I’ve now heard from three freelance journalists who’ve had pitches rejected or contracts questioned because AI detection tools flagged their work as machine-generated.

All three wrote their pieces themselves. The AI detectors were simply wrong.

This isn’t an aberration. It’s the predictable result of deploying fundamentally unreliable technology to solve a problem that’s better addressed other ways.

It’s time for media companies to stop using AI writing detectors. Here’s why.

The Technology Doesn’t Work

Let’s be clear about the technical reality: AI writing detectors are not accurate enough for professional use.

Every major detector—GPTZero, Turnitin’s AI detection, Originality.ai, and others—produces false positives at rates that would be unacceptable in any other context. Studies have found false positive rates ranging from 10% to over 30% depending on the tool and text type.

That means one in five to one in three human-written texts might be incorrectly flagged as AI-generated. For a freelancer whose livelihood depends on their reputation, this is devastating.

The detectors also produce false negatives—failing to catch AI-written text. Light editing of AI output typically evades detection. Paraphrasing tools defeat most detectors. Anyone who actually wants to hide AI use can do so easily.

So you have a technology that wrongly accuses innocent writers while being easily circumvented by those who are actually cheating. What exactly is the point?

The False Positives Problem

False positives aren’t random. They’re systematically biased in ways that make them particularly unfair.

Writers for whom English is a second language are flagged more often. Their writing patterns—often more formal, more cautious with idioms—resemble AI output more closely than native speakers’ writing.

Writers with certain styles—formal, technical, academic—get flagged more frequently than those with casual, conversational voices.

Early-career writers, still developing their distinctive voices, get flagged more than experienced writers with established styles.

The technology is essentially biased against writers who are already marginalized or disadvantaged. That should be disqualifying.

The Chilling Effect

Even when detectors don’t produce false positives, their mere existence creates problems.

Freelancers I’ve spoken with describe changing their natural writing style to avoid triggering detectors. They’re adding contractions they wouldn’t normally use. They’re varying sentence structure artificially. They’re making their writing worse to appear more “human.”

This is absurd. Writers shouldn’t be optimizing their work to satisfy flawed detection algorithms.

Others report anxiety about legitimate AI use. A journalist who uses AI to transcribe interviews worries: if they pull a quote from the AI transcript, will detection software flag the entire piece? A reporter who uses ChatGPT to brainstorm angles worries about contamination.

The chilling effect extends to editors, too. Some have stopped commissioning certain writers because detectors flagged their work—not because they believe the writers cheated, but because they don’t want the hassle.

What Problem Are We Actually Solving?

The case for AI detectors rests on a concern about authenticity: that AI-written content published under human bylines is fraudulent.

That’s a legitimate concern. Readers have a right to know when content is machine-generated. Publications have an interest in ensuring the quality and originality of what they publish.

But AI detectors are a terrible way to address this concern.

First, they catch the wrong people—honest writers who don’t use AI or use it appropriately.

Second, they miss the actual cheaters, who can easily evade detection.

Third, they address authenticity mechanistically rather than through the relationship of trust between editors and writers.

A better approach: ask writers directly about their AI use. Require disclosure of significant AI involvement. Build relationships where deception has consequences. Trust but verify through traditional editing processes that catch quality problems regardless of origin.

This is how media has always worked. Editors don’t use plagiarism detectors to check every piece—they rely on professional relationships, reputational stakes, and editorial processes that catch problems.

The Signal Problem

Here’s what bothers me most: AI detectors give false confidence about a problem they can’t actually solve.

An editor who runs a piece through a detector and gets “0% AI probability” might assume the piece is clean. But light AI editing would evade detection. The detector has provided false assurance.

Conversely, an editor who gets a high AI probability might distrust a piece that was entirely human-written. The detector has created false suspicion.

In neither case has the detector actually helped. It’s added noise to a signal that wasn’t useful in the first place.

What Media Companies Should Do Instead

Abandon detector-based screening entirely. The technology isn’t reliable enough for professional use.

Develop clear policies on acceptable AI use. Define what’s okay (transcription, research, brainstorming) versus what’s not (having AI write content published under a human byline). Communicate these policies to freelancers and staff.

Require disclosure. Ask writers to note any significant AI involvement in their process. Make this a contractual requirement for freelancers.

Trust editorial judgment. Good editors can often tell when something’s off—when prose is strangely generic, when knowledge seems surface-level, when the piece lacks a distinctive perspective. This human judgment is more reliable than algorithmic detection.

Focus on quality, not origin. A piece that’s poorly written, factually wrong, or lacks insight should be rejected regardless of whether a human or machine wrote it. A piece that’s excellent should be published regardless of what tools contributed to its creation.

Build relationships of trust. Work with writers whose integrity you’ve verified over time. New relationships require more scrutiny—but scrutiny of quality and accuracy, not algorithmic detection.

The Bigger Picture

The impulse to use AI detectors reflects a broader anxiety about authenticity in the AI age. That anxiety is understandable. The tools available to fake competence have never been more powerful.

But the solution isn’t technological counter-measures that don’t work. It’s building systems of trust, accountability, and quality control that work regardless of what tools exist.

Media has navigated previous disruptions—digital production, online distribution, social media, mobile consumption—by adapting its practices while maintaining core values. The AI disruption requires the same approach.

Algorithmic detection isn’t the answer. Human judgment, professional ethics, and editorial rigor are. Always have been. Always will be.

Retire the detectors. They’re causing more harm than good.