One study in forty-one was properly blinded, and it found nothing
We've graded 41 studies of sound and the brain against a public rubric. Blinding is the weakest criterion by a wide margin, not one study has been independently replicated, and the best-built paper in the library found beats made performance worse.

If you read the headlines about sound and the brain, the research sounds settled. If you read the papers, it isn’t, and the reasons are structural rather than a matter of opinion. We’ve now graded 41 studies against a public rubric, scoring eight things about each one from 1 to 5. The weakest of the eight, by a clear margin, is whether anybody was blinded. Exactly one of the 41 was double-blind. It hid the beat inside a piece of music so that neither the participants nor the experimenter could tell which track was which, and it found no effect on any of four cognitive measures.
Here’s what the whole library looks like as of 16 November 2026, across studies published between 1973 and 2026.
| What we score | Mean out of 5 | How bad the tail is |
|---|---|---|
| Outcome measurement | 4.27 | Only 2 of 41 rely on ad-hoc self-report or worse |
| Peer review and venue | 4.54 | 38 of 41 in indexed journals |
| Study design | 3.85 | 5 of 41 are uncontrolled pre/post or narrative reviews |
| Conflicts of interest | 3.59 | 22 of 41 make no funding or conflict statement at all |
| Control condition | 3.32 | 11 of 41 at 2 or below; 2 have no control of any kind |
| Sample size | 3.24 | 14 of 41 at 2 or below; 5 have fewer than ten participants |
| Replication | 2.71 | Not one of the 41 has been independently replicated |
| Blinding | 2.41 | 26 of 41 at 2 or below; one study is double-blind |
The mean weighted score across the library is 3.45, which lands just inside the band we label moderate. One study carries a strong grade, and it’s a meta-analysis rather than a trial.
These numbers describe our library, not the whole literature, and our library is not a random sample: it’s what a weekly sweep plus a historical backfill has turned up, with a deliberate bias toward foundational papers and toward the studies that consumer pages cite. Treat the numbers as a well-documented slice rather than a census.
Nobody can be blinded to a sound they can hear
This is the field’s central problem and it isn’t anyone’s fault. If your intervention is audible, the participant knows they got it. Silence doesn’t sound like a 10 Hz beat, and a beat doesn’t sound like pink noise. So the person rating how calm they feel knows which arm they’re in, and so does the researcher handing them the headphones.
Twenty-six of our 41 studies score 2 or below on blinding, which on our rubric means either unblinded with an objective outcome, or unblinded with a subjective one. Five sit at the floor: somebody was asked how they felt after hearing a sound they knew they were being given.
Yang and colleagues show how that plays out. Sixty-five students listened to twenty minutes of nature sounds carrying beats at 6, 10 and 25 Hz. Blood pressure and heart rate fell after all three, equally. The authors read this as beats lowering anxiety at any frequency. The other reading is that twenty minutes of nature sounds in headphones is relaxing and the beat frequency did nothing, and the study can’t separate them, because there was no condition without a beat.
Then there’s Pluck and López-Águila, the one study in the library that solved this. They embedded a 6 Hz beat inside a piece of music, so both conditions sounded the same, and ran each participant through both. Afterwards they asked the participants whether they’d heard a rapid regular beat in either track, and asked the experimenter to guess which track carried it, to confirm the blinding had held. Then they measured four kinds of cognitive fluency.
All four came back flat, with effect sizes between 0.01 and 0.11, which is as close to nothing as this kind of measurement gets. The one thing that did move was fear, which went up by a medium-sized amount in the beat condition, from a stimulus nobody could consciously detect. That’s one significant result among a dozen mood ratings and it needs repeating before anyone leans on it. But it’s the only effect in our library measured from a beat the listener couldn’t hear as a beat, and it points away from the thing people buy these recordings for.
Twenty-four people is a small study and a real cognitive effect could hide in it. Set that against the shape of the field, though, and the pattern is uncomfortable: the better the blinding, the smaller the effects. Five further studies got participants blind without reaching double-blind, and three of the five are among the most negative or most qualified results we hold.
The control is usually “nothing”, not “the same thing minus the beat”
The question anybody actually wants answered is whether the beat adds something to the audio carrying it. Answering it needs a control that’s identical except for the beat. Most studies compare against silence, a waiting list, or nothing at all, which answers a much easier question: is listening to pleasant audio better than sitting there?
Two of our 41 have no control condition whatsoever. Mujib and colleagues is the instructive one, because it’s widely cited as showing beats improve memory. Sixty people were split into three groups getting 10, 14 or 30 Hz beats for fifteen minutes, with a digit span test before and after. The 10 Hz group’s score went up with a large effect size. There was no silence group, no sham, and nobody who just took the test twice, which is exactly what a digit span test improves from. The authors name the problem themselves. Whatever that study measured, it can’t tell you the beat did it.
Eight studies do have a properly matched active control, and the effects there are smaller and stranger. Amd put 6 Hz beats inside pink noise and compared them against the same pink noise alone: the beats came out ahead on self-rated calm, measured with a single slider. Venkatesan and colleagues mixed monaural beats into music and compared against the identical music: quietly mixed, the beats lowered anxiety beyond plain music, and mixed in at four times the music’s volume they raised it. Both studies are single sessions, and the second is funded by a record company whose business is selling music.
And the best-controlled study we hold, Klichowski and colleagues, ran 1,000 adults against silence, classical music and a pure tone, and found that beats lowered reasoning scores by 5 to 7 per cent while none of the controls changed anything. It has the highest weighted score in our library, 4.3 out of 5, and we grade it unsupported, because it tested the claim and the claim failed.
So when this field builds the comparison properly, the effects shrink, split or reverse.
