It is a Tuesday in the kind of late evening that has stopped being evening and become the soft early portion of the next day. The phone is on the bed. The thumb is moving without instruction. A child has been killed in a place whose name the thumb does not need to learn because the next post is already loading, and the next, and the one after that is a recipe, and the one after that is a friend's grief about a parent who is dying, and the one after that is a brand. The body is on the bed. The body has not moved in forty minutes. Something has happened in the body that does not have a clinical name yet, and it is the same thing that is happening in roughly four billion other bodies, on roughly four billion other beds, at roughly the same hour, every night.
Compassion fatigue researchers would recognize the posture. They would not be surprised to learn it has scaled.
The condition the literature describes, first by Charles Figley in 1995 and then across thirty years of subsequent work, is the predictable consequence of sustained exposure to other people's suffering through a witnessing relationship that the witness has no power to act inside [1, 2]. Figley named it the cost of caring, and the framing has held up because the framing was correct. The cost is paid by the witness, not by the source of the suffering. The witness, given enough exposure at sufficient intensity over sufficient duration, develops a measurable reduction in the capacity to be moved by the next exposure. The reduction is not a moral failure. The reduction is the mechanism the body uses to keep functioning when the demand on the witnessing system exceeds what the witnessing system was built to absorb.
The clinicians who have spent careers measuring this condition have measured it almost exclusively in helping professionals. Trauma therapists. Emergency-room nurses. Child protection workers. Hospice staff. Sexual-assault counselors. The instrument the field uses, the Professional Quality of Life Scale developed by Beth Hudnall Stamm, has been validated across more than two hundred studies and translated into twenty-eight languages, and it measures three things: compassion satisfaction, burnout, and the secondary traumatic stress that accumulates from repeated exposure to other people's traumatic material [3, 4]. The cutoffs are clinical. The findings are consistent across cultures, across professions, across decades.
A 2024 systematic review and meta-analysis published in PubMed synthesized 196 studies covering 73,034 nurses and 4,551 nursing students, and found that post-pandemic, compassion fatigue scores had risen, with intensive care and emergency-department staff carrying the highest burden. A 2023 cross-sectional Spanish study of 710 nurses across nine high-complexity hospitals reported burnout prevalence above 20% and secondary traumatic stress prevalence at 30%, with each subscale significantly associated with the intention to leave the unit, the career, or both. A 2025 Chinese study of 581 clinical nurses found burnout in 31% and secondary traumatic stress in 34.8% of the population, with sleep quality and social support emerging as the strongest predictors of who could withstand the exposure and who could not.
The professional literature is unsentimental. The instrument measures what a body does when it has been asked to attend to suffering it cannot resolve.
What was supposed to happen
The literature was not supposed to apply outside the clinic.
Compassion fatigue was theorized as an occupational hazard: a condition specific to a particular kind of work, performed by a particular kind of trained person, in a particular kind of institutional setting that could in principle be modified to protect the worker. The hospitals that have moved the dial on it have done so by limiting shift exposure, by mandating peer-support sessions, by funding clinical supervision, by treating the witnessing capacity of the staff as a resource that must be replenished or it will be depleted. The interventions are real. The pre-pandemic data showed measurable improvement in units that took the framework seriously [3, 4]. The framework assumed a perimeter. Inside the perimeter, the witness was attending to suffering. Outside the perimeter, the witness was a person again, eating dinner, sleeping, watching something stupid on television, recovering. The perimeter was the recovery zone. Compassion fatigue was a condition of working inside the perimeter for too long, and the recovery zone was where the witnessing capacity rebuilt.
The recovery zone has stopped being a zone.
