The moderator we interviewed—using the pseudonym Alex—worked six months in 2025 as an English-language content reviewer for a platform with more than a billion users. Alex signed an NDA, received three weeks of training, and left after a mental health crisis their employee assistance program did not adequately address. This interview describes systemic conditions, not individual user content.
We verified employment through redacted pay stubs and training materials Alex provided. We cannot name the vendor or platform without risking identification. Alex’s account aligns with academic research and whistleblower reporting on moderation work from 2018 through 2026.
Alex requested we not describe specific content categories in detail beyond general labels—both for NDA compliance and because readers deserve warning without voyeurism toward trauma workers absorb.
Alex spoke slowly during the interview, pausing before answers—a rhythm we interpreted as care, not hesitation. Secondary trauma does not disappear when employment ends; Alex wanted that noted for anyone considering the work as temporary gig without psychological preparation.
Day One vs. Day Ninety
Training emphasized policy categories: hate speech, nudity, violence, self-harm, misinformation tiers. Trainers warned content would be disturbing. Alex thought prior retail and security work had hardened them. Day ninety Alex flinched opening the review queue. The volume and severity exceeded preparation.
Shifts were eight hours with two fifteen-minute breaks and a thirty-minute lunch. Metrics tracked: decisions per hour, accuracy against auditor samples, time off tool. Falling below thresholds triggered warnings; three warnings meant termination. Alex says bathroom breaks beyond scheduled times required supervisor approval in practice despite official policy saying otherwise.
Training included ten minutes on wellness resources that felt like checking a box. Alex later learned those resources were counted in vendor RFPs as “worker support” while queue targets stayed unchanged.
“Training warned me. Nothing prepared me for volume and severity combined.”
What the Queue Actually Contains
Alex will not describe specific videos—NDA and secondary trauma concerns—but categories dominated: graphic violence, child exploitation reports routed for immediate escalation, hate slurs in multiple dialects, livestream self-harm attempts requiring emergency escalation protocols that felt ritualized rather than supportive.
Escalation paths existed for law enforcement referrals and self-harm. Alex felt trained to click correctly, not cared for after clicking. Wellness coaches appeared monthly with breathing exercises while queue targets unchanged.
Auditors re-reviewed random samples of Alex’s decisions weekly. One bad week dropped accuracy score and triggered a warning meeting with thirty people waiting outside the door for their turn—humiliation by design.
“Metrics tracked decisions per hour. Bathroom breaks needed supervisor approval in practice.”
Pay, Classification, and Powerlessness
Alex earned twenty-two dollars hourly as a US-based contractor through a third-party vendor—above minimum wage, below livable in their metro after self-employment taxes and therapy copays. No paid sick leave. Health insurance purchased independently. Coworkers in other countries performed identical tasks for fractions of that wage; Alex learned this from a leaked internal wiki screenshot and felt complicit.
NDAs prohibited discussing client names, policy details, and queue contents. Alex felt muzzled discussing working conditions with family. Union organizing chatter on private Discord channels feared infiltration.
Contractor status meant Alex could be deactivated without cause. No unemployment benefits followed. Savings covered one month; family help covered another.
“Twenty-two dollars an hour to look at what users never see—then therapy copays.”
The Breaking Point and Exit
Alex broke down after a week of consecutive high-severity escalations without reduced queue load. A supervisor suggested “resilience training.” Alex’s therapist recommended leaving. Notice period was two weeks; Alex used PTO they had accrued and did not return.
Six months later Alex works retail management—lower status on LinkedIn, higher sleep quality. They still startle at certain notification sounds. They applied for a scholarship-funded counseling program for tech workers leaving harmful roles.
Coworkers traded dark humor Alex now finds disturbing in retrospect. At the time it felt like the only valve. Therapy helped unpack that normalization.
Alex’s retail manager job pays less but includes PTO used for therapy appointments without gig-schedule guilt. Trade accepted consciously.
Moderation vendors rotate training cohorts constantly—high churn disguised as growth hiring. Alex wants applicants to ask tenure metrics in interviews the way tech workers ask about engineering retention.
“Resilience training after a breakdown. Reduced queue load never came.”
What Platforms Could Change
Alex is not an engineer or executive and does not pretend to know moderation at scale is easy. They argue current models optimize for throughput and contractor liability minimization over worker sustainability. Shorter shifts on high-severity queues, mandatory rotation, W-2 employment with benefits, and trauma-informed counseling integrated—not EAP box-ticking—would reduce harm without eliminating moderation jobs.
Users who report content rarely think about the human who views it. Alex asks for that thought briefly—not to solve capitalism in a comment section, but to acknowledge labor platforms treat as invisible.
Alex still believes moderation is necessary work done badly by systems optimized for cost. Humans should not be disposable filters for infinite content uploads.
“I wish I quit at week eight when nightmares started—not month six.”
Would They Recommend the Job?
Alex says no for anyone with trauma history, untreated mental health conditions, or financial alternatives. Maybe temporarily for someone with strong support, strict time limits, and exit plan— but six months was too long for Alex. They wish they had quit at week eight when nightmares started.
Policy and AI automation shifted some categories to machine pre-filtering during Alex’s tenure. Humans remained the backstop for edge cases machines mishandled—the hardest cases, routed to the same underpaid queue.
Would Alex return for double pay? No. For triple? Still no. Money does not unwind what eyes cannot unsee—Alex’s words, not ours.
Alex compares moderation to other invisible infrastructure jobs—sanitation, overnight stocking—except those workers are not asked to view trauma as productivity input. That distinction matters when platforms compare themselves to “hard jobs” without acknowledging hard content.
Alex supports regulation requiring platforms to publish moderation worker counts and average tenure by region—not names, aggregates. Sunlight might not fix wages immediately but would end plausible deniability about churn.
“Humans get the edge cases machines fail—the hardest cases in the queue.”
Content moderation keeps platforms usable and legally defensible. Workers like Alex pay costs executives rarely line-item. Anonymous testimony cannot reform an industry alone—but silence guarantees nothing changes. If you use social platforms daily, someone like Alex reviewed what you never had to see.
Alex asked readers one thing: next time a platform brags about AI moderation, ask how many humans still review what the AI misses—and what they are paid. Visibility is not unionization, but it is a start.
Alex still follows moderation worker forums anonymously—not to relive trauma, but to warn newcomers when vendors repost the same job ads with higher promised pay and unchanged queue policies. Patterns repeat until someone documents them.
Alex does not regret speaking anonymously despite fear. Silence felt like complicity with a system that treats moderators as interchangeable filters. If one reader reconsiders a six-month plan without support, the interview served its purpose.
Moderation work will not disappear as AI filters expand—it will concentrate on worst cases routed to humans least equipped and paid to absorb them unless policy changes.
Alex hopes future moderators get shorter shifts on high-severity queues before they need Alex’s hindsight—prevention beats testimonial.