Damaged plants release volatile chemicals, and neighbours that detect them mount defences sooner. The chemistry is real; the framing usually is not — the best explanation is that plants are signalling to their own distant leaves, and neighbours are eavesdropping.
A plant attacked by insects releases a blend of volatile compounds, and a neighbouring plant exposed to that blend defends itself better when its own turn comes. This has been shown outdoors, at natural spacing, with damage measured over a whole season — it is not a chamber artefact, though it started as one and was dismissed for over a decade on those grounds. What has changed most is not whether it happens but why. Blocking airflow between leaves of a *single* plant abolishes the effect within that plant, which suggests volatiles evolved so that one part of a plant can signal another part it is poorly plumbed to reach. If so, warning the neighbours is not altruism requiring an evolutionary explanation; it is a leak that neighbours exploit. That reading is deflationary and much cleaner, and it is why NatureHQ treats "plants talk to each other" as the wrong summary of a real phenomenon.
Developed record · 98% complete · reviewed 2026-08-11
What this page covers
A phenomenon rather than an organism. Covers volatile and below-ground signalling wherever it has been directly tested — sagebrush, wild tobacco, lima bean, broad bean and several trees.
Quick facts
Airborne signalling
Demonstrated in the field, over a full season
Below-ground route
Signals also travel through shared mycorrhizal networks
A damaged plant releases chemicals into the air, and nearby plants prepare their defences
Well supported
Good evidence backs this, though some details remain open.
Herbivore or mechanical damage causes plants to emit volatile organic compounds. Undamaged plants exposed to these volatiles under field conditions up-regulate defensive chemistry or defensive services and sustain measurably less subsequent herbivore damage, with the effect travelling through air rather than through soil or root contact.
Who this applies to
demonstrated in several wild systems, principally sagebrush–tobacco and lima bean
“Plants talk to each other and warn their friends”
The chemistry is real and the framing is not. Talking implies a sender addressing a receiver for the sender’s benefit; what is documented is a plant releasing compounds — quite possibly to signal to its own distant leaves — and neighbours picking them up. Overhearing is not conversation, and the best current explanation for why plants emit at all has nothing to do with warning anyone.
Why we rate it this way, and what the caveats are
Well supportedHigh confidence
Field experiments at natural spacing with damage measured over a season, separating the airborne route from the soil route. The phenomenon began as a contested laboratory result and was established by better-designed outdoor work.
How far it can be extended
Volatile-induced resistance has been shown independently in unrelated plant families and in the field as well as the laboratory, though it is not universal and effect sizes vary widely.
Caveats
Effect sizes vary greatly between systems, and negative results are under-reported.
The neighbour is eavesdropping on a cue, not being addressed — no plant is plausibly signalling to another species for that species’ benefit.
Within-plant signalling may be the evolved function, with between-plant effects incidental.
Where researchers disagree
Whether between-plant volatile signalling is adaptive for the emitter, or simply a leak that neighbours exploit, is unresolved; the within-plant signalling evidence favours the latter.
Still unanswered
Which specific compounds carry the effect, and does the blend encode the type of attacker?
Is there any circumstance in which emitting is advantageous to the emitter?
A parallel below-ground route: signals travelling through a shared mycorrhizal network rather than through air.
How we know
Clipping sagebrush to see if the neighbours notice
Plant-to-plant signalling had been shown in sealed laboratory chambers and dismissed as an artefact. Does it happen outdoors, at natural spacing, in a real season?
Sagebrush growing wild in the Great Basin was experimentally clipped to simulate herbivore damage. Wild tobacco plants growing beside the clipped and unclipped sagebrush were then monitored for the rest of the season, and the herbivore damage they sustained was measured. Crucially, air contact and root contact were manipulated separately, so the route any effect travelled could be identified rather than assumed.
What happened
Tobacco growing next to clipped sagebrush sustained significantly less herbivore damage across the season, and the effect travelled through the air rather than through soil or roots.
What it shows
Volatile signalling between plants is a real field phenomenon and not a chamber artefact. An initially weak study proposing a real thing is a different animal from a wrong one, and this is what rescued the idea.
What it does not show
It does not show that sagebrush is warning tobacco, or benefiting from doing so — these are unrelated species, and the receiver is best described as eavesdropping. Later work suggesting volatiles evolved for signalling within a single plant makes the neighbour effect look like a leak rather than a message.
The controls — what makes this evidence rather than a story
Tobacco beside unclipped sagebrush gave the baseline — the plants differed only in whether their neighbour had been damaged.
Separating airborne from below-ground contact identified the route, which the earlier chamber studies could not do.
Damage was measured over a full season in the field, rather than chemistry measured in a box over days.
The field design is what makes this trustworthy. Sagebrush growing naturally in the Great Basin was clipped; wild tobacco growing beside it sustained measurably less insect damage over the following season than tobacco beside unclipped sagebrush. Air contact was manipulated separately from root contact, so the route is identified rather than assumed.
Note that these are two different species. Sagebrush is not plausibly signalling to tobacco for sagebrush’s benefit, which is the first hint that "communication" is the wrong frame — and the word researchers in the field use for the receiver is eavesdropping.
There is a second route entirely. Broad beans connected by a shared mycorrhizal network mount aphid defences when a connected neighbour is attacked, and severing the fungal connection abolishes it. Air is not the only channel.
Words used here
Volatile organic compound
A carbon-based chemical that evaporates readily at ordinary temperatures, so it can travel through air.
