Neural Entrainment: How the Brain Locks Onto a Beat Without Being Asked

Introduction

Somewhere in the last week you nodded to something.

A song in a shop. A drum fill in a car advert. The rhythm of a train over the rails. You did not decide to do it, and if someone had asked you a second earlier whether you were listening, you would probably have said no.

That is the thing worth explaining. Not that people like music. That your head found the beat while you were thinking about something else.

Neuroscience has a name for the phenomenon and a large literature behind it. Play a steady rhythm to a person and their brain produces activity at the frequency of that rhythm. Put electrodes on the scalp, run the recording through a frequency analysis, and there it is: a peak sitting exactly where the beat is, and smaller peaks at its multiples. Nobody in the field disputes that this happens.

What happens next in most articles is a small, quiet leap. The peak gets described as your brain waves synchronising to the music. Your internal oscillators, the story goes, are being captured by the external rhythm, the way two pendulum clocks hung on the same wall drift into step. It is a lovely image. Christiaan Huygens noticed the clock version of it in 1665, while ill in bed and watching two of his own pendulum clocks on a shared beam.

That leap is where the field splits, and it has been split for a long time.

Because there is a second explanation, and it is much less romantic. Maybe nothing is being captured at all. Maybe the auditory cortex simply fires a short, stereotyped response to each acoustic event, the way it fires to any sudden sound, and because the events arrive regularly the responses line up in the recording and produce a peak at the beat frequency. No oscillator. No capture. Just a row of reactions, evenly spaced.

Both accounts predict a peak. That is the problem.

This article is about what sits underneath that peak. It covers what the evidence actually shows, which parts are settled, which parts are argued over by named researchers who disagree in print, and which parts are being sold to you on the strength of a claim the literature does not support. The short version: something real and involuntary is going on, it is already running two days after birth, and the mechanism is not resolved.

The Oldest Version of the Experiment Used a Flashing Lamp

The founding observation in this field is not about music. It is about light.

In 1949 V. J. Walter and Grey Walter published a study of what they called the central effects of rhythmic sensory stimulation [1]. They sat people in front of a flickering lamp, varied the flicker rate, and watched the EEG. The brain's electrical activity picked up the rate of the flashes. They called it photic driving, and it became a standard clinical tool, because in some people a particular flicker rate provokes a seizure.

Two things about that experiment still matter.

The first is that it works across senses. The effect was never a music phenomenon. It is a rhythm phenomenon, and sound is simply the modality where it is strongest and easiest to study, partly because the auditory system encodes timing with more precision than any other sense. Touch works too. A rhythm delivered as vibration to the skin produces beat-related responses of its own [2].

The second is that photic driving was described decades before anyone had a theory of what it meant. That gap between a reliable effect and an agreed mechanism has never really closed. It just moved.

The theory arrived properly in 1999, when Edward Large and Mari Riess Jones published an account of attention as something rhythmic rather than steady [3]. Their claim was that attention pulses. It rises and falls, it can be entrained by structure in the world, and it delivers its peaks where the next important event is expected. On that account the brain is not following the beat. It is betting on it.

Hold on to that word. Betting. It comes back repeatedly, and it is the single idea that separates this topic from a curiosity about music.

If you want the underlying biology of how populations of cells produce these rhythms in the first place, the piece on how neurons communicate covers the machinery this article assumes.

What the Measurement Actually Measures

Almost every study you will read about here rests on one method, and the method repays a minute of your time, because most of the argument is about how to interpret it.

It is called frequency tagging. Sylvie Nozaradan and colleagues made it the standard approach for beat research in 2011 [4] and extended it the following year to show that the response tracked not just the beat but the meter, the grouping of beats into bars [5].

The recipe is simple. Play a rhythm at a known rate. Record the EEG. Transform the recording into the frequency domain, which converts a wiggly line over time into a set of peaks showing how much activity sits at each rate. Then look at the height of the peak sitting exactly at the beat frequency.

A rhythm at 120 beats per minute is a rhythm at 2 hertz. If the brain is tracking it, there is more activity at 2 hertz than there would otherwise be, plus activity at 4 and 6 hertz, the harmonics.

That is the whole measure. It is elegant, it is cheap, it works on people who are not doing anything, and it has one large weakness that the field has spent fifteen years arguing about: a peak in the frequency domain tells you that something in the brain is periodic at that rate. It does not tell you what.

Molly Henry, Björn Herrmann and Jessica Grahn laid out that limitation directly in 2017, in a paper about what can and cannot be learned by comparing brain signals against the sound that produced them [6]. It is a careful piece and its conclusion is uncomfortable. Much of what looks like brain activity in these recordings is a reasonably faithful copy of the stimulus, and separating the copy from anything the brain added is harder than it looks.