Everything is one study that nobody has repeated
Not one of the 41 studies in our library has been independently replicated with consistent results. Not one. Six score a 4, meaning a meta-analysis supports them, and all six are reviews or pooled analyses themselves rather than trials whose protocol somebody re-ran. Eighteen sit at 2 or below: a single study, no replication found.
The library is full of designs that exist exactly once. Derner and colleagues is the only head-to-head comparison of monaural against binaural beats on a behavioural outcome ever published, and it ran in fifteen patients awaiting epilepsy surgery who already had electrodes in their brains. Lee and colleagues swept the beat frequency instead of holding it steady and halved sleep-onset latency in ten good sleepers. Rakhshan and colleagues measured two kinds of working memory at once and found a 10 Hz beat helped one and hurt the other. Each of those is a good idea, well executed. None has been done twice.
Small samples make this worse rather than better. Fourteen of the 41 have fewer than 25 participants and five have fewer than ten. López-Caballero and Escera found no EEG change at any of five beat frequencies and said plainly that their fourteen participants might be too few to detect one. Goodin and colleagues recruited forty-five people and analysed thirty-one, losing fourteen to corrupted files, EEG artifact and two who couldn’t perceive the beats at all.
Sample sizes, at least, have improved. In our library they rise steadily by era, from a mean of 1.33 out of 5 for papers before 2000, to 2.67 for 2000 to 2019, to 3.76 for 2020 onwards. Blinding hasn’t moved: 2.00, then 2.67, then 2.45. Fifty years on, the studies are bigger and no better concealed.
The core measurement is contested
Underneath all of this sits a measurement problem that most readers never hear about. The standard evidence that a sound entrained your brain is a peak in the EEG at the beat frequency. Duecker and colleagues’ methods review sets out why that isn’t enough: a brain producing a fresh evoked response to each pulse makes a scalp signal that can look exactly like an ongoing oscillation being pulled into step. Telling them apart takes tests most studies skip, such as checking whether the rhythm outlasts the sound.
Dos Anjos and colleagues found that plain white noise, which contains no beat frequency at all, raised gamma power, so part of what every one of these stimuli does is simply be a sustained sound in your ears. And in Ávila’s 2026 study the apparent gamma increase at 40 Hz disappeared once the truly rhythmic part of the signal was separated from the broadband background. What had changed wasn’t rhythm. A large share of 40 Hz audio claims rest on the measurement that study took apart.
Most papers don’t say who paid for them
Twenty-two of our 41 papers, more than half, make no statement about who funded them or whether anyone had a competing interest. On our rubric that scores a 3, the neutral middle, because an absent statement isn’t evidence of a problem. It’s still an odd gap in a field where a lot of the research sits close to people selling apps and recordings.
The counter-example is the paper with the heaviest commercial tie in the library. Venkatesan and colleagues’ 2026 trial was funded by Universal Music Group and all four authors work there, and they say so in the paper and told the ethics board and the participants. We score their conflict at the floor and their disclosure is exemplary. Those aren’t in tension. Declaring a tie is what lets a reader weigh it.
What this means if you’re reading a headline
Nothing here is medical advice, and none of it says sound does nothing. But some things follow from the numbers themselves, not from anyone’s opinion.
- “Studies show” is doing a lot of work. For most outcomes in this field, it means one study, in a few dozen people, that nobody has repeated.
- Ask what the comparison was. If a study compared a beat against silence, it measured listening to something. If it compared the same audio with and without the beat, it measured the beat. Only eight of our 41 did the second thing.
- A brainwave graph is not an outcome. Power changing at the beat frequency is compatible with entrainment and with an evoked response, and sustained sound of any kind moves it.
- The strongest result in the field isn’t about any of this. It’s a meta-analysis of fifteen randomised trials of anxiety before and after surgery, where the effect is large and the patients are a specific, acutely stressed group.
- Weak evidence is not evidence against. Most of these grades mean nobody has run the study properly yet. That’s a different statement from “it doesn’t work”, and we try hard to keep the two apart.
The field’s strongest work is good. There just isn’t much of it, nobody has repeated any of it, and the better a study is built the more modest its findings get.
Sources
All figures recomputed from our library on 16 November 2026, across 41 graded summaries. The rubric and its weighting are on our methodology page.
- Pluck and López-Águila (2019), a double-blind null on cognitive fluency, and induced fear
unsupported - Klichowski and colleagues (2023), home-use beats lowered reasoning scores in 1,000 adults
unsupported - López-Caballero and Escera (2017), five beat frequencies, no change in EEG power
unsupported - Goodin and colleagues (2012), high-density EEG, no effect on vigilance or cortical power
unsupported - Mujib and colleagues (2021), digit span after 10 Hz beats, with no control arm
weak - Yang and colleagues (2025), theta, alpha and beta beats all lowered blood pressure, and equally
weak - Lee and colleagues (2024), a sweeping beat and sleep-onset latency in 10 good sleepers
weak - Derner and colleagues (2018), monaural against binaural beats on item and source memory
mixed - Rakhshan and colleagues (2022), one beat, two kinds of working memory, opposite results
mixed - Venkatesan and colleagues (2026), the same music with and without the beats, in 466 anxious adults
mixed - Duecker and colleagues (2024), is “entrainment” what the brain is actually doing? A methods review
mixed - Ávila and colleagues (2026), binaural beats added to visual flicker
mixed - Amd (2026), self-administered beats against pink noise and silence
moderate - Dos Anjos and colleagues (2024), isochronic-style tones shift EEG power more than gamma binaural beats
moderate - Xiong and colleagues (2025), binaural beats for perioperative anxiety and pain: a meta-analysis of 15 trials
strong