Priming
Being made ready to respond faster, rather than responding now. A primed plant defends more quickly when attacked.
The result that reframes the whole field, and is rarely reported.
How we know
Blocking the air between two leaves on the same plant
When a damaged plant defends its undamaged parts, does the message travel inside the plant — or outside it?
Wild lima bean plants were damaged on one leaf, and the response of an undamaged leaf on the same plant was measured by how much extrafloral nectar it produced, which in turn recruits defensive ants. The manipulation was simple and decisive: for some plants, airflow between the two leaves was physically blocked, while the internal vascular connection between them was left completely intact.
What happened
Undamaged leaves downwind of damaged ones increased nectar secretion and attracted more ants. Blocking the airflow abolished the effect — even though the two leaves were still fully connected inside the plant.
What it shows
Volatiles carry a signal *within* a single plant, reaching parts its own plumbing serves poorly. That reframes the whole field: if this is what volatile signalling is for, then neighbouring plants benefiting is a by-product of a broadcast rather than an act of warning requiring an evolutionary explanation.
What it does not show
One species with an unusually convenient measurable defence. Whether within-plant signalling is the general explanation for volatile communication across plants is not established, and blocking airflow may alter humidity and temperature as well as volatile transport.
The controls — what makes this evidence rather than a story
The vascular connection remained intact in every condition, so any difference could not be internal transport being severed.
Undamaged plants gave the baseline nectar production.
Ant recruitment was counted as well as nectar volume, tying the chemistry to a consequence.
Lima bean leaves are connected to each other by vascular tissue, but poorly — a signal travelling internally from one leaf may not reach another leaf a few centimetres away on a different branch. Blocking airflow between leaves of the same plant abolished the defensive response in the undamaged leaf, even though the vascular connection was intact.
So the volatile is doing a job *within* the individual: getting a message to a part of itself that its plumbing cannot reach. Every evolutionary difficulty about why a plant would warn competitors dissolves, because on this account it is not warning anyone. It is shouting across a gap in its own body, outdoors, where others can hear.
The word used in this literature for a neighbouring plant that benefits is "eavesdropper" — a receiver the signal was not sent to.
Sagebrush responds more strongly to chemical cues from its own relatives
Emerging evidence
Real findings exist, but too few or too recent to be settled.
Sagebrush plants receiving volatile cues from genetically related donors sustain less herbivore damage than those receiving cues from unrelated donors. Volatile blends are heritable, so a plant most responsive to blends resembling its own will respond preferentially to kin without any recognition of another individual.
Who this applies to
sagebrush in the Great Basin
Studied in
Artemisia tridentata
Why we rate it this way, and what the caveats are
Emerging evidenceModerate confidence
A well-designed field experiment with a clear result, but a single species, and kin similarity is confounded with blend similarity by design — which is also the proposed mechanism, making the effect and its explanation hard to separate.
Caveats
Self-matching to a familiar chemical profile is not recognition of another individual.
One species; heritability of volatile blends elsewhere is largely unknown.
Field damage is a noisy endpoint with many other influences.
Still unanswered
Is kin-biased response found in any other species?
Does the effect require the receiver to have any representation of its own blend?
Karban et al., 2013 · Proceedings of the Royal Society B: Biological Sciences
Field exposure to related versus unrelated donors, with damage measured over a season.
The kin-recognition result has the same deflationary shape. If volatile blends are heritable, and a plant responds most strongly to a blend resembling its own, kin-biased response follows automatically with nothing recognising anything. Self-matching, not recognition.
Words used here
Vascular connection
The internal plumbing — xylem and phloem — that carries water, sugars and some signals between parts of a plant.
Four words that get used interchangeably, and are not interchangeable at all.
Almost every argument about plant communication is really an argument about vocabulary, and it can be settled by keeping four things apart. They are not synonyms and they make different claims about the world.
The four things "plants communicate" can mean
Term
What it requires
Example in plants
Signal
A trait that exists *because* of its effect on a receiver — which means the sender benefits from being received
Contested. Nectar and floral scent qualify; whether any plant-to-plant volatile does is unresolved
Cue
Information a receiver can use, leaking out whether or not the source gains anything
Volatiles from a damaged neighbour; ultrasound from a stem under drought stress
Physiological signalling
Information moving within one organism's own tissues
The calcium and electrical wave from a wounded leaf to the rest of the plant
Communication between organisms
A sender, a receiver, and a benefit to the sender from the receiving
Requires evidence about the emitter's fitness, which is very rarely available
The reason this is not pedantry: the chemistry underlying "talking plants" is not in dispute, and neither is the effect on neighbours. What is in dispute is whether anything is being *sent*. A signal has to have been shaped by selection because of what it does to a receiver — which requires the sender to gain something. On the present evidence a plant is signalling to itself, in the open, and the neighbours are overhearing. The word researchers in this field use for the beneficiary is eavesdropper.
Two habits follow, and they resolve most of the confusion on this page. Ask who benefits from the sending, because if nobody does then communication is the wrong word however real the chemistry. And treat memory, learning and intelligence as claims that need their own evidence, rather than as descriptions of any behaviour that responds to something.
Words used here
Signal
A trait or action that evolved because of its effect on a receiver. Requires a benefit to the sender.
Cue
Information a receiver can use, which was not shaped for that purpose. A footprint is a cue; a shout is a signal.
Eavesdropping
Benefiting from information that was not addressed to you. The standard term for what a neighbouring plant is doing.
