The AI Detection False Economy: Fueling FOW (Fear of Writing) via @sejournal, @andybetts1

For the last 10 years, AI has helped with everything I do, content included. Last week I decided to take a break from it. No assistant in the corner, no prompts; nothing polished after the fact. One article, the old-fashioned way, all me.

Once I finished, I did the obvious thing and turned back to AI – running it through several of the best-known AI detectors. The results were all over the place:

Same article. Same words. Different answers every time. Which left me with a simple question: If these tools cannot agree on something I know is entirely human, what exactly are they detecting?

And if businesses, editors and clients are making decisions based on these scores, what happens when every detector gives them a different answer?

To be clear, I am not here to name and shame individual detectors. Some are better than others, and there will always be good and bad products. My point is wider than any one tool, because the problem sits with the whole idea of human vs. AI detection tools and the fear-based false economy built around it.

I Ran AI Detectors on Writing From Before ChatGPT, Claude and Co.

If AI detectors really work, there should be one thing they are good at: writing that predates generative AI. It could not have come from a machine, so it is the cleanest test there is of whether these tools can recognize human writing. After more than 25 years of writing, much of it as a ghostwriter, I had plenty of older work to dig into.

The first article I pulled out was from 2014, eight years before ChatGPT launched.

Most tools cleared it, though one would only go as far as a 66% probability that a person wrote it. Another declared it 0% AI (more on that one later).

Back in 2019, I wrote an article for Search Engine Journal called “Technical SEO Is a Necessity, Not an Option,” published years before ChatGPT arrived. The results weren’t much better.

It was funny at first. Then it stopped being funny, because this was not just one strange result anymore. If detectors can’t reliably identify writing created before generative AI existed, how much faith should we really put in the scores they are producing today?

On the flip side, you could easily argue it’s a clear sign of how far AI has come, getting so close to matching human language that many detectors can’t tell the difference.

AI Detector False Positives Keep Showing Up

The more I tested, the less this looked like a one-off. I also went back to another Search Engine Journal article I had written in July 2021, more than a year before ChatGPT arrived. The tools split again. One read it as 86% human. Another went the other way entirely, calling it 76% AI-generated. Same article, completely different verdicts.

By this point I’d tested writing from 2014, 2019, 2021 and 2026, and the detectors kept contradicting each other.

There is a lot of real research out there (mine was just a little test). Studies that ran historically human-written documents, including newspaper opinion pieces, through leading AI detectors found many were falsely flagged as machine-written.

One comes with a real twist: Stanford researchers found the problem was even worse for non-native English speakers, with some detectors incorrectly flagging large amounts of their work as AI. If you are a business using international writers, that’s an uncomfortable thought.

At the same time, a 2026 study published in ScienceDirect found that small human edits allowed much of the genuinely AI-generated content they tested to bypass detection altogether. So that leaves many in an awkward position: the tools can wrongly accuse human writers while missing some of the content they are supposed to detect. And a few human tweaks can bypass them anyway.

That doesn’t build much clarity or trust.

The False Economy of AI Detection

Bigger picture, what bothers me is that these detectors trade on a writer’s fear. Scores swing wildly from tool to tool. Add it all up and two things fall out of it: confused writers, and fear with a price tag on it.

FOW, the fear of writing. Writers are second-guessing work they know is good, worried some tool will decide it “looks AI.” Content used to be judged on whether it was useful, original, or well written. Now, more and more, the first question is whether a detector thinks a machine wrote it. Somewhere there, things went wrong.

That fear is running right through the whole industry. Writers worry about being accused of using AI. Agencies worry about clients running detector scores. Businesses worry about the impact on Google and about LinkedIn flagging them. Detectors, good or bad, feed on it. People panic-buy products based on emotion – because they suddenly feel essential.

The cost is jobs. Writers are losing work because nervous clients run scans and treat the numbers as gospel, turning an unreliable technology into a confidence crisis for the content industry. Increasingly, the verdict itself sits behind a paywall too, sold per scan, per word, per seat. Pay up, and they’ll tell you whether you’re human.

The Humanizer Upsell

Several companies in this market sell an AI detector with one hand and an AI “humanizer” with the other, built to help text beat detectors like their own. Many of you (my human prediction) will have run into this: a clean verdict, then an upsell. As I shared earlier, they love to tell you that your writing looks like AI when they can. And when they don’t, they still go for an upsell.

I did, on an article I wrote in 2014. One detector scored it 100% human, then in the same moment offered me the option to humanize it, with a paid upgrade to learn more. Humanize what exactly? The human? Irony doesn’t get much better.

And it sums up the whole AI detector tool business: an industry charging writers, content marketers and editors to humanize content that was human to begin with. Call it what it is, a toll nobody needs.

