment was made for the number of comparisons, and their interpretation is that it may reflect a modest effect on the vibration of the upper airway rather than on airway collapse. Reductions in snoring without changes in AHI have shown up in earlier trials too, and snoring is a symptom, not the disease.

Tolerability was reasonable. Nine participants reported transient muscle or mouth soreness; two had worsening of jaw-joint symptoms and two had flare-ups of pre-existing headaches. No severe events were judged likely to be related to the exercises.

What the trial could not settle

The authors are candid about the limits, and they matter for how far the null result should be pushed. Per-protocol numbers were smaller than the 26-per-group minimum their own power calculation required, because 15% of participants who finished the study lacked a valid sleep recording at one or both time points, and recruitment stopped for administrative reasons. A type 2 error — missing a real but modest effect — cannot be excluded.

Nor was the group selected for the traits that might predict a response. No endotyping was done, nasal resistance was not measured, and craniofacial features that were collected went unanalysed. Participants were also relatively uncomplicated: body mass index under 30, no prior sleep apnea treatment. The authors suggest people with more severe apnea, those who cannot tolerate CPAP, or those with particular anatomical or physiological subtypes might still respond.

That matters because the standard treatment has its own adherence problem: the trial’s introduction puts adherence to positive airway pressure therapy at 30–60%, which explains why a cheap exercise programme would be attractive if it worked. The NHLBI’s treatment page, last updated in January 2025, still lists orofacial therapy among the options, describing it as something that “may also be an effective treatment for sleep apnea in children and adults”. One well-run trial does not overturn an evidence base, but that sentence now rests on weaker support than its placement suggests.

Where this leaves the everyday question

Snoring is common, and most of it is not sleep apnea. MedlinePlus notes that people with apnea often snore loudly, but that not everyone who snores has the condition — and the useful distinction is not the noise but whether breathing actually stops, whether sleep fragments, and whether the daytime consequences show up as drowsiness or poor concentration.

Untreated apnea is not a cosmetic problem. The NHLBI points to higher risks of stroke and heart attack, MedlinePlus to more car crashes and work accidents, on top of the everyday cost of broken sleep. Diagnosis runs through a medical history, an examination and a sleep study, not an app. We have covered separately what the evidence says about sleep regularity and about a digital insomnia treatment that was tested properly and worked; the contrast is instructive, because the interventions that hold up are the ones measured against a control group.

For anyone tempted to try tongue exercises before mentioning their snoring to a clinician, the honest summary is that a supervised 12-week programme produced no measurable change in breathing events or sleepiness in the population most likely to be offered it. That is not grounds to dismiss the therapy everywhere, and it is certainly not grounds to buy anything. It is grounds to have the snoring assessed rather than self-treated. As always, this is information rather than advice: anyone with symptoms suggesting apnea should talk to a healthcare professional, and our medical disclaimer sets out the limits of what we publish.