A plant has no nerves and still gets a message from one end to the other in seconds.
Wound one leaf and a signal travels through the plant to the others
Established
Specialists would state this without hedging. Multiple independent lines of evidence agree.
Wounding or herbivory on one leaf of Arabidopsis triggers a wave of elevated cytosolic calcium that travels through the vasculature to distant leaves at approximately one millimetre per second, arriving within seconds to minutes and followed by jasmonate-dependent defence gene expression in the leaves reached. The wave requires glutamate receptor-like channels, and applying glutamate directly to a wound is sufficient to trigger it. A propagating change in membrane potential accompanies the calcium wave and is likewise abolished when the same channels are disabled.
Who this applies to
demonstrated in Arabidopsis; long-distance wound signalling is general to vascular plants
Studied in
Arabidopsis thaliana, Plantae
You may have heard
“Plants have a nervous system and use the same chemicals our neurons do”
The molecular coincidence is real and the conclusion does not follow. Glutamate is a neurotransmitter in animals, and the channels carrying the plant signal are relatives of animal glutamate receptors — which is genuinely striking and is why the coverage went the way it did. But there are no neurons, no synapses, no brain and no network doing anything with the signal. The wave travels at a millimetre a second, against tens of metres a second in a nerve, through plumbing rather than dedicated cells. An old molecular component reused as a wire in a body with no nervous system is a stranger fact than the headline, not a smaller one.
Why we rate it this way, and what the caveats are
EstablishedHigh confidence
Two independent laboratories, using imaging and electrophysiology respectively, converged on the same channel family, with necessity shown by mutants and sufficiency by direct glutamate application.
How far it can be extended
Systemic wound signalling and electrical signalling are documented across vascular plants; the molecular components are characterised chiefly in one model species.
Caveats
One model species; signalling parameters differ substantially between plant groups.
Calcium, electrical and hydraulic signals travel at similar speeds and are tightly coupled, so isolating one causally is difficult.
Fluorescent reporters can perturb the calcium handling they are used to measure.
Still unanswered
Does the signal encode anything about what caused the wound, or only that a wound occurred?
The accompanying electrical signal, and loss of both signal and distant response when the channels are disabled.
How we know
Watching a signal cross a plant in real time
How does one leaf tell the rest of the plant that something is eating it?
That a wounded plant mounts defences elsewhere had been known for decades; what carried the message was not. Speed was the clue — the response arrives far too quickly for a hormone to be transported the distance. Arabidopsis plants were engineered to express a protein that fluoresces in proportion to calcium concentration, turning an invisible chemical event into something a camera can film. A leaf was then wounded, or given to a caterpillar, and the whole plant filmed. Two further manipulations turned an observation into a mechanism: mutants lacking glutamate receptor-like proteins were run alongside to test whether the wave still occurred, and glutamate was applied directly to a wound to test whether it was enough on its own.
What happened
Wounding one leaf sends a wave of elevated calcium travelling through the vasculature to distant leaves at about a millimetre per second, arriving within seconds to a couple of minutes and followed by defence gene expression where it lands. The wave requires the glutamate receptor-like channels, and glutamate alone triggers it.
What it shows
A long-distance signalling system in a plant, filmed from end to end, with the molecular component identified by both necessity and sufficiency. Glutamate leaking from broken cells is the alarm, and receptor-like channels propagate it.
What it does not show
It does not show anything resembling a nervous system, whatever the shared molecular vocabulary suggests. There are no neurons and no network computing on the signal — it is a wire in plumbing. Calcium, electrical and hydraulic signals also travel together at similar speeds, so which is doing the causal work is genuinely hard to separate. And it is one model species in a laboratory.
The controls — what makes this evidence rather than a story
Unwounded plants filmed identically, so the wave could not be an artefact of handling or of the imaging.
Mutants lacking the candidate channels, testing necessity.
Glutamate applied to a wound without further damage, testing sufficiency.
Real herbivory by caterpillars alongside mechanical wounding, so the result is not an artefact of scissors.
Defence gene expression measured in the distant leaves, so the wave is connected to an outcome rather than only observed.
That a wounded plant defends itself elsewhere had been known for decades. What carried the message was not, and the clue was speed: the response arrives far too quickly for a hormone to be transported the distance. The answer came from making the invisible filmable. Plants engineered to fluoresce in proportion to calcium concentration were wounded, or given to a caterpillar, and the whole plant filmed.
A wave of elevated calcium travels through the vasculature to distant leaves at roughly a millimetre per second, arriving within seconds to a couple of minutes, and defence genes switch on in the leaves it reaches. Disable the glutamate receptor-like channels and the wave does not happen; apply glutamate to a wound and it happens without any wounding. Necessity and sufficiency, in the same paper.
Glutamate is a neurotransmitter in animals and the channels are relatives of animal glutamate receptors, which is why this was reported as plants having nerves. They do not. There are no neurons, no synapses, no brain, and nothing performing computation on the signal. The wave moves at a millimetre a second against tens of metres a second in a nerve, through plumbing rather than dedicated cells. An ancient molecular component reused as a wire in a body with no nervous system is a stranger fact than the headline, not a smaller one.
Electrical — propagating changes in membrane potential, travelling through vascular tissue, abolished when the same channels are disabled
Calcium — a wave of elevated cytosolic calcium, directly imaged, triggering defence gene expression where it lands
Hydraulic — pressure changes moving through the water column, fast and hard to separate from the others
Chemical — jasmonates and related hormones, slower, doing the sustained work once the fast signals have arrived
Those four travel together and at similar speeds, which is the main reason assigning causation to any one of them is difficult and why the field argues about it. What is not in doubt is that a plant moves information around its own body quickly, and that this — rather than anything between plants — is the best-evidenced signalling that plants do.
A warned plant does not defend itself — it gets ready to, cheaply
Established
Specialists would state this without hedging. Multiple independent lines of evidence agree.
Defence priming leaves a plant able to respond faster and more strongly to a subsequent attack without mounting the full response in advance. It operates through accumulation of inactive signalling components and through chromatin modifications at defence gene loci, so the primed state carries a low metabolic cost. Primed states persist for weeks and in some systems are transmitted to offspring. Priming carries a fitness cost when attack does not follow, and deliberate priming of crops is an established agricultural technology.
Who this applies to
vascular plants; mechanisms characterised chiefly in model and crop species
Studied in
Plantae, Arabidopsis thaliana, Zea mays
You may have heard
“Plants warn each other so their neighbours can defend themselves”
The measured effect is a step short of that and better for it. An exposed plant does not put up defences — it stocks the shelves. Inactive signalling components accumulate and the defence genes are marked for faster expression, at low cost, and nothing visible happens until something actually attacks. Then the response arrives sooner and harder. That is why the standard experiment finds no difference between exposed and unexposed plants until the caterpillars turn up, and it is a much cheaper strategy than defending against an attack that may never come.
Why we rate it this way, and what the caveats are
EstablishedHigh confidence
Priming is demonstrated across many species and attacker types, has an identified mechanistic basis, and is applied commercially — which is a demanding test of a laboratory effect.
How far it can be extended
Priming has been demonstrated across a wide range of plant families and against diverse attackers, though mechanistic detail comes from a few species.
Caveats
Laboratory effects have often been larger than field effects.
Transgenerational priming rests on a smaller and more contested evidence base than within-generation priming.
Priming costs are demonstrated in some systems and assumed in others.
Still unanswered
How long does a primed state persist under field conditions, and what resets it?
Heil and Karban, 2010 · Trends in Ecology & Evolution
Places priming as the actual measured outcome of volatile exposure, in place of an immediate defensive response.
What the receiving tissue then does is quieter than expected, and better. It does not mount a defence — that would be paying for an attack that may never come. It gets ready: inactive signalling components accumulate and defence genes are marked for faster expression, at low metabolic cost. Nothing visible happens until something actually bites, and then the response arrives sooner and harder. That is why the standard experiment shows no difference between exposed and unexposed plants right up until the caterpillars arrive, and it is now a commercial agricultural technique rather than a curiosity.
Words used here
Action potential
A self-propagating change in membrane voltage. Plants have them; they are slow, travel through vascular tissue, and connect to no network.
Priming
Being made ready to respond faster rather than responding now. The cheap option, and what plant signalling actually produces.
Jasmonate
A plant hormone family central to defence against chewing insects. The slower chemical layer beneath the fast electrical and calcium signals.
Plants spend a serious part of their income on chemicals pushed into the soil
Well supported
Good evidence backs this, though some details remain open.
Plant roots release a substantial fraction of photosynthetically fixed carbon into the surrounding soil as a chemically diverse exudate. Specific compounds attract or repel particular microbes, initiate symbiotic associations including nodulation and mycorrhizal colonisation, and in some species inhibit the germination or growth of neighbouring plants. Exudate composition varies with the plant's physiological state and with the organisms present. Fungal networks are one channel among several rather than the whole of below-ground interaction.
Who this applies to
vascular plants and their soil communities
Studied in
Plantae, Bacteria, Fungi
Why we rate it this way, and what the caveats are
Well supportedModerate confidence
Exudation and its role in establishing symbioses are firmly established. Claims about chemical suppression of neighbours are much weaker, having frequently failed to reproduce in field conditions.
How far it can be extended
Exudation is universal among rooted plants; the specific compounds and their effects are characterised in a minority of species.
Caveats
Much of the underlying work uses hydroponic or sterile systems that differ greatly from soil.
Allelopathic claims in particular have often not survived field testing.
The proportion of fixed carbon exuded varies widely with species, age and conditions.
Still unanswered
Which exudate effects are signals shaped by their effect on a receiver, and which are metabolic leakage that other organisms exploit?
Heil and Karban, 2010 · Trends in Ecology & Evolution
Supplies the signal-versus-cue framework, which separates recruiting a partner from poisoning a competitor.
The below-ground story is usually told entirely through fungal networks, and that is one channel out of several. A plant releases a substantial fraction of everything it fixes by photosynthesis straight into the soil around its roots, as a chemically diverse exudate whose composition changes with the plant's condition and with what is living nearby. That is a serious investment of income, and it is regulated rather than leaked.
Some of it recruits: specific compounds attract particular bacteria, initiate nodulation with nitrogen-fixing partners, and signal to mycorrhizal fungi to begin colonisation. Some of it suppresses, inhibiting the germination or growth of neighbouring plants. Those two are worth separating for the same reason as everything else on this page — recruiting a partner that benefits from responding looks a great deal more like signalling than poisoning a competitor does.
The honest caveat is that much of this literature comes from hydroponic or sterile systems that are nothing like soil, and the suppression claims in particular have often failed to reproduce in the field. Soil is crowded, chemically complex and full of microbes that eat the compounds in question.