There is a further wrinkle, and it is a practical one. Katharina Duecker, Keith Doelling, Assaf Breska and Emily Coffey wrote a 2024 review of the methodological state of this field with the flat title Challenges and Approaches in the Study of Neural Entrainment [7]. It is not a takedown. It is a list of the ways this measurement can mislead you, written by people who use it.

The Beat That Is Not in the Sound

If the frequency peak were nothing but an echo of the sound, the peak would have to be where the sound is. Sometimes it is not, and those are the results that keep the interesting version of this story alive.

Start with syncopation. In a syncopated rhythm the strongest perceived pulse can fall where nothing is played. You feel a beat in a gap. It is the basis of a great deal of dance music, and it is a genuinely strange thing for a nervous system to do.

Idan Tal, Edward Large and colleagues built rhythms designed to exploit this and published the result as the missing-pulse phenomenon [8]. The stimulus had essentially no acoustic energy at the pulse rate people reported feeling. The brain response had a peak there anyway.

Jan Stupacher and colleagues came at the same question from another direction, with drum patterns containing silent breaks [9]. The rhythm stops. The beat-related response does not stop with it.

Take those two together and one conclusion is hard to avoid. Whatever is producing that peak, it is not a straightforward transcription of the waveform. Something is filling in.

Then there is bass. Tomas Lenc, Peter Keller, Manuel Varlet and Sylvie Nozaradan showed that low-frequency sounds strengthen neural tracking of the beat, using EEG from fourteen listeners [10]. That is a small sample and the finding needs replication before anyone leans on it.

It is the sort of result that fits something people already suspect, which is that a bass line moves you when the same rhythm on a piccolo does not. Fitting a suspicion is not confirming one. A dozen-odd listeners in a lab cannot tell you why nightclubs are built the way they are.

Even your own movement changes the recording. Baptiste Chemin, André Mouraux and Sylvie Nozaradan gave people an ambiguous rhythm, one that can be heard in two different metres, and had them move their bodies to one interpretation of it [11]. Afterwards, the brain response matched the interpretation they had moved to. The rhythm in the air had not changed. What the listener did with their body changed what their auditory cortex did with it.

And how much you enjoy the music matters as well. Wiebke Trost and colleagues found that entrainment to musical rhythm varied with how pleasant listeners found the piece [12]. Which raises an awkward possibility for the tidy version of this story: the measurement may be partly a measurement of your mood.

Two Explanations That Fit the Same Data

The oscillator account says the brain contains genuine self-sustaining rhythms, generated by populations of neurons, and that a regular external rhythm can pull those internal rhythms into step. On this view entrainment is capture. The brain's own clock gets adjusted, its high-excitability phases start landing on the beats, and sensory processing is sharpened at exactly the moments something is likely to happen.

The evoked account says there is no capture. The auditory cortex produces a short response to each acoustic event. Play events regularly and the responses repeat regularly. The frequency analysis then shows a peak at the repetition rate, because that is what a frequency analysis does to any repeating waveform. The rhythm is in the input and in the arithmetic. It is not in an oscillator.

The oscillator account is not hand-waving, either. It has been built into working models. Edward Large, Jorge Herrera and Marc Velasco described networks of coupled nonlinear oscillators that reproduce beat perception from an audio signal, including the case where the perceived pulse is not the loudest thing in the sound [13]. Feed those models a rhythm and a pulse falls out. That is a real achievement, and it is also the reason the argument is hard: a model that reproduces the data is not proof that the brain uses that model.

Both are respectable. Both are held by serious people. Here is who says what.

The oscillator side has a specific and testable prediction, and it is a good one: if a real oscillator was captured, it should keep going for a moment after the driver is removed, the way a pushed swing keeps swinging. Sander van Bree and colleagues went looking for exactly that and reported sustained neural rhythms outlasting the stimulus in a speech task, working with eighteen participants who completed both of their experiments [14]. Benedikt Zoefel, Sanne ten Oever and Alexander Sack had already laid out the broader case in a 2018 review whose subtitle is the whole argument: more than a regular repetition of evoked neural responses [15].

The evoked side has an answer to the persistence evidence, and it is not a dodge. A response that outlasts the sound can also be produced by the tail of the last evoked response, and by the analysis itself: narrowband filtering smears activity forwards in time, so a rhythm can appear to continue after the stimulus stops even when nothing in the brain is still oscillating. That is one of the specific traps the methodological reviews warn about [7], and it is why the persistence test has not closed the argument on its own [16].

