You think exercise is something you do for your body — the body just hands the benefit over to your brain.
Climb six flights of stairs and unlock the door out of breath, or walk briskly for forty minutes and get home, and for a little while your head is unusually clear — not a vague sense of being "energized," but the kind of clarity where you actually sit down and finish the thing you've been putting off for three days. That window isn't imaginary, and it isn't just "more blood up there" either. Every time a muscle contracts it releases a batch of molecules into the blood; several of them reach the brain and turn up the dial on the brain's ability to change itself. Exercise acting on the brain isn't a metaphor — there's a messenger, a recipient, and a downstream reaction.
Muscle used to be treated as a pure engine. We now know it is also a secretory organ: every contraction releases small molecules into the bloodstream (collectively, myokines). Several of them cross into the brain, or hand their signal off at the brain's border.
The most important downstream reaction on the receiving end is called BDNF (brain-derived neurotrophic factor). In plain terms it's fertilizer for the brain: it keeps neurons alive longer, makes synapses (the junction between two neurons) sturdier, and — most importantly — makes the act of strengthening a connection easier in the first place (synaptic plasticity). The founding 1995 study found that giving rats a running wheel visibly raised BDNF in the hippocampus.
In humans you can't open the skull, so you measure BDNF in blood: a single aerobic session does raise it. But how much "higher in blood" really means "higher in brain" is an inference, not a direct measurement — one link in this chain is not yet nailed down, and that's worth remembering.
What makes BDNF interesting is that it teaches no content at all; it only tunes how easy it is to be changed. AI training has an almost exact counterpart: the learning rate — a dial that decides nothing about what is learned, only how big a step each update takes. Too small and nothing moves; too large and what you already knew gets shaken apart. Exercise turns that dial up for a while, opening a window in which the brain is easier to rewrite. What gets put through that window still depends on you actually learning and practising during it. More plasticity ≠ smarter — it only lowers the cost of learning.
"Adult brain cells only decline, never increase" — you've probably heard this. It roughly holds for most of the brain, with one famous exception: a small region of the hippocampus called the dentate gyrus. In rodents the case is closed: give a rat a running wheel and newborn cells in the dentate gyrus clearly increase, alongside better learning and a stronger capacity for synapses to strengthen.
In humans it's messier. This is a fight still going on:
In 2018 two top-journal papers landed in the same year with opposite conclusions: one found almost no newborn neurons after adulthood, the other found them still present at seventy-nine. Much of the difference came down to method — how long the tissue sat between death and fixation determines how unreliable the markers used to spot new cells become. A 2025 study took a different route, combining single-cell gene sequencing with spatial mapping, and found actively dividing neural progenitor cells in the adult human hippocampus. It also found something important: some people have many, others almost none, and nobody yet knows why.
So the honest statement is: newborn neurons probably do exist in humans, but nowhere near as abundantly as in rodents, and they vary from person to person. Their main proposed job is pattern separation — storing two very similar experiences (yesterday's parking spot and today's) separately so they don't blur into each other.
Volume needs the same honesty. One much-cited trial had older adults do a year of aerobic exercise and found the anterior hippocampus about 2% larger — roughly rolling back one or two years of shrinkage. But pooling all the controlled trials gives a far more modest picture: no significant overall increase, more like "held the line against shrinking" than "grew it back." For an organ that shrinks year on year, holding the line is already a real achievement — it just isn't the same claim as "exercise makes your brain bigger."
The problem newborn neurons are thought to solve has a name in AI: catastrophic forgetting. A network trained on a new task tends to paint straight over what it learned on the old one — it only has so many parameters, and new knowledge competes with old for the same space. One family of fixes is to add fresh units to the network, so new material moves into new space instead of overwriting the old. Pattern separation via newborn cells is the biological version of exactly that idea: rather than rewriting an old trace, hand the new experience a batch of brand-new, unclaimed cells. Both are answering one question — how does a system with finite capacity keep learning without erasing itself?
If you only trust one conclusion here, make it this one: the evidence that exercise improves executive function is considerably stronger than the evidence for memory.
Executive function is the "management layer" run by the prefrontal cortex — braking (wanting to check your phone and not doing it), switching (moving from one task to another without stalling), and holding things in mind (keeping the three points you're about to make). None of it is as showy as memorising vocabulary, but you use it in most of the moments that matter in a day.
It works on two timescales. Acutely: in the half-hour to hour after a single moderate session, tasks that require braking and focus improve slightly — that's the measurable version of the "clear head." Chronically: months of training bring more durable gains, and the effect is largest in children and older adults. Healthy young adults are already near the ceiling, so there's much less room to move. That isn't exercise failing; that's the test failing to detect it.
Aerobic or resistance? Don't pick sides. Aerobic has the thickest evidence — it improves cardiorespiratory fitness, cerebral blood flow and vasculature at once, and the BDNF line was mostly worked out on aerobic exercise. But resistance training has independently been shown in randomised trials to improve executive function in older adults, probably via a different (IGF-1-related) route. What actually separates outcomes is never the modality; it's whether you keep doing it. An exercise you'll sustain beats a theoretically optimal one you quit in three weeks.
"Exercise is good for your mood" is a true platitude. The interesting part is one layer down: what exactly is it moving?
First, the evidence, stated fairly. A 2024 synthesis pooling a large number of randomised trials found that walking or jogging, yoga and strength training all reduce depressive symptoms, with more intense exercise doing more. But there's an unavoidable ceiling: exercise trials can't be double-blinded — you can't keep someone from knowing they're running. Participants' expectations and researchers' expectations are baked in, so the true effect is probably smaller than the published numbers. That doesn't make it useless; it means don't treat it as a settled dose that replaces treatment.
On mechanism there are several candidates, and the most counterintuitive runs like this. Chronic stress pushes an amino acid down a metabolic route toward kynurenine; kynurenine crosses the blood-brain barrier (the checkpoint around the brain that lets only certain molecules through), and its accumulation there is associated with depression. Trained muscle expresses extra enzymes that convert kynurenine, out in the bloodstream, into a different form — one that cannot cross the barrier.
In other words, muscle intercepts a stress metabolite out in the body on the brain's behalf — the benefit isn't "exercise made the brain secrete something happy," it's "the legs are taking out the brain's rubbish." This full causal chain was demonstrated in mice; in humans there's only correlational evidence so far. But it gives "the antidepressant organ might be your legs" a story specified down to the molecule.
While we're here, one very widespread claim needs correcting. "The joy of running is endorphins" — probably wrong. Endorphin molecules are too large to cross the blood-brain barrier, so what you measure in blood says little about what's happening in the brain. In experiments, blocking mice's endocannabinoid receptors (the receptors cannabis acts on — the body makes similar molecules of its own) abolished the anxiety-reducing effect of running, while blocking opioid receptors (the endorphin route) did not. More likely it's cannabinoids you made yourself, not endorphins.
Two more mechanisms are in play: exercise makes the stress-hormone system less easily maxed out, and it lowers chronic low-grade inflammation throughout the body — and the link between inflammation and depression is one of the more solid findings of the past fifteen years. Finally, regular exercise weakens the kind of rumination that won't stop (going over "why am I so terrible" again and again), which corresponds to the default mode network idling in overdrive.
"Moving the body is the shortest route to changing the mind" — several traditions brushed up against this, and one of them was looked down on by its own school: