The AI Detection False Economy: Fueling FOW (Fear of Writing)

Aug 11, 2026 07:00 PM - 56 minutes ago 1

For the past 10 years, AI has helped pinch everything I do, contented included. Last week I decided to return a break from it. No adjunct successful the corner, nary prompts; thing polished aft the fact. One article, the old-fashioned way, each me.

Once I finished, I did the evident point and turned backmost to AI – moving it done respective of the best-known AI detectors. The results were each complete the place:

One came backmost 100% AI-generated.
Another scored 78% AI.
Another landed connected 42%.
One thought a quality wrote it.
A mates would not show maine the consequence unless I paid first.

Six AI detector results for the aforesaid human-written article, ranging from 100% AI to very apt human-written

Same article. Same words. Different answers each time. Which near maine pinch a elemental question: If these devices cannot work together connected thing I cognize is wholly human, what precisely are they detecting?

And if businesses, editors and clients are making decisions based connected these scores, what happens erstwhile each detector gives them a different answer?

To beryllium clear, I americium not present to sanction and shame individual detectors. Some are amended than others, and location will ever beryllium bully and bad products. My constituent is wider than immoderate 1 tool, because the problem sits pinch the full thought of quality vs. AI discovery devices and the fear-based mendacious system built astir it.

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

If AI detectors really work, location should beryllium 1 point they are bully at: penning that predates generative AI. It could not person travel from a machine, truthful it is the cleanest trial location is of whether these devices tin admit quality writing. After much than 25 years of writing, overmuch of it arsenic a ghostwriter, I had plentifulness of older activity to excavation into.

The first article I pulled retired was from 2014, 8 years earlier ChatGPT launched.

Most devices cleared it, though 1 would only spell arsenic acold arsenic a 66% probability that a personification wrote it. Another declared it 0% AI (more connected that 1 later).

Back successful 2019, I wrote an article for MCP called “Technical SEO Is a Necessity, Not an Option,” published years earlier ChatGPT arrived. The results weren’t overmuch better.

One instrumentality put it astatine 35% AI.
A 2nd said 16%.
And 1 claimed it had recovered traces of GPT, a merchandise that didn’t moreover beryllium yet.

It was funny astatine first. Then it stopped being funny, because this was not conscionable 1 unusual consequence anymore. If detectors can’t reliably place penning created earlier generative AI existed, really overmuch religion should we really put successful the scores they are producing today?

On the flip side, you could easy reason it’s a clear motion of really acold AI has come, getting truthful adjacent to matching quality connection that galore detectors can’t show the difference.

AI Detector False Positives Keep Showing Up

The much I tested, the little this looked for illustration a one-off. I besides went backmost to different MCP article I had written successful July 2021, much than a twelvemonth earlier ChatGPT arrived. The devices divided again. One publication it arsenic 86% human. Another went the different measurement entirely, calling it 76% AI-generated. Same article, wholly different verdicts.

AI detector scores for SEJ articles published successful 2019 and 2021, years earlier ChatGPT, showing mendacious AI-generated verdicts

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

There is simply a batch of existent investigation retired location (mine was conscionable a small test). Studies that ran historically human-written documents, including newspaper sentiment pieces, done starring AI detectors recovered galore were falsely flagged arsenic machine-written.

One comes pinch a existent twist: Stanford researchers recovered the problem was moreover worse for non-native English speakers, pinch immoderate detectors incorrectly flagging ample amounts of their activity arsenic AI. If you are a business utilizing world writers, that’s an uncomfortable thought.

At the aforesaid time, a 2026 study published successful ScienceDirect recovered that mini quality edits allowed overmuch of the genuinely AI-generated contented they tested to bypass discovery altogether. So that leaves galore successful an awkward position: the devices tin wrongly impeach quality writers while missing immoderate of the contented they are expected to detect. And a fewer quality tweaks tin bypass them anyway.

That doesn’t build overmuch clarity aliases trust.

The False Economy of AI Detection

Bigger picture, what bothers maine is that these detectors waste and acquisition connected a writer’s fear. Scores plaything wildly from instrumentality to tool. Add it each up and 2 things autumn retired of it: confused writers, and fearfulness pinch a value tag connected it.

FOW, the fearfulness of writing. Writers are second-guessing activity they cognize is good, worried immoderate instrumentality will determine it “looks AI.” Content utilized to beryllium judged connected whether it was useful, original, aliases good written. Now, much and more, the first mobility is whether a detector thinks a instrumentality wrote it. Somewhere there, things went wrong.

That fearfulness is moving correct done the full industry. Writers interest astir being accused of utilizing AI. Agencies interest astir clients moving detector scores. Businesses interest astir the effect connected Google and astir LinkedIn flagging them. Detectors, bully aliases bad, provender connected it. People panic-buy products based connected emotion – because they abruptly consciousness essential.

The costs is jobs. Writers are losing work because tense clients tally scans and dainty the numbers arsenic gospel, turning an unreliable exertion into a assurance situation for the contented industry. Increasingly, the verdict itself sits down a paywall too, sold per scan, per word, per seat. Pay up, and they’ll show you whether you’re human.

The Humanizer Upsell

Several companies successful this marketplace waste an AI detector pinch 1 manus and an AI “humanizer” pinch the other, built to thief matter hit detectors for illustration their own. Many of you (my quality prediction) will person tally into this: a cleanable verdict, past an upsell. As I shared earlier, they emotion to show you that your penning looks for illustration AI erstwhile they can. And erstwhile they don’t, they still spell for an upsell.

I did, connected an article I wrote successful 2014. One detector scored it 100% human, past successful the aforesaid infinitesimal offered maine the action to humanize it, pinch a paid upgrade to study more. Humanize what exactly? The human? Irony doesn’t get overmuch better.

