Independent author T.M. Kirk did not expect a design choice to turn into an argument about honesty. She shared a post online, and the comments filled up with people accusing her of using artificial intelligence. Nothing in the post had been flagged as AI-made. The problem, as far as her critics were concerned, was the font.
Kirk raised the episode on Instagram, where she explained that the post carried no AI label and that she had made it herself. That did not settle it. Readers had decided the lettering looked machine-made, and the design itself became the evidence. It is a strange place for a working author to find herself: defending not a sentence she wrote, but the shape of the letters on the page.
The incident says less about one post than about the mood around AI-made content right now. Tools such as ChatGPT and Claude have made it fast and cheap to produce graphics and adverts, and the internet is now full of them. Much of that output is low effort, churned out in bulk with little human judgement, and audiences have given it a name: AI slop. Merriam-Webster went as far as naming slop its word of the year for 2025. The resentment is real enough that people have started calling for boycotts of businesses that use it. With so much machine-made imagery around, people have started hunting for clues, and the search has moved well beyond the obvious mistakes. Early AI pictures gave themselves away with garbled words and extra fingers. Those errors are rarer now, so the hunt has moved to smaller details, and a familiar design choice can draw the same suspicion as a mangled hand.
What Makes a Font Look AI-Generated?
When an image generator draws lettering, it does not pick a font like a designer does. It produces the letters as pixels, copying patterns it has seen in its training data. That data is full of adverts, posters, banners and video thumbnails, so the lettering it produces tends to resemble the styles used in exactly those places: bold, polished and designed to grab attention. Small inconsistencies in spacing or in the shape of individual letters can give the text an uneven look.
Some of that lettering looks like Inter, the clean sans-serif face now common across websites and apps. The more talked-about example is the slanted brush lettering seen in dramatic adverts and flyers, which many people now compare to the display face Road Rage. It turns up so often in AI-made posters that readers treat it as a signature of the machine. Designer Ramin Nasibov picked up on the pattern in a November 2 post, rendering the words Comic Sans in that same brush style. Comic Sans was designed in 1994 to mimic informal handwriting, yet a style meant to look handmade has become, in many readers’ eyes, the mark of something made by a machine. Human designers choose these styles too, so the lettering alone tells a reader very little about how an image was made.
Creators Are Changing How They Work
The suspicion has already spread from pictures to words. The em dash, used by writers for years, became a supposed giveaway once people noticed how often chatbots use it, and the word delve suffered a similar fate. Some writers now write more casually on purpose, or leave small typos in place, just to sound human.
The stakes are higher than a few angry comments. In early November, Bloomberg reported that authors have already lost million-dollar book deals over AI accusations. The same month brought a sharper example from Spain, where El Pais ran a prize-winning novel through the Pangram detector, which reported 99 percent of its roughly 26,000 words as AI-generated. The publisher suspended presentations of the book while it reviewed the claim, and the author insisted he had written the novel without generative AI tools, gathering earlier drafts and notebooks to prove it. Whether the detector was right is still unestablished.
Faced with that climate, many creators now document their process as they go. Artists share early sketches and record time-lapse videos of a piece taking shape. The trouble is that even that evidence is no longer solid ground, because AI can now generate footage that imitates a time-lapse, so such a video proves less than it used to. Some designers have gone further and simply avoid typefaces that readers associate with AI, steering clear of styles that might invite an accusation. It narrows the choices available to people who did nothing wrong.
The deeper problem is that the clues are fading while the suspicion is not. Image generators are getting better at clean, consistent lettering, so the uneven spacing and odd letter shapes that once gave them away are becoming less common. What may last longer is the stigma attached to particular styles, which falls hardest on small creators. Kirk’s post was ordinary work, yet it drew an accusation she had to answer in public. A font is a choice, not a confession, and treating it as proof of machine authorship punishes the very people who still do the work by hand.
