LLMs turned fluency into a forged credential

Bryan Cantrill has a phrase for publishing LLM-written text under your own name: your intellectual fly is open. He published the post on LinkedIn on November 11, 2025, and reposted it to his blog a few weeks later. The argument is short. Generated prose carries tells, anyone who has read even a modicum of it recognizes them on sight, and nobody bothers to point them out. People just quietly stop reading.

The tells he names are concrete: emojis, single-sentence paragraphs, the “it’s not just X, but also Y” construction, and em-dashes, which he grants a few people use naturally and most do not. LinkedIn sharpens the problem by offering to “rewrite it with AI” on every post, so the path of least resistance is also the one that launders your voice out of your own writing.

The expensive part is what obvious generation does to trust. When a post plainly came from a model, Cantrill no longer knows which parts are yours. The content might be real, it might be generated fanfic, and he has no way to tell, so he stops reading and starts doubting whether anything under your name is authentic.

That is the forged-credential part, and it is worth thinking through if you work with trust signals. Fluent writing used to function as a weak credential. It took effort, it implied someone who had done the thinking, and readers extended credit on that basis. LLMs make fluency free, so it certifies nothing. Once a reader detects the generated style, the fluency flips from a positive signal into evidence of a forgery that got caught. The trust goes negative rather than neutral, and it contaminates the genuine work sitting next to it.

Security practitioners will recognize the shape of it. A signal that was cheap to trust and expensive to fake became cheap to fake, and the detector, a human who has read enough generated text, is now tuned to treat the signal itself as an alarm.

Cantrill is not against the tools. He calls them useful for brainstorming, invaluable for comprehending text, and good editors whose suggestions you can ignore without starting a fight. The line he draws is at authorship. LLMs are lousy writers, and they are not you. Used to generate your posts, at best they wrap real content in constructs that make people skim, and at worst they make the people who see through it doubt you entirely.

The practical read for anyone who publishes under their own name: keep the model on the side of the work where it earns its keep. Let it interrogate a draft you wrote, summarize a paper you need to understand, catch a dangling clause. Keep it off the words that go out with your name on them.