That’s a false economy. Money keeps pouring into a verification layer that can’t reliably verify anything, and the monetization continues. Writers second-guess themselves. Editors and contributors eye each other with suspicion. And every new model release resets the arms race. Fear goes in, revenue comes out, and nobody is any closer to the truth.

Why AI Struggles to Human and AI  Writing

More irony here. Generative AI and writing assistants were built to sound like humans. That was the whole point. These models were trained on human writing so they could produce something natural enough to pass for it. The fact that detectors now struggle to tell the difference might be the strongest evidence of how well that worked.

That’s what makes the whole detector debate so strange. We’re asking one AI system to tell us whether another AI system sounds too much like a human, when the second system was designed to sound human in the first place.

The mimicry worked, which was always the goal.

When purpose-built detectors can’t consistently pull human writing apart from AI-assisted writing, on what grounds does anyone accuse a writer of doing something wrong because AI helped?

LinkedIn, AI Detection, and Slop: The Social Side

The detection tool debate is spreading everywhere, and getting confused along the way. Publishing is going through its own reckoning over AI-written work. LinkedIn, meanwhile, has just added a “seems like AI slop” button so members can flag posts they think used AI. It’s detection again, but with humans as the “detectorists,” though its goal and approach are different.

Patrick Coffee at The Wall Street Journal has just written a timely piece on exactly this, digging into third-party AI detector findings on LinkedIn post content. It was interesting to see LinkedIn question the vendors’ numbers while declining to provide comparable data of its own. A few quotes worth sharing.

“Their research appears to treat any content that AI touches as AI slop, which is not how we look at it,” said a LinkedIn spokeswoman. “… AI can be a great support tool in helping people in articulating ideas, refining language, or making language more concise.”

And Dan Roth, LinkedIn’s editor in chief and vice president of content, makes a good point: “Obviously, a big share of LinkedIn’s content is AI-assisted, but nobody can know exactly how much, including LinkedIn, and anyone claiming they do is selling a detection tool.”

AI slop is a real problem, but I’m not sure a button fixes it. Hand people a flag and some will point it at competitors and posts they simply don’t like. A feature built to clean up the feed could just as easily end up turning people on each other.

And there’s a harder question underneath. Will LinkedIn use the flagged posts to tune future AI-assisted writing models?. If they are, then millions of members are effectively training AI, unpaid, to write in ways that stop getting flagged. Time will tell.

Evaluating Content With Humans in the Full Loop

None of this is me defending lazy AI-generated content. I’ve read plenty of it. I’ve also read plenty of poor content written entirely by humans.

And to be clear the other way: Use AI. It has earned its place in every modern marketing workflow, mine included. For content writing, it genuinely helps with ideation, research, insights, and creative assistance. AI for marketing and content creation is not the issue here. The issue is the accuracy of AI detection tools, the fear they generate, and scores pretending to be the truth.

That’s where I think many have lost their way. Instead of asking and over-indexing on whether AI wrote it and “false scoring,” ask the questions that matter:

Questions like those catch weak content far more reliably than a detector score ever will. They always have.

That’s also much closer to Google’s position. Google’s guidance has been steady on this for years: Reward helpful, reliable content, however it was produced. Its enforcement goes after low-quality content at scale. It doesn’t sit there deciding whether each sentence came from a person or a model.

If organizations genuinely care about how content is created, they’re better off having a clear AI policy than relying on detector scores.

In academia, Indiana University’s Kelley School of Business bans AI detection tools outright in its faculty AI playbook. It calls them unreliable, tells staff not to upload student work to them at all, and points everyone toward clear policy instead. When a leading business school won’t trust these scores on student essays, why would a brand trust them on marketing copy?

And if provenance really matters to you, lean on humans and your work colleagues and peers. Draft history. Version control. Editorial review. A writer’s body of work. Any of those carries more context than a percentage spat out of a black box.

The Rise of FOW: The Fear of Writing

What started as a small personal experiment points to something much bigger, and it’s working its way through the industry. AI detection is loading writing with fear. Fear that honest human work gets questioned. That AI-assisted work gets rejected. That a single score from a single tool somehow counts as the truth. Fear creates demand. Demand creates products. There’s the false economy.

And that’s how FOW takes hold. Honest work doubted because a machine produced a number. None of that is healthy, not for writers, not for editors, not for the industry.

Always challenge poor content. Check the facts, and back human judgment ahead of tools. Push writers for original thinking and real expertise. Just don’t confuse any of that with a detector score. And if you do want to run checks, point them at the non-negotiables, like plagiarism, where a result actually means something.

I wouldn’t spend my money trying to prove writing is human. I’d spend it making the writing better.

Just for transparency, I used AI to help me with this article. Here are some of the results from running this through the detectors.

The last time I bother, to be honest. We’ll see, eh?

More Resources:

Featured Image: La Terase/Shutterstock