The evoked side has been just as direct. Giacomo Novembre and Gian Domenico Iannetti published a piece in 2018 whose title is a question aimed squarely at the beat literature, asking whether tagging the musical beat measures neural entrainment or ordinary event-related potentials [17]. Yulia Oganian and colleagues went further in 2023 with magnetoencephalography from twelve participants, concluding that phase alignment to the envelope of speech reflects evoked responses to sharp acoustic edges rather than oscillatory entrainment [18].

Notice how this disagreement is being conducted. Nobody says the peak is an artefact. Nobody says the recordings are wrong. Everyone agrees on what the data look like. The fight is over which of two mechanisms puts them there, and that is a far more specific question than the popular version of it.

Assaf Breska and Leon Deouell had reported something similar in 2017 from a different angle. Delta-band phase-locking, they found, tracked how predictable the timing was rather than reflecting rhythmic entrainment as such [19]. Lars Meyer, Yue Sun and Andrea Martin summed up their position in a 2020 title that has been quoted a great deal since: synchronous, but not entrained [20].

Between the two camps sits a critical review from Saskia Haegens and Elana Zion Golumbic, which examined how much of the rhythmic-facilitation literature actually supports the claims made on its behalf, and found the answer to be less than advertised [21]. Peter Lakatos, Joachim Gross and Gregor Thut tried to organise the whole mess in 2019 with a framework paper, A New Unifying Account of the Roles of Neuronal Entrainment [22].

None of this means the effect is fake. It means the label is doing more work than the evidence licenses.

The bottom of that diagram is where the trouble lives. Each branch was supposed to predict something the other could not produce. The next section is about the second attempt to find such a prediction, and what happened when somebody tested it.

The Test That Was Supposed to Settle It

By 2019 there was a serious attempt to decide between the accounts by modelling rather than arguing.

Keith Doelling, Florencia Assaneo, Dana Bevilacqua, Bijan Pesaran and David Poeppel took recordings of people listening to music and asked which of two computational models reproduced the data better [23]. One model contained an oscillator. The other did not. They reported that the oscillator model fit better, particularly at slow rates, and the paper's title says so without hedging: an oscillator model better predicts cortical entrainment to music. The dataset they modelled came from twenty-seven listeners.

For a while that looked close to decisive.

Then in 2025 Atser Damsma, Mitchell de Roo, Keith Doelling, Pierre-Louis Bazin and Fleur Bouwer published a study in Cerebral Cortex with twelve participants, aged twenty-one to thirty-one, that tested a different signature [24]. The signature was tempo dependence. Neural responses at the beat frequency are selectively enhanced, and the size of that enhancement varies with tempo, and this had been treated as something an evoked model could not produce.

It can. Both an oscillator model and an evoked model reproduced the same tempo-dependent enhancement.

That result deserves a moment, and it is almost entirely absent from the popular coverage. The measurement that was supposed to separate the two accounts does not separate them. Not "the evidence is mixed". The specific fingerprint does not distinguish the specific hypotheses.

There is a detail in the author lists that says something about how this field works. Keith Doelling led the 2019 paper that made the oscillator case, and he is also a co-author of the 2025 paper that dissolves it. People here are arguing with their own earlier results rather than defending them.

The same group followed up with a broader piece on modelling rhythmic expectation, laying out where these models can mislead you and why several published comparisons prove less than they appear to [25]. A 2026 paper in eNeuro added a blunter warning still, arguing that many naturalistic studies of entrainment are missing a control for the stimulus itself, so that properties of the sound get read as properties of the brain [16].

And there is a quieter problem underneath all of it. Yuranny Cabral-Calderin and Molly Henry examined how reliable the entrainment measure is when you record the same person twice [26]. If a measure is noisy across sessions, then differences between groups, between musicians and non-musicians, between children with and without dyslexia, get harder to trust than the published p-values suggest.

This does not make the field disreputable. Fields that check themselves like this are the healthy ones. But it does mean that anyone telling you confidently what your brain waves are doing while you listen to a playlist is ahead of the evidence.

Three Different Things Are Wearing the Same Word

Part of the confusion you meet online is not scientific at all. It is vocabulary. The word entrainment gets used for three separate claims of very different quality, and almost nobody separates them.

Mixing these three is how a genuinely open scientific question turns into a marketing claim. The first is a measurement. The second is a puzzle. The third is a product category, and it comes back near the end of this article, because the answer there is short and not very flattering.

Fourteen Newborns

If you only remember one experiment from this article, make it this one.

