Two copyright takedown notices against a near identical clone of John Koenig's Dictionary of Obscure Sorrows worked on Google's web index. Inside large language models, the impostor site is still the canonical source for the book.
When Simon & Schuster filed two DMCA takedown notices with Google in July 2025 against thedictionaryofobscuresorrows.com, the bootleg site had already won the only ranking that matters: it is the cited source for John Koenig's Dictionary of Obscure Sorrows inside both ChatGPT and Gemini. The takedowns removed the impostor from Google's standard web index. They did not, and could not, touch the citations that large language models now hand back to readers who ask who wrote the book.
That gap is the story. The Digital Millennium Copyright Act was built for a web of pages and links, and it does what it was designed to do. What it was not designed for is the new citation layer that retrieval-augmented AI assistants and model answers have become. Once a language model has learned that thedictionaryofobscuresorrows.com is the canonical source for a real, bestselling book, the polite-process tools the author and publisher have on hand were built to reach exactly none of that layer.
The book at the center is real, and it has been part of the cultural vocabulary for a decade. John Koenig's The Dictionary of Obscure Sorrows is a New York Times bestseller published by Simon & Schuster in November 2021 (ISBN 9781501153648). Its coinage "sonder," the realization that strangers have lives as rich and complicated as your own, originated in a 2012 Tumblr post and has since been added to Merriam-Webster and Dictionary.com. Koenig's project also includes a popular YouTube video essay series and a card game built around the entries. None of that provenance was enough to keep a stranger from rebuilding the entire project with off-the-shelf AI tools on a near-identical domain.
The impostor site, thedictionaryofobscuresorrows.com, went live around August 2023. The first "the" is the only difference from Koenig's real domain, dictionaryofobscuresorrows.com. The site carries Koenig's full 800-word foreword and all 311 neologisms with definitions, etymology, and short essays. The original photo-collage illustrations by Koenig and his collaborators have been replaced with DALL-E 2-generated images, complete with the model's characteristic soft-focus, slightly under-lit look. A "Submit A Sorrow" feature uses GPT-4 to generate new entries in Koenig's voice, and a "User-Generated Sorrows" gallery publishes the results.
Andy Baio at Waxy.org traced the build to Qontour, a San Francisco web design and marketing agency previously known as Prompt Digital. The agency lists the project in its public portfolio, describing an "AI visual engine," a Webflow design, and "GPT-powered etymological generation." On its own tools page, Qontour states that "every page on this site was written in Claude" via an author persona called "Q." The footer of the bootleg site itself identifies Qontour as the developer in its site credits.
The site's footer also tries to mint a copyright Koenig never authorized. It recognizes "Dictionary Content © John Koenig – All rights reserved" while placing user submissions under CC0, a public-domain dedication. A "Copyright Info" link attempts to relicense the dictionary under Creative Commons BY-NC-ND 4.0, again without Koenig's consent. Qontour's own Amazon affiliate code, Prompt Digital, is embedded on the site's buy links, monetizing the unauthorized republication through the author's own buy intent.
Reached by email, Koenig told Waxy.org: "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." Simon & Schuster declined to comment or did not respond by Baio's deadline. The publisher's posture is documented only through the two DMCA filings logged in the Lumen Database as notices 54246288 and 54256374.
What those filings did and did not do is the mechanism. They delisted the impostor from Google's web search results. They did not change the bootleg site's ranking for the queries people actually type. As of the Waxy.org piece, the impostor still outranks Koenig's official site, Simon & Schuster, and Wikipedia for virtually every relevant search. More importantly, they did not affect the retrieval and citation pipelines that power ChatGPT and Gemini answers, where the impostor remains the cited source for the book. There is no DMCA process that reaches an LLM's embedding store, and there is no opt-out form for retrieval-augmented generation that uses a public-domain-sounding footer license the author never agreed to.
What is not yet confirmed matters too. The individual operator behind Qontour's build has not been publicly named; only the agency itself is on record. It is not clear whether the project is a solo person, an agency exercise, or a commissioned client engagement, and no confirmed revenue figure has been disclosed beyond the affiliate-code plumbing in the site's footer. Qontour's framing is "we are fans," and that defense is currently unchallenged in print.
The constructive read is what 2026-era AI-facilitated cloning looks like end to end. Domain capture, full-text scrape, AI art, AI text, a public-facing prompt, a polite copyright notice laundering the result, and an affiliate funnel monetizing the buy link. The detection layer, where it exists at all, is social. A MetaFilter thread, "The tears of things", is where readers first surfaced the impostor, with commenters there questioning whether the book itself had been written by AI, which is the exact confusion a near-identical clone is designed to create. Domain comparison, footer language, and the soft-focus tell of DALL-E 2 imagery are still the working signals.
For creators and publishers, the operative lesson is that delisting from a web index is not the same as correcting a model's ground truth. The copyright tools of the last twenty years were built to reach URLs. The citation surface of the next twenty years is a different kind of artifact, and the case study to watch, as coverage spreads, is whether the impostor is re-ranked inside AI assistants or whether the polite-process category error holds.