Andy Baio stumbled across it the way many do these days. A link on MetaFilter pointed to what looked like an official new home for The Dictionary of Obscure Sorrows. The site appeared polished. Professional. It carried the full text of John Koenig’s 2021 book. It even offered a tool to generate fresh sorrows with artificial intelligence.
But something felt wrong. The original project lives at dictionaryofobscuresorrows.com. This version sat at thedictionaryofobscuresorrows.com. One extra article. One big difference.
Koenig launched the endeavor on Tumblr in 2009. He coined terms for emotions that lack names. Sonder. The realization that every stranger possesses a life as intricate as your own. Anemoia. Nostalgia for a time or place you never experienced. Vellichor. The peculiar melancholy of used bookstores. Monachopsis. That nagging sense of not quite fitting in.
Those words traveled. Dictionary.com and Merriam-Webster eventually added sonder. Businesses adopted the term. A music act, a venture firm, even a bar near Baio’s home. The project grew into video essays on YouTube. Then Simon & Schuster published the physical book in November 2021. It became a New York Times bestseller.
Koenig described his goal simply. The work aims to shine a light on the fundamental strangeness of being a human being. He never authorized the slick new website. When Baio asked, Koenig replied within the hour. “Yeah man, I had nothing to do with it. Don’t know what to think or do about that, as the site is pretty slick. Nicer than my own, really.”
The operator revealed itself in the footer. Qontour, a San Francisco marketing and web design agency formerly known as Prompt Digital. The firm built the site in Webflow. It swapped Koenig’s careful photo-collage illustrations for images created by DALL-E 2. Those pictures show the familiar flaws. Warped hands. Strange artifacts. Generic dreamlike haze.
Visitors can submit descriptions of feelings. GPT-4 then invents a word, etymology, and definition. Those entries join a public gallery under a CC Zero license. The agency also embedded its own Amazon affiliate links. Any purchase through the site funnels a commission back to Qontour.
On its portfolio page the agency called itself a fan. It boasted about creating an interactive platform, an AI-powered image library, and a submission feature. The project supposedly consolidated scattered content into one destination. Yet the site now dominates Google results. Search for “dictionary of obscure sorrows,” for “sonder,” for “anemoia,” even for Koenig’s own name. The unauthorized version appears first. The original creator’s site trails behind.
Baio calls this wholesale plagiarism. Not the subtle laundering of text through large language models that often evades detection. This copies an entire living author’s book. It replaces the art. Adds generative tools. Monetizes the traffic. And then ranks higher than the genuine article almost everywhere. A flagrant move. One that treats copyright as optional when the technology looks impressive enough.
The footer attempts to cover legal bases. It notes that dictionary content belongs to Koenig with all rights reserved. User-generated material falls into the public domain. But copyright does not work that way. Publishing someone else’s complete work to showcase your design skills requires permission. Re-licensing it under Creative Commons when you hold no rights changes nothing.
This episode fits a larger pattern. Publishers worry about manuscripts assembled with large language models. A recent nonfiction title on truth and artificial intelligence contained multiple fabricated quotes attributed to real people. Author Steven Rosenbaum admitted using ChatGPT and Claude for research, writing, and editing. The New York Times exposed the issue in May. Rosenbaum later took responsibility and promised corrections.
Newspapers have suffered similar embarrassments. The Chicago Sun-Times and Philadelphia Inquirer published AI-generated summer reading lists that recommended books which do not exist. Both outlets apologized after readers noticed. The lists mixed plausible titles with complete fabrications.
Academic researchers document another dimension. A recent arXiv paper examined autonomous research agents that claim to produce novel ideas. Many simply paraphrase or borrow heavily from existing work. Plagiarism detectors often miss the sophisticated rewriting. The authors found more than 24 percent of sampled documents showed clear signs of unacknowledged copying.
Koenig’s case stands out because the source material is so distinctive. His neologisms carry emotional precision. They emerge from years of careful observation and craft. An AI generator cannot feel vellichor in a dusty bookstore. It cannot experience monachopsis at a crowded party. It can only remix descriptions scraped from the very text it now replicates.
Yet the public may not notice. The fake site looks authoritative. Its AI features feel interactive and modern. Search engines reward the cleaner design and faster load times. Users land there first, absorb the words as authentic, and move on. Koenig receives neither credit nor control over how his creations spread.
Baio’s reporting, published just days ago, has sparked fresh conversation across social platforms. Writers and artists see echoes of their own concerns. A marketing firm can seize a cultural project, dress it in generative imagery, and profit from the attention economy. Fans who discover the material through the impostor site may never realize they missed the human touch behind the original.
The incident raises uncomfortable questions about discovery online. When an unauthorized replica outranks its source, what happens to authorship? How many similar projects exist undetected? And how long before readers simply accept polished AI slop as the canonical version of creative work?
Koenig continues his project on the original Tumblr and YouTube channel. The book remains in print. But the internet now carries two versions. One crafted over more than a decade by a single careful voice. The other assembled quickly by a marketing agency eager to demonstrate its command of new tools. One seeks to illuminate human experience. The other seeks to rank.
So far the algorithm favors the latter. That choice tells its own story about what the internet values in 2026. Speed over substance. Traffic over truth. And the convenience of generated content over the harder labor of genuine creation.
WebProNews is an iEntry Publication