In 2009 István Winkler, Gábor Háden, Olivia Ladinig, István Sziller and Henkjan Honing tested fourteen healthy full-term newborns [27]. The infants were between thirty-seven and forty weeks gestational age, and they were tested on day two or three after birth. Most of them were asleep.

They played rock drum patterns. Occasionally they left out the downbeat, the first and strongest beat of the bar. Not a quiet beat. An absent one.

The newborn brains produced a mismatch response, the signature the brain generates when a sensory expectation is violated. Something in a two-day-old had worked out where the downbeat should be, noticed it was missing, and objected.

Sit with what that rules out. No training. No practice. No instruction. No culture, no exposure to a particular musical tradition, no motivation, no task, and in most cases no wakefulness. The capacity to extract a pulse from a pattern of sounds is there before almost anything else is.

The result needs stating precisely, because it is easy to stretch. Fourteen infants is a small sample. The measure is a mismatch response, which tells you the brain registered a violation of an expectation, not that the babies felt a groove. And a downbeat can be inferred from the pattern of loud and quiet sounds without anything oscillating anywhere. What the result establishes is narrower than the headline and still remarkable: two days into a life, the auditory system is already building expectations about when the next event should arrive, and complaining when it does not.

Laura Cirelli, Christina Spinelli, Sylvie Nozaradan and Laurel Trainor followed this line into infancy, measuring neural entrainment to beat and meter in babies and finding that musical background shaped the response [28]. So the capacity is early, and experience then works on it.

Fleur Bouwer put the underlying question directly in the title of a 2022 paper: is neural entrainment to auditory rhythms automatic or top-down driven [29]. The answer is closer to both than to either. There is a floor that runs without you. Above that floor, what you attend to and what you know push the response around.

That floor is why the title of this article says "without being asked". It is not a figure of speech. Nobody asked those newborns.

What Turns It Up, and What Does Not

If entrainment were purely automatic it would be uninteresting for anything except a textbook. It is not purely automatic, and the pattern of what moves it is revealing.

Attention moves it. Aeron Laffere, Fred Dick, Lori Holt and Adam Tierney showed attentional modulation of entrainment to sound streams in children with and without ADHD, which matters because attention is the thing that has to be recruited if this response is going to be useful for anything [30]. Peter Lakatos and colleagues had already made the case in a 2008 macaque study that entrainment is a mechanism of attentional selection, one of the most-cited results in this whole area [31]. Jonas Obleser and Christoph Kayser wrote the review that keeps attention properly in the frame when people talk about the listening brain [32].

Training moves it. Keith Doelling and David Poeppel found that cortical entrainment to music was modulated by musical expertise [33]. Musicians are not simply better at reporting the beat. Their recordings look different.

Both of those results cut towards the oscillator side, and it is worth seeing why. A pure chain of reactions to sound should not care much about training or about where you point your attention, because the sound is identical either way. Something that changes with what the listener knows and what the listener attends to is doing more than reacting. That argument is suggestive rather than decisive, since an evoked response can be modulated too, but it is the reason these findings keep getting cited in the mechanism dispute rather than filed as trivia.

Individual differences track behaviour. Sylvie Nozaradan, Isabelle Peretz and Peter Keller found that how strongly a person's cortex tracked the beat correlated with how well they predicted it in a synchronisation task [34]. María Noboa, Csaba Kertész and Ferenc Honbolygó reported in 2025 that entrainment to the beat and working memory together predicted sensorimotor synchronisation skill [35].

That link to behaviour is what keeps this measurement from being a curiosity. If cortical tracking predicted nothing a person actually does, it would be a number with no owner. Bruno Repp's review of the tapping literature remains the reference point for what that behaviour looks like when measured properly [36], and Nozaradan and colleagues later captured the coupling between the neural and the behavioural side in a single EEG paradigm [37].

Real music moves it too, which is less obvious than it sounds. Most of this research uses metronomes and synthesised drum patterns, because they are controllable. Adam Tierney and Nina Kraus recorded listeners hearing an actual song, a Bo Diddley track, and found the brain response tracked the rhythmic structure of the music rather than only its loudest onsets [38]. It is one of the most widely covered studies in this area, and the popular pieces built on it are now more than ten years old.

So metronome studies and music studies are not interchangeable, and the difference runs the way any musician would guess. A real performance carries timing a grid does not have.

Groove moves it, but not in the way you would guess. Daniel Cameron, Ioanna Zioga, Job Lindsen and Marcus Pearce found that entrainment tracked how groovy listeners rated a rhythm for rhythms that had been performed by a human, and not for mechanically generated ones [39]. Something about the microtiming of a real performance is doing work that a perfect grid does not do.