The 2025 Reuters Institute Digital News Report, drawing on responses from nearly one hundred thousand people across forty-eight markets, recorded that 40% of global audiences now sometimes or often actively avoid the news, up from 29% in 2017. In Bulgaria the figure is 63%. In Croatia, 61%. In Turkey and Greece, both above 60%. The report's lead authors, Nic Newman and the Reuters Institute team at Oxford, describe the avoidance with care: it is selective, it is rising, and it tracks most strongly with younger audiences who report that the news feels overwhelming, confusing, or far removed from their lives [8, 9]. The avoidance is the same thing the ICU nurse does when she cannot, on the seventh shift in a row, allow the eighth dying patient to land. It is not a failure of civic duty.
The condition is no longer specific to the helpers. The condition is the population.
A note about the body on the bed
Here a single voice is required, because what happens at the level of the population is not legible at the level of the population. It is legible only at the level of one person inside it. I have been a witness for an hour. The witnessing has been nominal. I am not on the ground in the place where the child was killed, I cannot help the friend whose parent is dying, I cannot intervene in the catastrophe in the second post or the seventh post or the nineteenth, and the nominality is the part the older framework cannot account for. A trauma therapist sits in a room with a single client for fifty minutes. A hospice nurse is at one bedside at a time. The witness in those frames has work to do and a person to do it with, and the doing is what gives the witnessing its shape. I am performing nothing. I am receiving. The receiving has no shape and no end. The thumb keeps moving because the alternative to the thumb moving is a stillness I cannot, on this Tuesday, hold.
I notice, as the hour ends, that I have stopped registering distinction between categories of grief. The dying parent and the dead child and the friend's divorce are landing on the same surface in me, with the same weight, which is not weight any longer. They are landing on a surface that has become acoustically dead. The technical language for what I am experiencing exists in the literature on emergency medicine, where it is called emotional numbing, and it is one of the listed symptoms of secondary traumatic stress [1, 2, 6]. I have all the symptoms. I have not provided care to anyone today. I have not been near another person's suffering in any direct sense. I have only been receiving, for forty minutes or sixty, the broadcast of suffering that the architecture is designed to deliver to me at the rate the architecture can sustain, which is faster than I can metabolize.
I close the phone. I do not feel better. The condition does not invert when the source is removed; the literature is consistent on this, and the body confirms it [2, 5]. I will sleep. The sleep will be shallow. I will wake on Wednesday and my capacity to attend to my actual life, the people I love, the work I have committed to, the small obligations of being a person, will be slightly lower than it was on Monday, in a way that is not yet measurable to me but that is, across an aggregated population, measurable to anyone who looks.
This is one body on one bed on one Tuesday. The aggregation is the thing the literature has not yet caught up to.
What the platforms measure
The platforms see the aggregation in their dashboards. They have a different name for it. The 2025 Buffer State of Social Media Engagement report, drawn from analysis of more than 52 million posts across 200,000 accounts, documented year-over-year engagement collapses across most major platforms. Instagram fell roughly 26%, from a median engagement rate of around 7.3% to approximately 5.4%. Threads dropped about 18%. LinkedIn dropped 5%. The longer view from DataReportal's October 2025 dataset shows the structural pattern: 5.66 billion social media identities, an increase of 4.8% year-over-year, while average daily time spent on platforms dropped to 2 hours and 21 minutes. A separate Financial Times analysis by John Burn-Murdoch, drawing on GWI survey data covering 250,000 adults across more than 50 countries, found that average time on social media peaked in 2022 and has since declined nearly 10%, with the steepest fall concentrated among teens and adults in their twenties, the cohorts the advertising market is built around.
In April 2025, testifying under oath in the U.S. Federal Trade Commission's antitrust trial against Meta, Mark Zuckerberg acknowledged that Facebook's and Instagram's share of the time people spend on social media apps had "gone down meaningfully". The platforms had no language for the condition. The platforms had dashboards that registered the condition as engagement loss, monetization loss, the inability of the system to extract from its users the sustained attention the system had been engineered to convert into revenue.
The dashboards are looking at the same body the clinicians are looking at, from the opposite side. The clinician sees a person who has stopped being able to bear witness. The platform sees a person who has stopped being able to be reached. Both are correct. The instrument that documents the condition from the clinical side is the ProQOL. The instrument that documents the condition from the platform side is the engagement dashboard. Neither instrument is built to acknowledge what the other one is measuring. They are reading the same depletion at different layers of the same body.
This is the place the older framework breaks. Compassion fatigue was theorized as the cost of caring. The condition the population is now in is the cost of being unable to stop receiving the inputs that would have produced caring if the receiving had not exceeded the capacity to convert input into care.
The trigger that has stopped firing
The trigger-action architecture of the contemporary internet was built on a single behavioral assumption: signal in, response out. The user receives content. The user reacts to the content: a click, a comment, a share, a purchase, an emotion that produces an algorithmically detectable downstream behavior. The infrastructure was engineered around the conversion. The economic model is the conversion. The advertising market is priced against the conversion. The political-economic theory of how public attention now produces public action depends on the conversion still occurring.
The conversion is failing.
A 2023 study published in Scientific Reports by Saifuddin Ahmed at Nanyang Technological University and Muhammad Ehab Rasul at UC Davis, drawing on cross-national survey evidence from eight countries, found that social media fatigue directly increases the likelihood that a user will both believe and share misinformation. The mechanism the authors propose is depletion. The fatigued user does not have the cognitive resources for the deliberative processing the verification of a claim would require, and the claim travels onward through the user's network because the user has been emptied of the capacity to evaluate it. The study's contribution to the field is to identify the mediating variable. The mediating variable is exhaustion.
The 2018 Vosoughi, Roy, and Aral study at the MIT Media Lab, the largest empirical analysis of rumor diffusion ever conducted on a single platform, published in Science and covering Twitter from 2006 through 2016, documented that false news traveled approximately seventy percent faster than true news, reached fifteen hundred people six times faster, and at a cascade depth of ten was moving twenty times faster than facts. The drivers were human. The Ahmed and Rasul study clarifies the mechanism by which the human drivers are now operating: the population doing the spreading is the same population that has been receiving the broadcast for years and has run out of the cognitive resources required to do anything with the broadcast except pass it on.
The trigger is firing. The action it produces is no longer the action the architecture was designed to produce. The action has become forwarding without metabolizing, sharing without believing, consuming without responding, and, increasingly, across the demographics whose attention the architecture was built to convert, closing the application and lying still.
What the helpers learned
The clinicians learned something the platforms have not.
The hospitals that reduced compassion fatigue in their staff did not do it by reducing the volume of suffering arriving at the unit. The volume of suffering arriving at the unit is, by definition, exogenous to the hospital. The hospitals reduced compassion fatigue by intervening on the architecture of the witness's exposure to it. They built rotation systems that limited the number of consecutive shifts in high-acuity environments. They funded peer-support programs that gave staff a place to discharge what they had absorbed. They mandated supervision sessions in which what had been witnessed was processed with another trained witness. They treated the witness's capacity as a renewable resource that required infrastructure to renew, and they built the infrastructure.
The interventions were not therapy on top of the work. The interventions were modifications of the work. The unit that did not build them watched its staff degrade and leave. The unit that did build them retained staff who could continue to be present at the bedside of the next dying patient.
The contemporary internet has built none of this. The architecture has no rotation system. There is no shift cap on exposure to other people's grief. There is no peer-support system that the platform sponsors. There is no supervision session built into the product. The user is alone with the broadcast, on the bed, on the Tuesday, and the only intervention available to the user is the decision to close the phone. That decision the engagement dashboards now register as failure, even though it is the same decision the ICU nurse makes when she walks out of the unit because she has reached the limit of what one body can bear and remain useful inside.
The hospital learned that the limit was a structural property of the witness. The platform has not learned this yet. The platform's product roadmap continues to assume that the limit can be engineered around, that better personalization or richer content or more compelling formats will reactivate the trigger that the body is no longer firing. The body has moved past the place where engineering can reach it. The body has done what bodies do. The body has decided that the cost of the next witnessing is higher than the value of the witnessing, and has stopped paying.
