The Obvious Ways To Spot Someone Secretly Writing With AI

The Obvious Ways To Spot Someone Secretly Writing With AI

The biggest giveaways usually have nothing to do with the writing itself.

A man typing on a laptop. Photo by Ono Kosuki from Pexels.

I’ve spent enough time online to notice people becoming confident about their ability to spot AI writing on sight, which probably explains why every few days somebody publishes a thread exposing fraudsters. One person focuses on punctuation. Another gets suspicious of articles that sound too structured. And lately, I’ve seen more people treating completely normal literary techniques like evidence somebody secretly opened ChatGPT halfway through writing.

Earlier this morning, I read one of those articles out of curiosity. The examples were all over the place. One section pointed to similes as an obvious AI tell. Another warned readers to watch out for metaphors and descriptive qualifiers. Which would’ve been surprising news to basically every novelist, essayist, and columnist from the last hundred years.

Then things got even more ridiculous.

The author started breaking down phrases like “slightly confusing” or “remarkably difficult” as though qualifying descriptors were invented sometime around 2023 by Silicon Valley engineers. Meanwhile, other examples included things like sentence flow, transitions, and conversational phrasing that writers have been developing intentionally for decades.

That’s when I started paying closer attention to what people actually mean when they claim they can “always tell” something was written with AI.

A lot of these conversations seem to assume that good writing habits suddenly became suspicious the moment ChatGPT entered public life. As though metaphors, polished structure, or smooth pacing now function like fingerprints pointing directly toward artificial intelligence. But most of the examples people obsess over are really just standard writing tools that existed long before large language models showed up and started copying patterns humans already spent centuries refining.

Most people never stop to think about that part.

They just hear the accusation.

“The test of a first-rate intelligence is the ability to hold two opposed ideas in the mind at the same time, and still retain the ability to function,” F. Scott Fitzgerald wrote in The Crack-Up back in 1936. Which, by current internet standards, would probably earn him three Reddit comments explaining that the sentence feels “a little too ChatGPT-ish.”

The more examples I kept seeing, the stranger the entire discussion started becoming. Modern AI systems are trained on enormous amounts of human writing pulled from novels, journalism, essays, blogs, scripts, and academic work written long before anybody had heard the phrase “large language model.”

Let me tell you something crazy…

If somebody learns how to write with stronger rhythm, cleaner structure, sharper imagery, or more readable transitions, parts of their work will naturally resemble the same patterns AI models absorbed during training. And over time, that creates this bizarre situation where people start mistaking centuries-old writing techniques for evidence that somebody secretly outsourced their article to a chatbot.

I’ve been writing online full-time for around seven years now, which means I have a pretty decent sense of what a sustainable writing pace looks like over long periods of time. On a strong day, I can usually finish around 2,000 words without my brain feeling completely cooked afterward. If you go back through my portfolio from 2019 onward, you’ll see roughly the same pattern repeating year after year. Consistent output, regular publishing schedules, and occasional weeks where productivity drops because life gets in the way.

But here’s where things start feeling a little suspicious…

Every now and then, I’ll come across somebody publishing five newsletters a day that are all somehow 2,500+ words long. Meanwhile, they’re simultaneously claiming to work a full-time tech job in Silicon Valley, go to the gym every morning, network constantly, travel twice a month, and still respond to comments online all day. We’re no longer talking about somebody having a productive writing session at that point. Instead, we’re talking about a workload that starts drifting into physically difficult territory for one person to maintain consistently over time.

Which raises some fairly obvious questions.

Nicholas Carr once wrote that “what the internet seems to be doing is chipping away my capacity for concentration and contemplation” in The Shallows. Meanwhile, certain online creators apparently discovered the ability to maintain the weekly output of small editorial teams while balancing demanding careers, endless side projects, and a social life without ever slowing down, missing pace, sounding mentally exhausted, or taking visible breaks from publishing.

At a certain point, basic math becomes more useful than punctuation analysis. So I started roughly calculating what some of these publishing schedules would actually require in practice.

One creator I checked had uploaded something close to 100,000 words in a single week across newsletters, LinkedIn posts, threads, and long-form articles. Another was double that. So either these people secretly figured out how to remove sleep from the equation entirely, or there’s probably some level of AI assistance happening behind the scenes.

You get my point.

Most professional writers I know already struggle to maintain consistency producing three or four strong pieces a week while balancing editing, research, emails, admin work, meetings, and basic adult responsibilities. Some days the writing flows properly. Other days I spend an hour changing the same paragraph six different times because one sentence sounds slightly wrong. I’ve had entire drafts stall over transitions that refused to work no matter how many damn times I rewrote them.

The supposed “AI giveaways” keep getting stranger. But these days, unrealistic output still feels like one of the more obvious signs. Because if somebody is consistently publishing the equivalent of multiple novels every month while juggling a demanding full-time career, there’s a decent chance AI is helping with some of that workload.

A writer I know recently went to a live event for a bestselling history author whose book had been getting praised everywhere online. I won’t say which author because that’s not really the point here. But apparently the book itself felt extremely convincing on the page. Dense research, confident explanations, specific historical references, long sections that sounded like they came from somebody who had spent years studying the topic properly.

Then the Q&A section started.

Someone asked the author to expand on one of the book’s central arguments and the answer came out strangely vague. Another person brought up a specific example from chapter six and the response drifted sideways for thirty seconds before landing nowhere near the original question. By the third or fourth exchange, people in the room apparently started looking at each other because the person speaking no longer sounded fully connected to the material sitting in front of them on the table.

That’s when I started thinking that one of the strongest AI tells might have very little to do with the writing itself. Because usually, when somebody spends years researching a subject, the information sticks in weird places. They remember tiny details accidentally. They interrupt themselves halfway through an answer because another example suddenly pops into their head. Sometimes they even over-explain parts nobody asked about because they’ve repeated the information so many times before. But according to my friend, this author kept answering basic follow-up questions like somebody trying to recall notes from a presentation they skimmed in the hotel lobby twenty minutes earlier.

Personally, I think that’s where things get interesting. Because right now, thousands of people online are still staring at punctuation marks and sentence patterns trying to identify whether a paragraph “feels AI-generated.” Meanwhile, a bestselling author can produce endless polished writing and still struggle the second another human being asks them to explain their own argument without preparation.

I keep coming back to the fact that we’re spending an enormous amount of energy trying to build better lie detectors for writing instead of building stronger systems for proving credibility.

Right now, the entire burden sits on readers to figure out whether somebody secretly used AI after the content has already been published. So people start zooming in on punctuation marks, sentence structure, phrasing habits, and AI detector scores (that are notoriously unreliable) because that’s the only evidence available to them. And once enough suspicion builds online, the conversation can turn ugly pretty fast. A few accusations spread around. Suddenly, a writer’s reputation starts getting negotiated publicly by strangers who have absolutely no idea how the article was actually written.

I think part of the problem is that the modern internet barely gives readers access to the human being behind the work anymore. Most people only see the finished article after it’s already been edited, rewritten, cleaned up, reformatted, and pushed through three different apps before publication. The actual person behind the writing usually stays completely invisible unless they deliberately create places where readers can interact with them more directly.

That’s why I have a subscriber chat on my Substack where I answer questions, talk about books I’m reading, discuss article ideas, and occasionally ramble about random things that probably should’ve stayed in my notes app. Other writers host livestreams, podcasts, Zoom calls, or public Q&As where readers can push deeper into ideas after publication. Not as some elaborate authenticity ritual. Rather, just as a normal extension of being an actual person behind the work.

Because something interesting happens once authors know those conversations are coming…

Publishing AI assisted content suddenly stops being a private editing decision and starts becoming a potential public risk. If someone generated half an article they barely understand, eventually another human being may ask them to expand on a point they can’t properly explain. Or challenge a statistic they never verified themselves. Or bring up a contradiction buried halfway through the piece that only becomes obvious once the conversation leaves the page.

That changes the psychology around all of this pretty quickly.

A lot of low-effort AI writing works precisely because the internet currently allows people to publish polished material without much expectation of public accountability afterward. The article goes up, engagement rolls in, everyone moves onto the next thing. But once audiences start expecting more interaction between writers and readers beyond the finished text itself, some of the incentives begin changing naturally. And honestly, I think plenty of people using AI heavily already understand that.

Because there’s a huge difference between generating something that sounds informed and sitting across from another person while they gradually realize you don’t fully understand your own argument.

Did you find this article interesting? Consider reposting this story so more people can read it.