Now the part that gets left out.

Joshua Hoddinott, Molly Henry and Jessica Grahn asked in 2026 whether familiarity built through training changes the response. They trained people on half a set of rhythms across four sessions, then compared the EEG. The title states the finding: experience-driven predictability does not influence neural entrainment to the beat [40]. Fifteen participants had usable data across both sessions, one analysis came down to thirteen, and the authors flag the small sample themselves.

That is a negative result on a small sample, and it should be read as such. It is in this article because articles that only report the findings pointing one way are how a field's uncertainty gets laundered into confidence. If you want more on how attention and memory interact in general, the piece on attention and memory goes deeper than there is room for here.

It Was Never Only About Hearing

There is a fact about beat perception that surprises people every time, and it reframes the whole topic.

When you listen to a rhythm without moving at all, your motor system activates.

Jessica Grahn and Matthew Brett established this in 2007 with a study of rhythm and beat perception in motor areas of the brain [41]. Listeners lay still. The basal ganglia and supplementary motor area lit up anyway, and more for rhythms with a strong beat than for rhythms without one. Grahn and Rowe followed up on the premotor and striatal interactions involved [42], and Grahn wrote up the role of the basal ganglia specifically [43].

That direction of travel had been anticipated. Robert Zatorre, Joyce Chen and Virginia Penhune had already written the review that framed music perception as an auditory-motor problem rather than an auditory one [44].

This is the finding that changes how the whole topic feels. You are not receiving a rhythm. Some part of you is producing one, quietly, and checking it against what arrives.

So the beat is not something your ears hand to your brain. It is something your movement system helps construct, whether or not you move.

Takako Fujioka, Laurel Trainor, Edward Large and Bernhard Ross found the clearest neural signature of that construction. The internalised timing of a steady sequence shows up in beta-band oscillations, in a pattern that rises and falls in step with the beat and does so even when a beat is omitted [45]. Beta is a motor-system rhythm. It is doing timekeeping.

Aniruddh Patel and John Iversen turned this into a specific proposal, the action simulation for auditory prediction hypothesis, which says the motor system simulates the rhythm in order to generate predictions that are then sent back to auditory regions [46]. Jonathan Cannon and Patel updated the argument in 2021 under a title that captures it neatly: how beat perception co-opts motor neurophysiology [47]. Giorgio Lazzari and colleagues added causal evidence in 2025, mapping the functional organisation of beat perception in human premotor cortex [48].

That loop is why beat perception recruits the cerebellum and the rest of the procedural timing machinery rather than living in auditory cortex alone.

And here too there is a result pushing back. In 2026 Samantha O'Connell and colleagues reported that listening to groovy music did not enhance primary motor cortex activation [49]. Primary motor cortex is not the same as premotor cortex, and nobody claims the whole motor hierarchy behaves identically. Still, the tidy story that groove drives the motor system straight through to its output stage did not survive that test.

Prediction, Not Reaction

Everything above starts to make sense once you stop thinking of the brain as following the beat and start thinking of it as forecasting the beat.

The difference is measurable, and the cleanest demonstration comes from rats. Vani Rajendran, Jan Schnupp and colleagues showed that rats synchronise predictively to metronomes rather than reacting to each click [50]. Their movements anticipate. A reaction arrives after the stimulus. A prediction arrives with it or slightly before, and that timing signature is what distinguishes the two.

The distinction is not a technicality. A system that reacts is always late by the length of its own reaction. A system that predicts can be on time, and can also be wrong in a way a reactive system never is. Both of those are things you have felt.

Human listeners do the same thing, which is why you can clap on the beat instead of just after it, and why a song that suddenly drops a beat feels like a stumble. The bet was already placed. It lost.

That predictive framing is also what connects rhythm to attention. If your brain expects an event at a particular moment, it can prepare, and preparation costs less than constant vigilance. This is the functional argument for why entrainment would be worth having at all. A brain that knows when to look does not have to look all the time. The same logic runs through everything in how sensory information is briefly held and sampled before it either becomes something you notice or disappears.

Note what this does to the mechanism argument. Prediction is easier to build from an oscillator than from a chain of reactions, which is a point in the oscillator column. It is not decisive, because an evoked-response system with a timing model attached can also anticipate. But it explains why the oscillator account has been attractive for so long.

The Same Argument, Playing Out in Speech

Rhythm research and speech research have been fighting the same battle in parallel, and the speech half matters here because that is where the strongest causal evidence lives.