Yes. No human has ever heard one, and nothing is known to be listening.
Plants do click under stress, in a range no human can hear, and nothing is known to listen
Emerging evidence
Real findings exist, but too few or too recent to be settled.
Tomato and tobacco plants subjected to drought or cutting emit airborne ultrasonic clicks in the 20–100 kilohertz band at rates of tens per hour, against near-silence from unstressed controls, detectable at a range of several metres. A classifier trained on these recordings distinguishes drought-stressed from cut from control plants above chance, including under greenhouse conditions. The mechanism is understood as cavitation — water columns under tension breaking within the xylem. No receiver has been shown to use the emissions, and no evidence indicates the sounds are produced for a receiver.
Who this applies to
demonstrated in two crop species under controlled conditions
Studied in
Solanum lycopersicum, Nicotiana tabacum
You may have heard
“Plants scream when you cut them”
Everything in that sentence except the fact that a sound occurs. It is ultrasonic, so no human has ever heard one. It happens at a rate of a few dozen an hour, which is closer to a dripping tap than to a scream. The mechanism is water columns snapping under tension inside the stem — the same physics as a cracking knuckle — and nothing suggests it is produced in order to be heard. The paper's own word is "informative", which means a listener could in principle learn something from it. That is the definition of a cue. Whether anything actually listens is being investigated and is not yet answered.
Why we rate it this way, and what the caveats are
Emerging evidenceModerate confidence
The emissions themselves are measured cleanly and the classifier result is solid. Everything about biological function is open: no receiver has been tested, and whether anything uses the information is unknown.
Caveats
Two crop species under controlled conditions; the greenhouse extension rests on a smaller dataset.
Classifier performance above chance is not evidence that any organism can use the information.
Nothing indicates the emissions are produced for a receiver, which is what would make them a signal rather than a cue.
Still unanswered
Do moths, or any other organism with ultrasonic hearing, actually use these emissions when choosing where to lay or feed?
Do the emissions occur at comparable rates in the field, where wind and other noise sources are present?
Heil and Karban, 2010 · Trends in Ecology & Evolution
Supplies the signal-versus-cue distinction that determines how emissions with no demonstrated receiver should be described.
How we know
Listening to plants at frequencies nobody can hear
Do stressed plants make sounds, and does the sound say anything about the stress?
Plants had been known to produce ultrasound *within* their tissue when water columns snap under tension — but a sound in the stem is not a sound in the air, and nobody had shown any of it escaping into the room. Tomato and tobacco plants were placed in an acoustically isolated box with ultrasonic microphones and subjected to drought, to cutting, or to nothing at all. Because the raw recordings are sparse clicks that no human can hear or classify, a machine-learning classifier was trained to distinguish recordings by treatment — and then the whole thing was repeated in a greenhouse, where the sounds of pumps, fans and the outside world could all interfere.
What happened
Stressed plants emit airborne ultrasonic clicks between 20 and 100 kilohertz at rates of tens per hour, against near-silence from controls. The classifier could tell drought-stressed from cut from control plants above chance, and could still do so in the greenhouse. The emissions are detectable several metres away by an animal with the right ears.
What it shows
That plants emit informative airborne sound. "Informative" is precise and load-bearing: the sound correlates with the plant's state closely enough that a listener could in principle learn something from it — which is the definition of a cue.
What it does not show
It does not show communication, and the paper does not claim it. Nothing indicates the sound is produced for a receiver, which is what would make it a signal; the mechanism is cavitation, a physical consequence of water under tension, not an emission system. No receiver was tested at all, so whether any animal uses these sounds is unknown and under investigation. And nothing screams: this is a few dozen ultrasonic clicks an hour.
The controls — what makes this evidence rather than a story
Untouched plants recorded under identical conditions, establishing the near-silent baseline.
Empty pots and the chamber itself recorded, so the apparatus could be excluded as the source.
Two stress types and two species, so the classifier had to separate more than presence from absence.
The greenhouse repeat, which tests whether anything survives outside an anechoic box.
Classifier performance judged against chance on held-out recordings rather than on the data it was trained on.
Plants were known to produce ultrasound inside their tissue when water columns snap under tension. A sound in the stem is not a sound in the room, though, and in 2023 somebody put ultrasonic microphones around tomato and tobacco plants in an acoustically isolated box and found the emissions escaping into the air — dozens of clicks an hour from drought-stressed and cut plants, against near-silence from untouched ones, detectable several metres away.
More than that, the sounds differ by cause. A classifier trained on the recordings could separate drought-stressed from cut from control above chance, and could still do it in a greenhouse full of pumps and fans. So the emissions carry information about the plant's state.
Now the part the headlines dropped. The frequencies are 20 to 100 kilohertz — no human has ever heard one and never will. The rate is a few dozen clicks an hour, which is a dripping tap rather than a scream. The mechanism is cavitation: water columns under tension breaking, the same physics as a cracking knuckle, not an emission system. And nothing indicates the sound is produced *for* a receiver, which is what would make it a signal rather than a cue.
The word in the paper's own title is "informative", and it is precise. It means a listener could in principle learn something. That is the definition of a cue, and it is a deliberately weaker claim than communication.
Whether anything actually listens is the open question, and it is a good one. Moths choosing where to lay, and herbivores choosing which plant to eat, would both benefit from knowing which plant is struggling. That work is under way and is not settled, which is exactly the state to report it in.
