I Wrote This Article. It Is “100% Human Written.” I’m Still Paranoid You’ll Think Otherwise.

Users A.I. Detectors Are Supposed to Make Our Writing Better. It Might Be Doing the Opposite. If you’ve been avoiding the em dash and words like delve and tapestry, you’re not alone. By Enter your email to receive alerts for this author. Sign in or create an account to better manage your email preferences. Unsubscribe from email alerts Are you sure you want to unsubscribe from email alerts for Alex Kirshner? Aug 18, 202610:30 AM Photo illustration by Slate. Photo by Getty Images Plus. Sign up for the Slatest to get the most insightful analysis, criticism, and advice out there, delivered to your inbox daily. I wrote this article. I am sure of it not because I remember writing this thing in fits and starts throughout Saturday, but because an A.I. detector called Pangram blessed these paragraphs with a “100% Human Written” score. Thank heavens. An A.I. detector’s value proposition is simple enough. For a teacher, it ensures that students’ work is their own. For the rest of us, it functions to preserve our last shreds of shared reality at a time of rapidly developing make-stuff-up machines. However, the accuracy of detectors is the subject of reasonable debate. University of Chicago researchers and a New York Times journalist have found Pangram, one of the buzziest and most popular A.I. detectors, to operate excellently. The company has made claims about its accuracy that strain my own credulity; a Pangram blog post last year stated that it had a “1 in 10,000 false positive rate.” I’ve found a far higher error rate in my own unscientific testing. Tech writer Alex Heath, who uses A.I. in a back-and-forth process with his own tweaking, explained recently that Pangram judged his text to be 100 percent human. I suspect that the detector would flag the most uninvolved “Write this for me” slop but not the text that involves a lot of human molding. As writer Freddie deBoer argues, Pangram doesn’t seem broken but does seem easy to break. Whatever the future holds for the accuracy of A.I. detectors, I am unsettled by what both the A.I. writing and the detectors are doing to me right now. I worry about how A.I. writing and efforts to sniff it out are changing fully human writing. I write all week, every week, and can feel this story’s arsonists (LLMs, particularly when used the wrong way) and firefighters (A.I. detectors) weighing on my process in negative ways. Like many journalists, my writing style comes from the books, news articles, blogs, and (yes) social media posts that have latched on to my brain over the years. A.I. writing is now in my water supply, whether I can identify each new gulp or not. It feels inevitable that I’ve started to internalize its ticks, even the most infamous ones, like incessant em-dashing and overreliance on the “It’s not x but y” structure. For one thing, I fear this will make me a worse writer, dragging a talent that was good enough to get me hired here (and elsewhere!) back toward the A.I.-generated median. Scarier yet, it’s possible that someone—or, heaven forbid, some A.I. detector—might mistake my text for ChatGPT’s. I have started to self-police against this outcome. A few times lately, I’ve caught myself modifying my own work to sound “less like A.I.” In the process, I’ve probably picked words I wouldn’t have settled on otherwise. It feels dystopian—and I doubt I’ve succeeded. One thing LLMs are great at is rapidly finding little acorns within huge blocks of text. When I set Claude’s Opus 5 model loose on about 64,000 words of my Slate writing from last September to this summer, it identified that my em-dash usage had declined by about 50 percent from the first few months to the last few. That makes sense; em dashes are easy to spot, and I’ve consciously used them less as the internet has flooded with them. But my use of antithetical framing had gone up even as I’d told myself I would stay away from it. Humiliatingly, in this story about reasonable uses of A.I., I leaned on it twice. “My anxiety about A.I. is not that it will make my job obsolete but that it will make me a little dumber,” I wrote. I was correct, I guess; here I am, being paid to write about feeling like a rube. The machine patted me on the head by noting that I had avoided words like profound and tapestry and only once indulged in telling all of you “Here’s the thing.” For now I am a (possibly naive) believer that A.I. will prove useful to writers and other creative media professionals despite the serious threats it poses to us. My mindset on this shifts often, but the one thing I’ll never want is for people to think that the machine made what I actually made. A time or two over the past few weeks, I’ve caught myself in a cursed loop: Draft a story with my human brain and human fingers, then put it into Pangram before it appears anywhere on the internet. Usually, the model concurs that I did it myself. In a few instances, it determined that single-digit percentages of my writing were either A.I.-generated or A.I.-assisted. According to Pangram, A.I.-assisted writing “is text a person wrote partly with help from an AI tool—for example, editing or rewriting parts.” I think this is where the detector is most likely to totally miss an LLM’s help. Earlier this month, Pangram flagged a small passage from a draft I had submitted to an editor at one of the other publications where I freelance. I forget if the story came up as 5 percent A.I.-assisted, 9 percent, or whatever. But I remember rewriting several sentences of my own work in the hopes of satisfying this new integrity god. I was worried that my editor, a cool guy who’s known me for years, might be on edge about this sort of thing. What if he put it through Pangram and wondered if he’d gotten me all wrong? That wouldn’t be good for business. I ran out of free credits before I could get the story up to a “100% Human” designation, then attached it to an email and pressed Send. I sent the editor a six-text blast explaining myself, as if I were getting ready for my own criminal trial. We were all good, he said. It was at that point that I had my only useful epiphany on this topic. This isn’t a problem to work out in the pages of Slate—it’s a problem for me to work out in therapy. Artificial Intelligence Media