Speech has a rhythm. Not a metronomic one, but a reliable rise and fall of energy at roughly the rate of syllables. Jonathan Peelle and Matthew Davis argued in 2012 that neural oscillations carry that speech rhythm through to comprehension [51]. Nai Ding, Lucia Melloni, Hang Zhang, Xing Tian and David Poeppel then showed something more striking: cortical activity tracks hierarchical linguistic structures, phrases and sentences, that have no direct acoustic signature at all [52].

Stop and consider how odd that is. A phrase boundary is not a sound. It is a fact about grammar. Yet the recording carries a rhythm at the rate the phrases arrive, in listeners who understand the language and not in listeners who do not.

That is the speech version of the missing pulse. The brain produces a response at a rate that is not in the sound.

The causal work is where speech pulls ahead of music. Lars Riecke, Elia Formisano, Bettina Sorger, Deniz Başkent and Etienne Gaudrain used electrical stimulation to manipulate the phase of ongoing activity and found that doing so changed how intelligible speech was [53]. Anne Kösem and colleagues showed that entrainment influenced which words listeners reported hearing, in a study whose title claims exactly that [54].

Manipulating the brain rhythm changed the percept. That is a stronger kind of evidence than any correlation, and it is why the oscillator account has not simply been abandoned in the face of the evoked-response critiques.

It is also, awkwardly, the same domain where Oganian's group found the strongest evidence for the evoked account. Both results are about speech. Both are careful. They have not been reconciled.

Which Animals Feel a Beat, and Why the Answer Kept Changing

This is the part of the story with characters, and it is the part where a confident hypothesis met a series of animals that had not read it.

In 2006 Aniruddh Patel proposed that the ability to synchronise movement to a musical beat depends on the brain circuitry for complex vocal learning, the machinery that lets a species learn to produce new sounds by imitation [55]. Humans have it. Parrots have it. Songbirds have it. Most mammals, including our closest relatives, do not.

The prediction was sharp and testable. Only vocal learners should be able to keep a beat.

Then came Snowball. In 2009 Patel, John Iversen, Micah Bregman and Irena Schulz published experimental evidence of beat synchronisation in a nonhuman animal, working with a single sulphur-crested cockatoo [56]. What made the result convincing was not the dancing but the tempo manipulation: when they sped the music up and slowed it down, the bird adjusted to stay with it, which is the difference between synchronising and simply moving rhythmically.

One bird is one bird. A single animal cannot tell you what a species can do, let alone what a clade can do, and the authors were careful about that. What it can do is kill the claim that nothing outside humans manages it at all. That is the work Snowball did.

A cockatoo is a vocal learner. So far the hypothesis was holding.

Then a sea lion called Ronan learned to bob her head to a beat, and Peter Cook, Andrew Rouse, Margaret Wilson and Colleen Reichmuth published it in 2013 [57]. A California sea lion is not a vocal learner in the sense the hypothesis required. One animal, again, and again the tempo transfer was the point: she generalised to tempos she had never been trained on, and to real music. Rouse and colleagues later reanalysed her behaviour as a coupled oscillator system [58].

Ronan mattered more than Snowball, and for an unglamorous reason. Snowball confirmed a prediction. Ronan broke one. A theory that survives its confirmations and fails its first serious test is in trouble, and this is the point where the vocal learning story stopped being tidy.

Meanwhile the primate results were going the other way, which made the hypothesis look better. Henkjan Honing, Hugo Merchant and colleagues tested rhesus monkeys in 2012 and found they detected rhythmic groups in music but not the beat [59]. Merchant, Grahn, Trainor, Martin Rohrmeier and Tecumseh Fitch reviewed the whole cross-species picture in 2015 [60], and Carel ten Cate, Michelle Spierings, Jeroen Hubert and Honing did the same for birds [61].

Then rats.

Yoshiki Ito and colleagues reported in 2022 that rats show spontaneous beat synchronisation, with head movements and auditory cortical activity both tuned to the same tempo range humans prefer, between 120 and 140 beats per minute [62]. The movement analysis covered ten animals and the neural analyses seven to nine, depending on the measure. Their explanation is the interesting part. They argue the tuning comes from the time constant of neural dynamics, which is conserved across species, rather than from body size or step rate.

That result undercuts a claim most people have heard. The idea that 120 beats per minute feels natural because it is near a resting heart rate, or near walking pace, has a problem: a rat's heart rate and stride are nothing like yours, and the rats peaked in the same window.

