New research on content moderators challenges a widespread assumption in tech and has important implications for every organization operating under sustained pressure.
When people think about psychological harm at work, they tend to think about exposure. What someone saw. What they heard. How often, and how bad. The logic feels obvious. The more distressing the input, the greater the damage. Manage the input, manage the risk.
But new research from the University of Washington puts that assumption under serious pressure. In a study of content moderators, workers reviewing graphic, violent, and disturbing digital content at scale, the number of hours spent on distressing material was not a robust predictor of mental health outcomes. What predicted those outcomes, consistently and across two separate samples, was something else entirely.
The researchers found that workplace culture, negative automatic thoughts, perceived stress, and avoidant coping were among the strongest and most consistent predictors of depression and PTSD severity. Poorer perceived workplace culture was associated with higher depression. Avoidant coping predicted both PTSD and depression, including prospectively three months later.
That matters because it shifts the conversation.
It suggests that in high pressure environments, the core question is not only what people are exposed to. It is what kind of system they are expected to carry that exposure inside.
The scale of harm is hard to ignore
The findings are not marginal. Across the international and U.S. samples, the study found elevated rates of probable mental health difficulties among content moderators, including PTSD, depression, somatic symptoms, and alcohol misuse. In the international sample, 48.5 percent met the threshold for probable major depression, 25.9 percent met the threshold for probable PTSD, and 68.7 percent reported moderate to severe somatic symptoms. In the U.S. sample, 42.1 percent met the threshold for probable depression and 26.3 percent for probable PTSD.
The interviewer rated findings were just as striking. In the U.S. group, content moderators had higher interviewer rated PTSD severity than the comparison group and were over eight times more likely to meet criteria for a current mood disorder. They were also more than twice as likely to report a lifetime history of major depressive disorder.
These are not small effects. And importantly, they appear in a workforce where employers have already tried to reduce risk through exposure limits and access to support. The paper notes that moderation hours may already have been constrained by organizational policy, which may explain why the amount of distressing content reviewed each day was not the clearest driver of outcomes. Instead, once exposure is capped, the surrounding organisational context becomes more important.
That is a crucial point for leaders.
Because it suggests that many existing wellbeing efforts may be focused on the wrong lever.
The question organizations keep avoiding
Content moderation is an extreme example, but the underlying mechanisms are not unique to it.
The same pressures appear across Trust and Safety, fraud and risk teams, AI red teaming, cybersecurity operations, emergency response environments, and other roles where people are asked to maintain judgement, consistency, and performance under sustained psychological load. The specifics vary. The pattern does not.
High demand. Limited control. Repeated exposure to difficult material or decisions. Insufficient recovery. Cultures that talk about wellbeing but do not operationalise it.
The research effectively points to a conditional dose effect. When exposure is already managed to some degree, organizational conditions become the dominant driver of mental health risk. Those conditions include team support, supervisory practice, social connectedness, workplace culture, and whether people have meaningful ways to process strain before it accumulates.
This is the shift that remains stubbornly difficult for most organizations. It is easier to commission an EAP, run a mental health awareness campaign, or add a wellbeing app to the benefits package than it is to examine the structure of the work itself. But the evidence increasingly points toward that harder work as the thing that actually changes outcomes.
Wellbeing support that sits outside day to day work rarely changes what happens inside it.
What the research actually points toward
One of the strengths of the paper is that it does not stop at documenting harm. It identifies a set of modifiable targets that organizations can act on. These sit at both organizational and individual level.
At organizational level, the research points to the importance of culture, support, pacing, and work design. Moderators who perceived workplace culture more positively had lower depression severity. Those who disclosed aspects of their work to fewer people tended to report worse outcomes. Distress linked to intrusive memories also emerged as a strong predictor, suggesting that the way work is structured and processed has real clinical implications.
At individual level, negative automatic thoughts were among the strongest predictors across both samples, both cross sectionally and over time. Perceived stress and avoidant coping were also consistently associated with worse mental health. The paper explicitly points toward interventions such as cognitive restructuring, stress reduction, adaptive coping support, and changes to moderation systems that may reduce intrusive memory formation.
That distinction matters.
Because these are not fixed vulnerabilities or personal weaknesses. They are patterns that can be shaped by the environment and addressed through evidence based intervention, if that intervention is timely, clinically grounded, and embedded into the reality of the work.
This is where many organizations still fall short. They offer support, but too late. Or too generically. Or in ways that depend on people self-identifying only after strain has become severe.
By then, performance has often already started to erode.
Early warning signs are there, if the system is built to see them
Another useful implication of the research is that deterioration is not random. It leaves signals.
The paper highlights the role of intrusive memories, avoidant behaviour, persistent negative thinking, ongoing stress, and low social connectedness. These are not abstract clinical concepts. In practice, they can show up as withdrawal, cynicism, difficulty switching off, increased disengagement, emotional numbing, reduced tolerance, or a shift in how people talk about themselves, others, and the work.
These are catchable moments.
But only if line managers, team leads, and People functions are trained to recognise them, and only if there is a system around the individual that can respond in a safe and credible way.
That means more than signposting. It means access to specialised support that understands the nature of the work, local clinical pathways, trusted escalation routes, and leadership capability that goes beyond general empathy training.
Why this matters beyond content moderation
The researchers explicitly connect these findings to adjacent forms of human in the loop work, including AI red teaming and data enrichment. That extension is especially important now.
These workforces are growing quickly. Red teamers, evaluators, and data labellers are often exposed to repetitive and sometimes disturbing outputs in the service of making AI systems safer. In many cases, the support structures around them are less mature than those built around large scale content moderation operations.
That creates a familiar risk.
As AI development accelerates, so does the need for human infrastructure to test, label, challenge, and contain what these systems produce. The wellbeing of that infrastructure deserves the same rigour being applied to the technology itself.
Otherwise the sector risks repeating a pattern it should already recognise: building critical safety functions on top of human systems that are not designed to sustain the people inside them.
What organizations should take from this
The lesson from this research is not that exposure does not matter. It does.
But exposure is not the whole story, and in many environments it is no longer even the most actionable one.
The stronger question is this:
What happens after the exposure?
What kind of team does someone return to?
What kind of manager notices the change?
What kind of culture shapes whether they speak?
What kind of support is available before strain becomes illness?
What kind of work design prevents pressure from turning into harm?
These are operational questions, not just wellbeing ones.
And for organizations serious about performance, retention, resilience, and risk reduction, they should be treated that way.
Final thought
Content moderation is often discussed as if the work itself is the whole problem.
This research suggests otherwise.
The deeper risk lies in what surrounds the work. The culture. The leadership. The coping patterns it encourages or ignores. The structures that either help people recover or quietly wear them down.
That is why the future of wellbeing in high pressure environments will not be defined by perks, platforms, or reactive support alone.
It will be defined by whether organisations are willing to redesign work around human capacity.
Because pressure may be inevitable.
But harm is not.