Words used here
Cavitation
A water column under tension breaking and forming a bubble. It makes an audible click, and it is what "screaming plants" reporting describes.
Ultrasound
Sound above roughly 20 kilohertz, beyond human hearing. Where all of the plant emissions sit.
Plants detect damage and respond to it; nothing indicates they feel it
Well supported
Good evidence backs this, though some details remain open.
Plants detect tissue damage and mount rapid local and systemic responses via calcium, electrical and hormonal signalling. They possess no nociceptors, no nervous system, no centralised integrative structure and no anatomical analogue of the neural architectures associated with pain or with conscious experience in animals. The historical claim that plants respond electrically to harm inflicted on nearby organisms did not survive controlled replication in which the timing of the harm was randomised and the analysis blinded.
Who this applies to
plants generally
Studied in
Plantae
You may have heard
“Plants feel pain when you cut them”
They detect the damage — quickly, specifically, and with a signal that reaches the rest of the plant within seconds. Detecting is not feeling. Pain in animals depends on nociceptors reporting to a nervous system that integrates the signal, and plants have neither: no pain receptors, no neurons, no brain, nothing that could be the place where it hurts. The founding evidence for the stronger claim was a polygraph examiner's houseplant in 1966, and when someone repeated it properly — randomising when the harm happened and analysing the traces blind — the effect vanished. That result is fifty years old and the claim has outlived it comfortably.
Why we rate it this way, and what the caveats are
Well supportedHigh confidence
The anatomical facts are not disputed by anyone, including researchers who argue for plant cognition. High confidence about mechanism; the philosophical question of what experience requires cannot be settled empirically, and this claim is careful to address only the mechanism.
How far it can be extended
The absence of neurons, synapses and centralised integrative organs is a general feature of plants, not a finding about particular species.
Caveats
Absence of a known mechanism is strong evidence and is not logical proof; this claim is about mechanism rather than metaphysics.
Nothing here diminishes the sophistication of plant damage responses, which are fast, specific and systemic.
The word "pain" carries a meaning about experience that "damage detection" does not, and the distinction is the entire claim.
Still unanswered
What would count as evidence of experience in an organism with no nervous system, and is the question answerable at all?
Shows what damage detection in a plant actually consists of — a calcium wave and gene expression — which is the mechanism the popular claim reinterprets.
A plant detects damage. It does so quickly, specifically, and with a signal that reaches the rest of the plant within seconds — all of which is on this page already. Detecting is not feeling. Pain in animals depends on nociceptors reporting into a nervous system that integrates the signal, and plants have neither: no pain receptors, no neurons, no synapses, and nothing that could be the place where it hurts.
How we know
The houseplant, the polygraph and the brine shrimp
Do plants react electrically when something is killed nearby?
The claim came from a polygraph examiner who attached his equipment to a houseplant in 1966, reported that it responded when he merely intended to burn a leaf, and became the origin of most "plants sense your feelings" material ever since. The design that tested it is a small lesson in what the original lacked. Brine shrimp were killed by an automatic mechanism at times chosen at random, so that no human decided when anything happened and no human was present to react. The plants were electrically shielded, since a person moving near an unshielded electrode produces exactly the sort of trace being claimed. And the recordings were analysed without the analyst knowing when the killings had occurred, so that "responses" could not be found by looking where they were expected.
What happened
Nothing. There was no relationship between the killings and the plants' electrical activity. Traces resembling the reported responses appeared throughout the records regardless of whether anything had been killed — which is the point: the original phenomenon was ordinary electrical noise, read by someone who knew when to look.
What it shows
That the founding claim of plant "primary perception" does not survive controlled testing, and precisely which three features it lacked: randomised timing, shielding, and blind analysis. It is a compact demonstration of why those three exist.
What it does not show
A negative result cannot exclude an effect smaller than the design could detect, and this addresses the claim as originally framed rather than every variant since. It also says nothing against the plant electrical signalling that is real — that is a wound response with an identified mechanism and known channels, a different phenomenon that happens to be measured with similar equipment.
The controls — what makes this evidence rather than a story
Killings triggered automatically at randomised times, removing the experimenter from the moment entirely.
Plants electrically shielded, removing the person as a source of the signal.
Analysis blinded to the schedule, so apparent responses could not be selected after the fact.
Long recording periods containing many non-killing intervals for comparison.
Replicated trials rather than a single striking session.
The stronger claim — that plants perceive harm done to others, or sense human intentions — has a specific origin. In 1966 a polygraph examiner attached his equipment to a houseplant, reported that it reacted when he merely thought about burning a leaf, and started a story that has never stopped circulating. In 1975 somebody tested it properly: brine shrimp killed automatically at randomised times, the plants electrically shielded, and the traces analysed by someone who did not know when the killings had happened. The effect disappeared entirely. Traces resembling the reported responses turned up throughout the records whether or not anything had been killed.
That result is half a century old and the claim has comfortably outlived it, which is worth noticing in itself. It also does not touch the plant electrical signalling that is real — that is a wound response with identified channels and a known mechanism, a different phenomenon that happens to be measured with similar equipment.
Plants habituate; whether they can learn to associate two things is disputed
Contested
Researchers actively disagree, and the disagreement is substantive.