Taken seriously, the tuning claim changes what a beat even is. If the preferred tempo range came from the body, it should differ wherever bodies differ. If it comes from how fast populations of neurons settle and recover, it should be shared. The rat data point at the second answer. That is one study in one species with about ten animals, so it is a lead rather than a conclusion, but it is the kind of lead that changes what you would look for next.

Patel has continued to develop that argument, and its mature form treats vocal learning as a preadaptation for beat perception rather than a strict requirement for it [63]. Rouse, Patel and Mimi Kao provided supporting evidence from songbirds, linking vocal learning ability to flexible rhythm pattern perception [64].

And then in 2025, macaques did it. Rajendran, Luis Prado, Juan Pablo Marquez, Merchant and colleagues published a study in Science under the title Monkeys have rhythm, reporting that macaques synchronised to a subjective beat in real music and spontaneously preferred that strategy over the alternatives [65]. The authors state that this contradicts the vocal-learning hypothesis. The paper sits behind a paywall and its sample size could not be established from any open record, so no number for it appears here.

Down that list, a clean hypothesis gets complicated by animals, one at a time. That is what a healthy field looks like from outside. It is also why the current answer to "which animals feel a beat" is a spectrum rather than a line.

When the Beat Does Not Land

Not everyone gets this for free, and the exceptions have been studied properly.

Beat deafness is real and rare. Jessica Phillips-Silver and colleagues described it in 2011 in a paper titled Born to Dance but Beat Deaf, characterising it as a form of congenital amusia [66]. The case they built it around could not synchronise to a musical beat and could not reliably detect when someone else was out of time, while ordinary hearing was intact.

That last detail is the important one. The problem is not the ears.

None of this makes clumsiness a diagnosis. Rhythm ability sits on a wide continuum in the general population, and being at the clumsy end of it is not a disorder. The published cases are the far tail, identified precisely because they are so unusual.

Jakub Sowiński and Simone Dalla Bella followed this up and found that poor synchronisation can arise from faulty auditory-motor mapping rather than faulty perception [67]. Some people hear the beat perfectly well and cannot get their movements onto it. Others have trouble further upstream.

If you have ever been told you have no rhythm, the reassuring news is that true beat deafness is uncommon, and that most people who believe they have it can hear a beat fine and have simply never practised putting a movement on one.

Where This Is Already Being Used

The clinical application arrived before the mechanism argument did, which is a pattern that recurs across neuroscience.

In 1996 Michael Thaut and colleagues ran a three-week home-based gait training programme for people with Parkinson's disease, using a rhythmic auditory cue as a pacemaker [68]. Fifteen patients trained with the rhythmic cue. Eleven more were split across two control conditions, one with no training and one with self-paced training. Walking speed, stride length and cadence improved more with the rhythmic cue.

Read that design again, because the controls are the point. One group did nothing. The other trained the same amount without the rhythmic cue. If three weeks of walking practice had been doing the work, the self-paced group would have improved as much. Keep the size in view while you read it, though: eleven people split across two control conditions is five or six each, which is thin ground for any comparison.

It is a thirty-year-old trial and a small one. It is also the origin of a technique that is now widely used. Thaut, McIntosh and Rice applied the same principle to gait training after stroke the following year [69].

Why it works is still argued about, and the arguing has become more careful. Yuko Koshimori and Thaut reviewed rhythmic auditory stimulation as a possible neuromodulator in Parkinson's disease and were considerably more measured than the popular coverage [70]. Their caution is the right response to a technique that works better than anyone can currently explain.

A second question has turned out to matter as much as whether to use rhythm at all, which is what rhythm to use. Emily Ready, Jeffrey Holmes, Eryn Lonnee and Jessica Grahn found that a plain metronome is not the best cue available: music with high groove, and music the person already knows, changes the size of the gait effect [71]. Prisca Hsu, Ready and Grahn looked further upstream and found that beat perception and production themselves differ in Parkinson's disease, and that music and dance training track those abilities [72]. Clara Ziane and Simone Dalla Bella argued in 2026 that walking is the most useful test bed the field has, because gait gives you a rhythmic behaviour that matters clinically and can be measured outside a laboratory [73].

None of that is advice. It is a description of a research area, and the decision about any treatment belongs with a clinician.

There is a second application, further from settled. Usha Goswami proposed in 2011 that developmental dyslexia involves atypical sampling of the speech signal at slow rates, the temporal sampling framework [74]. Lincoln Colling, Hannah Noble and Goswami later reported differences in neural entrainment and sensorimotor synchronisation to the beat in children with dyslexia [75]. It is a serious hypothesis with real evidence behind it, and it is not the consensus explanation of dyslexia.