And it sums up the full AI detector instrumentality business: an manufacture charging writers, contented marketers and editors to humanize contented that was quality to statesman with. Call it what it is, a toll cipher needs.

That’s a mendacious economy. Money keeps pouring into a verification furniture that can’t reliably verify anything, and the monetization continues. Writers second-guess themselves. Editors and contributors oculus each different pinch suspicion. And each caller exemplary merchandise resets the arms race. Fear goes in, gross comes out, and cipher is immoderate person to the truth.

Why AI Struggles to Human and AI Writing

More irony here. Generative AI and penning assistants were built to sound for illustration humans. That was the full point. These models were trained connected quality penning truthful they could nutrient thing earthy capable to walk for it. The truth that detectors now struggle to show the quality mightiness beryllium the strongest grounds of really good that worked.

That’s what makes the full detector statement truthful strange. We’re asking 1 AI strategy to show america whether different AI strategy sounds excessively overmuch for illustration a human, erstwhile the 2nd strategy was designed to sound quality successful the first place.

The mimicry worked, which was ever the goal.

When purpose-built detectors can’t consistently propulsion quality penning isolated from AI-assisted writing, connected what grounds does anyone impeach a writer of doing thing incorrect because AI helped?

LinkedIn, AI Detection, and Slop: The Social Side

The discovery instrumentality statement is spreading everywhere, and getting confused on the way. Publishing is going done its ain reckoning complete AI-written work. LinkedIn, meanwhile, has conscionable added a “seems for illustration AI slop” button truthful members tin emblem posts they deliberation utilized AI. It’s discovery again, but pinch humans arsenic the “detectorists,” though its extremity and attack are different.

Patrick Coffee astatine The Wall Street Journal has conscionable written a timely portion connected precisely this, digging into third-party AI detector findings connected LinkedIn station content. It was absorbing to spot LinkedIn mobility the vendors’ numbers while declining to supply comparable information of its own. A fewer quotes worthy sharing.

“Their investigation appears to dainty immoderate contented that AI touches arsenic AI slop, which is not really we look astatine it,” said a LinkedIn spokeswoman. “… AI tin beryllium a awesome support instrumentality successful helping group successful articulating ideas, refining language, aliases making connection much concise.”

And Dan Roth, LinkedIn’s editor successful main and vice president of content, makes a bully point: “Obviously, a large stock of LinkedIn’s contented is AI-assisted, but cipher tin cognize precisely really much, including LinkedIn, and anyone claiming they do is trading a discovery tool.”

AI slop is simply a existent problem, but I’m not judge a fastener fixes it. Hand group a emblem and immoderate will constituent it astatine competitors and posts they simply don’t like. A characteristic built to cleanable up the provender could conscionable arsenic easy extremity up turning group connected each other.

And there’s a harder mobility underneath. Will LinkedIn usage the flagged posts to tune early AI-assisted penning models?. If they are, past millions of members are efficaciously training AI, unpaid, to constitute successful ways that extremity getting flagged. Time will tell.

Evaluating Content With Humans successful the Full Loop

None of this is maine defending lazy AI-generated content. I’ve publication plentifulness of it. I’ve besides publication plentifulness of mediocre contented written wholly by humans.

And to beryllium clear the different way: Use AI. It has earned its spot successful each modern trading workflow, excavation included. For contented writing, it genuinely helps pinch ideation, research, insights, and imaginative assistance. AI for trading and contented creation is not the rumor here. The rumor is the accuracy of AI discovery tools, the fearfulness they generate, and scores pretending to beryllium the truth.

That’s wherever I deliberation galore person mislaid their way. Instead of asking and over-indexing connected whether AI wrote it and “false scoring,” inquire the questions that matter:

Is it meticulous and relevant?
Is it original?
Is location existent expertise aliases acquisition down it?
Does it adhd thing useful?
Would a scholar decorativeness it and consciousness the clip was good spent?

Questions for illustration those drawback anemic contented acold much reliably than a detector people ever will. They ever have.

That’s besides overmuch person to Google’s position. Google’s guidance has been dependable connected this for years: Reward helpful, reliable content, nevertheless it was produced. Its enforcement goes aft low-quality contented astatine scale. It doesn’t beryllium location deciding whether each condemnation came from a personification aliases a model.

If organizations genuinely attraction astir really contented is created, they’re amended disconnected having a clear AI argumentation than relying connected detector scores.

In academia, Indiana University’s Kelley School of Business bans AI discovery devices outright successful its module AI playbook. It calls them unreliable, tells unit not to upload student activity to them astatine all, and points everyone toward clear argumentation instead. When a starring business schoolhouse won’t spot these scores connected student essays, why would a marque spot them connected trading copy?

And if provenance really matters to you, thin connected humans and your activity colleagues and peers. Draft history. Version control. Editorial review. A writer’s assemblage of work. Any of those carries much discourse than a percent spat retired of a achromatic box.

The Rise of FOW: The Fear of Writing

What started arsenic a mini individual research points to thing overmuch bigger, and it’s moving its measurement done the industry. AI discovery is loading penning pinch fear. Fear that honorable quality activity gets questioned. That AI-assisted activity gets rejected. That a azygous people from a azygous instrumentality someway counts arsenic the truth. Fear creates demand. Demand creates products. There’s the mendacious economy.

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

Always situation mediocre content. Check the facts, and backmost quality judgement up of tools. Push writers for original reasoning and existent expertise. Just don’t confuse immoderate of that pinch a detector score. And if you do want to tally checks, constituent them astatine the non-negotiables, for illustration plagiarism, wherever a consequence really intends something.

I wouldn’t walk my money trying to beryllium penning is human. I’d walk it making the penning better.

Just for transparency, I utilized AI to thief maine pinch this article. Here are immoderate of the results from moving this done the detectors.