Mimosa pudica ceases its leaf-folding response to a repeated harmless drop within a small number of trials, continues to respond to a different stimulus, and retains the reduced response for up to a month, with faster habituation under conditions where folding is more costly — the standard criteria for habituation rather than fatigue. A separate and stronger claim, that pea seedlings can be conditioned to grow towards a fan previously paired with light, was not reproduced in a pre-registered replication with a larger sample. Habituation is the simplest form of learning and occurs in organisms without nervous systems; associative learning is a materially stronger capability.
Who this applies to
habituation demonstrated in Mimosa; associative learning claimed in peas and not reproducedDo not extend this beyond the taxa listed — the popular version over-reaches.
Studied in
Mimosa pudica, Pisum sativum
You may have heard
“Plants can learn and remember”
It depends entirely which claim is meant, and the two get merged. Mimosa stops reacting to a repeated harmless disturbance and keeps reacting to a new one, which is habituation — real learning in the technical sense, and the simplest kind there is, found in single-celled organisms with nothing resembling a nervous system. The claim that made headlines was much stronger: that pea plants could be taught to associate a breeze with light, which is Pavlovian conditioning. That one failed to replicate. Reporting the first while implying the second is how "plants learn" came to mean far more than the evidence carries.
Why we rate it this way, and what the caveats are
ContestedModerate confidence
The habituation result has appropriate controls and has not been overturned. The associative-learning result failed a careful replication, and the original authors dispute the conditions. A single failed replication is evidence rather than a verdict, so the honest state is disagreement.
How far it can be extended
Neither result has been extended across plants generally, and the two point in different directions, so no group-wide generalisation is available.
Where researchers disagree
The Mimosa habituation result stands with adequate controls; the pea associative-learning result, which is the one that would demonstrate something cognitively substantial, did not replicate.
The original authors argue that growth conditions, seed stock and apparatus placement differ between laboratories in ways that matter, and that the replication therefore did not test the same thing.
Plant physiologists argue that even the habituation result requires no learning mechanism, being explicable by depletion and recovery of the tissues involved in the movement — though the novel-stimulus control was designed to address exactly that.
Still unanswered
Can associative learning in plants be demonstrated in a design where the outcome is scored automatically and the experimenter is blind to condition?
Is habituation in Mimosa a change in the signalling pathway, or a change in the mechanical capacity to fold?
Argues that the behaviours reported require no cognitive explanation, and that the terminology imports assumptions the anatomy does not support.
How we know
Trying to teach a pea plant, again
Can a plant learn that one thing predicts another?
The original experiment was the most striking claim plant behaviour research had produced: pea seedlings in a Y-shaped maze were exposed to a fan blowing from the same arm that light later came from, and were reported afterwards to grow towards the fan alone — Pavlov's dogs, in a plant. Because the result carried so much weight, and because plant growth experiments are unusually sensitive to small differences in conditions, it needed an independent repeat. The protocol was followed with a substantially larger sample and with the analysis plan registered before the data were collected, so that the ways of slicing the results were fixed in advance rather than chosen once the growth directions were known.
What happened
No evidence of conditioned growth. Seedlings did not grow towards the fan-associated arm at rates differing from chance or from the controls.
What it shows
That the strongest claim in the plant-cognition literature did not reproduce. This matters more than an ordinary null result because of what the original was claiming: associative learning requires linking two unrelated stimuli, which is a substantially harder thing than habituation and is what made the result feel like evidence of something cognitive.
What it does not show
One failed replication is evidence, not a verdict. Seed stock, growth conditions and apparatus placement differ between laboratories, and the original authors argue those differences matter. It also says nothing about habituation in Mimosa, which is a separate result with its own controls and has not been overturned. NatureHQ records the question as contested for exactly this reason.
The controls — what makes this evidence rather than a story
Pre-registered analysis plan, fixing the outcome measure before any seedling grew.
A larger sample than the original, since a null result is only informative with enough power to detect the effect claimed.
Control groups receiving fan and light unpaired, so a general effect of airflow could be separated from an association.
Growth direction scored by position rather than by impression.
Learning is where NatureHQ has to say "we do not know" and mean it. Two claims travel together and they are in very different states. Mimosa pudica stops folding its leaves in response to a repeated harmless drop, keeps responding to a different disturbance, and still fails to respond a month later — habituation, with the control that distinguishes it from simple exhaustion. Habituation is real learning in the technical sense and it is the simplest kind there is, found in organisms with no nervous system at all.
The claim that made headlines was much stronger: that pea seedlings could be conditioned to grow towards a fan that had previously predicted light — Pavlov, in a plant. Associative learning means linking two unrelated stimuli, which is a categorically harder thing than habituation, and it is what made the result feel like evidence of cognition. A pre-registered replication with a larger sample found nothing. The original authors dispute the conditions, and one failed replication is evidence rather than a verdict, so this page records the question as contested and will not tidy it in either direction.
And plants do not communicate with people. Nothing detects your presence, your mood or your intentions. Talking to a houseplant is a pleasant thing to do and there is no mechanism by which it helps; the exhaled carbon dioxide is negligible against what a room already contains. What plants do detect — light, gravity, touch, chemistry, damage, day length — is an impressive list, and none of it is you.
Practical
Reading the next "plants are conscious" headline
Three questions handle almost all of it. First, is the effect a signal or a cue — did the plant produce this because of what it does to a receiver, or is it information leaking out? Second, has a receiver actually been tested, or only shown to be capable in principle? Third, is the strong claim resting on the same evidence as the weak one — because habituation and associative learning are routinely merged, and only one of them has held up. Nothing here requires dismissing plant behaviour research, which is producing genuinely surprising results. It requires reading the claim that was made rather than the one that got reported.