The Part That Is Being Sold to You

Search for anything in this article and you will meet products. Binaural beats. Isochronic tones. Focus audio engineered, so the copy says, to entrain your brain into a state of concentration.

Here is what the evidence says, without a stake in the answer.

Two things get bundled together in that pitch and they deserve separating before you evaluate either. One is a claim about perception. The other is a claim about what that perception does to the rest of your brain. The first is solid and the second is the one being sold.

Binaural beats are a real perceptual phenomenon. Play a tone at 200 hertz in one ear and 210 hertz in the other, and you hear a pulsing at 10 hertz that exists nowhere in the air. Your auditory system constructs it. That much is not in question.

The commercial claim goes further. It says that because the perceived pulse falls in the range of EEG frequency bands, your cortical activity will follow it into that band, and the mental state associated with that band will follow too.

Ruth Maria Ingendoh, Ella Posny and Angela Heine tested that claim against the literature in 2023 with a systematic review [76]. Fourteen studies met their inclusion criteria. Five reported results consistent with the brainwave entrainment hypothesis. Eight reported results that contradicted it. One was mixed.

Eight against five, with the studies themselves varying widely in how they were run. That is not a finding you can build a product page on, and it is not the same as saying binaural beats do nothing. Quiet consistent audio may well help you concentrate for reasons that have nothing to do with entraining anything. Hesham Elnazer's 2026 systematic review of music and binaural beat interventions in young adults covers the broader picture of what has been tested [77]. If you want the actual mechanics of sustained concentration, the neuroscience of deep focus is a better use of your time than choosing a frequency.

The accurate framing is narrow and easy to get wrong. The mechanism printed on the label has not been demonstrated. That is a different statement from saying the audio does nothing at all, and it is the one the evidence actually supports.

A separate line of work needs care for the opposite reason, because it is promising and gets overstated in the other direction. In 2016 Hunter Iaccarino and colleagues in Li-Huei Tsai's lab reported that driving gamma-frequency activity at 40 hertz reduced amyloid load and altered microglia [78]. That study was done in mice, which is the part that falls out of most retellings. Human work is under way and early: Tjaša Mlinarič and colleagues showed in 2025 that 40 hertz visual stimulation produces measurable oscillations in the human hippocampus [79], which establishes that the signal reaches the right structure, not that it treats anything.

Electrical stimulation is the other tool in this space. Randolph Helfrich and colleagues demonstrated in 2014 that transcranial alternating current stimulation can entrain brain oscillations [80]. That is a laboratory technique with a controlled dose, applied by researchers. It is not what you get from a phone app.

The gap between these two literatures needs naming. Careful stimulation studies, run at known frequencies with proper controls, show real effects on brain activity. Consumer audio marketed on the same vocabulary has a much weaker record. The word entrainment is what carries the credibility across the gap, and it should not.

What the Beat Actually Buys You

Step back from the argument and ask what this system is for.

The best answer available is timing. A brain that can extract a regular pulse from a messy signal can predict when the next important thing is likely to arrive, and can put its resources there instead of spreading them evenly. That applies to music, to speech, to a conversation where you need to know when it is your turn, and to any sound that arrives in a pattern.

Speech may be the case that paid for the machinery, though the animal evidence above should make anyone cautious about origin stories. Conversation runs on timing you never think about: when a syllable will land, when a turn is ending, when the person opposite has finished. Getting that wrong is socially expensive in a way that missing a downbeat is not.

Music may be the least important use of it. It is simply the one where the pattern is clearest and the pleasure is highest, which is why it is where the research happens.

There is one thing this article has deliberately not claimed. It has not told you that rhythm improves your memory or your studying. The relationship between music and remembering is a genuinely separate topic with its own evidence and its own complications, and it is covered in the piece on music and memory. Entrainment is about timing, not storage. Your biological rhythms interact with learning in other ways too, and the timing of your day turns out to matter more than the timing of your playlist.

So where does that leave the peak on the EEG?

It leaves it real. It leaves it involuntary, present in a two-day-old, driven by parts of your brain that plan movements you are not making. It leaves it useful in the clinic, on evidence that is thirty years old and still thinner than the headlines suggest.

And it leaves the mechanism open. Something in your head is keeping time. Whether it is a clock that got captured, or a series of reactions that only looks like a clock, is a question that a very good 2025 experiment failed to settle, and the researchers involved would be the first to say so.

That is an unsatisfying answer and it is the correct one. A field that could tell you the mechanism in a sentence would not still be publishing careful negative results about it in 2026.

Next time your head starts moving before you notice the music, that is the thing doing it. Nobody knows exactly what it is yet.