The instrument and what it knows
The researchers who built the ProQOL were not trying to describe a condition of civilization. Beth Hudnall Stamm spent the years between the late 1990s and her retirement refining an instrument that could distinguish between the satisfaction of helping, the burnout of being asked to help under structurally impossible conditions, and the secondary traumatic stress of absorbing other people's traumatic material [3, 4]. She licensed the instrument freely so it could be used by any clinician, in any country, who wanted to know whether the staff of a hospital or a counseling center or a child-welfare office was being depleted faster than the institution was replenishing them. The instrument is mundane. It is a thirty-item self-report. It takes ten minutes. The cutoffs are calibrated against more than two hundred validation studies. The construct is one of the most thoroughly tested in the trauma-research literature.
What Stamm gave the field was not a theory. The theory was Figley's, and it preceded her [1, 2]. What she gave the field was the instrument that made the theory measurable in a concrete population at a concrete moment in time. Without the instrument, compassion fatigue was a felt sense; with it, the felt sense became a measurable condition with a baseline, a trajectory, and a clinical threshold past which the institution had to act. The hospitals that act on the ProQOL data are the hospitals that retain their nurses. The hospitals that ignore the data lose them. The instrument has not yet been adapted for the population of the broadcast. There is no cross-validated version of the ProQOL that asks a teenager in Bulgaria how many hours she spent on Tuesday receiving suffering she could not act on, and how that exposure tracks against her capacity to attend to her own life on Wednesday. The instrument exists in form. The instrument has not been pointed at the condition the population is now in, because the field that built the instrument did not anticipate that the perimeter would dissolve.
The dissolution is the point. The instrument was built for a profession. The condition is now a civic baseline. The clinical literature has the vocabulary. The clinical literature does not yet have the population.
What is left when the witness stops
The Reuters team's 2025 finding is the leading indicator the older framework would have looked for. The 40% who avoid the news are the population's analogue of the nurse who has filed a request to transfer out of the trauma unit [8, 9]. The transfer request is not nihilism. The transfer request is the body's last remaining piece of agency, exercised at the only layer where the body still has any. The user closes the application. The reader stops following the war. The voter stops watching the debate. The friend stops asking how the cancer treatment is going, not because the friend does not care, but because the friend has been receiving fourteen separate streams of cancer treatment for fourteen separate friends for the last six years and has run out of the resource the asking would require.
This is the condition the platforms cannot price. The condition the platforms can price is engagement loss, which is the correct quantity at the wrong layer. Engagement loss is the failed conversion of attention into action. The condition underneath engagement loss is the conversion of a witnessing population into a depleted one. The first is a metric. The second is a state. The platforms read the metric and assume the state is recoverable through better content. The clinical literature reads the state and reports that recovery requires structural intervention into exposure conditions, which are the conditions the platforms have spent two decades engineering against [3, 4, 14].
The body on the bed on the Tuesday has not failed. The body on the bed has done what every clinical model of overload has predicted a body will do under those conditions. The body has converted from a responder into a receiver, and from a receiver into a substrate the broadcast passes through without leaving a mark.
The transparency was not a gift the population accepted. The transparency was a condition imposed on a witnessing capacity that was never sized for it. The capacity has answered the way capacities answer when they are asked for more than they can give. The capacity has gone numb. The numbness is not absence. The numbness is the architecture the body builds when the architecture above it has stopped acknowledging that the body has limits.
If the literature on the helpers is correct — and three decades of cross-validated data suggest that it is — then the population has been operating, for some time now, inside a condition the field already had a name for, on a scale the field did not anticipate, with no instrument deployed to measure what the population is silently carrying. The phone is on the bed. The thumb has stopped moving. The stillness is not rest. What is the population that has been told its silence is its consent doing, exactly, when it has stopped responding to anything at all?