Where this applies: global
Words used here
Nociceptor
A sensory receptor specialised for detecting damaging stimuli, feeding into a nervous system. Plants have none.
Habituation
Learning to stop responding to a repeated harmless stimulus. The simplest form of learning, present in organisms without nervous systems.
Associative learning
Learning that one stimulus predicts another. A substantially stronger capability than habituation, and the claim that failed to replicate in plants.
There is no good evidence that plants talk to each other using sound
Not enough evidence
Nobody has done the work needed to answer this properly.
Stressed plants emit ultrasonic acoustic energy, which is well explained by cavitation in water columns under tension. Claims that plants respond adaptively to specific acoustic frequencies, or use sound as a communication channel, rest on results that have not been robustly replicated.
Who this applies to
plants generallyDo not extend this beyond the taxa listed — the popular version over-reaches.
Studied in
Plantae
You may have heard
“Plants scream when they are cut”
Drought-stressed and damaged plants do emit ultrasound, and the mechanism is a water column snapping under tension inside the stem — the same physics as a cracking knuckle. It carries no message and is not produced for a receiver. Something might listen to it; nothing is being told.
Why we rate it this way, and what the caveats are
Not enough evidenceModerate confidence
Acoustic emission is real and mechanically explained without invoking a signal. The response side — plants doing something useful in reply to sound — has not accumulated replicated evidence, and NatureHQ records the area as open rather than as refuted.
How far it can be extended
The absence of evidence is stated for the specific claim of acoustic communication; it is not a claim about all possible plant sensing.
Caveats
Emission is genuine; the dispute is entirely about whether it functions as a signal.
Absence of replicated evidence is not the same as disproof.
Popular coverage of "screaming plants" describes cavitation noise, not distress.
Still unanswered
Do any organisms — insects, for instance — use plant cavitation noise as a cue?
Can reported growth responses to specific frequencies be replicated independently?
The main advocacy for the research programme, included as a marker of where the frontier is rather than as support for the claim.
A drought-stressed plant does emit ultrasound, and the mechanism is well understood: water columns inside the stem are under tension, and when one snaps it makes a noise — the same physics as a cracking knuckle. It carries no message and is not produced for a receiver. Something might listen to it. Nothing is being told.
Practical
Reading claims about plant intelligence
Three habits keep this area straight. Distinguish a signal from a cue — a signal is produced because of its effect on a receiver, a cue is information leaking out whether or not anyone benefits. Ask who benefits from the sending, because if nobody does then "communication" is the wrong word however real the chemistry. And treat memory, learning and intelligence as claims needing their own evidence rather than as descriptions of any behaviour that responds to something.
A field that was dismissed for a decade, vindicated by better experiments, and then given a completely different explanation.
1983
First observation
The first claim of plant-to-plant signalling
Poplar and maple saplings sharing sealed air with damaged neighbours increased their defensive phenolics. The design was weak — small samples, shared chambers, conditions nothing like a forest — and the idea was treated as unserious for over a decade.
Wild tobacco growing beside experimentally clipped sagebrush in the Great Basin sustained significantly less herbivore damage over a season, with the effect travelling through air rather than through soil.
Changes how the 1983 result reads
An initially weak study proposing a real phenomenon is a different thing from a wrong one. What rescued the idea was not repetition but a field design — outdoors, natural spacing, damage measured over a season, airborne and below-ground routes separated.
Blocking airflow between leaves of a single lima bean plant abolished the defensive response in the undamaged leaf, even with the vascular connection intact. Volatiles were reaching parts of the plant its own plumbing could not.
Changes how the 2000 result reads
This dissolves the evolutionary puzzle rather than solving it. If volatiles exist to carry a message within one plant, then neighbours benefiting is a leak rather than altruism — and "plants warning each other" stops needing an explanation because it stops being what is happening.
The below-ground literature is found to be over-read
A systematic review of common mycorrhizal network research found positive citation bias and overinterpreted results running well ahead of the underlying evidence — a caution that applies to the airborne literature too.
Is emitting volatiles ever advantageous to the emitter with respect to neighbours, or is it always a leak?
Why it matters: This is the difference between plant communication being a signalling system and being an overheard side-effect, and the framing of the whole field turns on it.
Does the blend encode which herbivore is attacking, or only that something is?
Why it matters: A blend that specifies the attacker would be a referential signal in plants, a substantially stronger claim than priming.
Is kin-biased response found outside sagebrush?
Why it matters: One species is a thin base for a phenomenon widely reported as a general property of plants.
Do insects use plant cavitation noise as a cue to find stressed plants?
Why it matters: It would make the sound biologically meaningful without making it a signal — the distinction the whole section rests on.
NatureHQ publishes its own gaps. This record is at 98% completeness against what we would call a finished subject.
2 high-priority search intent(s) not yet covered
Which specific volatile compounds carry which effects is not covered, and the chemistry is well studied.
Plant–pollinator signalling is treated on the pollination and flower subjects rather than here, which splits a subject that arguably belongs in one place.
Plant–microbe signalling is described through root exudates only; the molecular dialogue of nodulation and mycorrhizal colonisation deserves its own treatment.
Whether any organism uses the ultrasonic emissions is the most interesting open question on the page and cannot be answered yet.
Last reviewed 2026-08-11 · 12 claims · 193 search questions answered on this page