The freelancer we are calling "Alex" - a pseudonym, at their request - had been writing for the content agency for three years. It was their largest client, accounting for roughly 40% of their annual income. The arrangement was straightforward: Alex received assignments, conducted research and interviews, wrote the pieces, and submitted them through the agency's portal. The work was not glamorous - product roundups, how-to guides, industry analyses - but it was steady, it paid reasonably, and Alex took pride in doing it well.
In November 2025, the agency implemented an AI detection policy. All submitted content would be screened before payment. Writers whose work was flagged above 30% would have their submission rejected and would not be paid for that piece. The policy was communicated in a three-paragraph email with no opportunity for discussion.
The Flag
Alex's first rejection came on a 2,000-word guide to choosing accounting software for small businesses. They had spent four days on it: two days researching, interviewing a CPA and a small business owner, and testing three software products; two days writing and revising. The detection tool scored it at 47% AI probability.
The rejection email was automated. It contained the score, a reminder of the policy, and no mechanism for appeal. Alex replied with a detailed email explaining their research and writing process, attaching interview notes, draft timestamps, and source links. The agency's response came two days later: "We appreciate your dedication to quality. However, our policy is firm. We suggest reviewing your writing process to reduce AI-like patterns."
The Pattern
Over the next six weeks, three more of Alex's submissions were flagged. The scores ranged from 33% to 52%. Each time, Alex compiled evidence of their authorship. Each time, the evidence was acknowledged but the rejection stood. The agency's position was consistent: the tool's output was the policy, and the policy was the tool's output.
Alex began to notice which pieces were flagged and which weren't. The flagged pieces shared a characteristic: they were well-organized, clearly written, and structured according to the content formats the agency itself had specified. The pieces that passed were rougher - less structured, more conversational, with longer paragraphs and less predictable organization. Alex was being rewarded for writing worse and punished for writing well.
The End
After the fourth rejection, Alex requested a phone call with the agency's editorial director. The call lasted twelve minutes. The editorial director acknowledged that false positives were possible but said the agency could not make exceptions without undermining the policy's credibility. When Alex asked whether the agency had evaluated the detection tool's false positive rate, the director said they relied on the tool provider's published accuracy metrics.
Alex completed their remaining assignments and did not accept new ones. They lost approximately $2,400 in rejected work and $18,000 in projected annual income from the relationship. The agency replaced them within a week.
The Aftermath
Alex now works with three smaller clients, none of which use AI detection screening. Their income has not recovered to pre-rejection levels. They have considered legal action but were told by a lawyer that the agency's terms of service - which Alex had signed - gave the agency broad discretion over content acceptance criteria.
"The worst part isn't the money," Alex told us. "The worst part is knowing that somewhere, an algorithm decided I wasn't real, and a company I'd worked with for three years believed the algorithm over me. I have recordings of the interviews I conducted. I have the phone numbers of the people I talked to. I have timestamps on every draft. None of it mattered."
The agency continues to use the same detection tool. They have not published their rejection rate or false positive data. When we reached out for comment, they declined.