QWE AI Academy In June 2026, a marketing agency quietly republished an entire book – all 311 entries of John Koenig’s Dictionary of Obscure Sorrows – swapped out the human art for DALL-E 2 pictures, bolted on a GPT-4 word generator, and watched its clone outrank the original on Google. The case (broken by Andy Baio at Waxy.org) is the cleanest example yet of what AI-assisted plagiarism actually looks like in 2026: not a sloppy copy-paste, but a polished SEO-optimized site that beats the source.
If you write anything online, you should care. The same recipe works on a blog, a portfolio, a documentation site. This tutorial walks through the obscure sorrows plagiarism playbook in reverse – how to spot an AI-cloned site in 60 seconds, and what actually gets it taken down.
The takeaway, upfront
Three things matter, in this order:
- Detection is fast. A unique sentence in Google quotes will find most clones in under a minute.
- De-indexing is not removal. Filing a DMCA only with Google leaves the infringing site running. You need to hit the host too.
- The clone usually wins on rank because it’s technically newer and cleaner. The original being ‘first’ counts for almost nothing.
What actually happened with the Dictionary of Obscure Sorrows
John Koenig has been writing the project since 2009 – first a blog, then a YouTube channel, then a Simon & Schuster book in 2021. A marketing agency – later identified as Qontour, formerly Prompt Digital, a San Francisco web design firm – republished the entirety of the book without permission, including its 800-word foreword and complete archive of all 311 neologisms with their definitions, etymologies, and short essays.
The original photo-collage illustrations were stripped out and replaced with DALL-E 2 images, complete with the artifacts typical of that model. Then they added a generator. A banner at the top invites visitors to ‘Generate your own words using AI,’ feeding descriptions to GPT-4 which produces new words and art into a ‘User-Generated Sorrows’ gallery.
Koenig found out from a stranger’s email. His reply: ‘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.’
That last line is the real lesson. A clone can look better than the source.
Method A vs. Method B: how to detect a clone
Two approaches dominate. They’re not equivalent.
| Approach | Cost | Speed | Catches AI rewrites? |
|---|---|---|---|
| Google quoted-string search | Free | Seconds | Only verbatim copies |
| Copyscape (free or Premium) | Free tier limited; Premium ~$0.05/search (as of mid-2026, check copyscape.com for current pricing) | ~1 minute | Better, but not bulletproof |
For a verbatim wholesale rip like the Koenig case, the Google method wins. You don’t need a paid tool when 311 entries were copied character-for-character. Copy a unique sentence from your content, search for it in quotes, and any other site appearing in the results is a candidate.
Copyscape becomes useful when you want continuous monitoring or when the text has been lightly rewritten. Copysentry, their automated service, scans the web on a schedule and emails you whenever new copies of your pages appear. There’s a catch – see edge cases below.
The detection walkthrough (the method that actually works)
Here’s the sequence I’d run if I suspected my own site was being cloned tomorrow.
Step 1: The 60-second sniff test
Pick a phrase from your content that is distinctive – long enough to be unique, weird enough that nobody else would coin it independently. For Koenig, any neologism plus its definition would do. Search Google with quotes around it:
"the frustration of photographing something amazing when thousands of identical photos already exist"If a domain you don’t own appears, click it. Then look for these AI-clone tells:
- Images with the smeared-text, six-fingered-hand artifacts of older diffusion models
- A ‘Generate with AI’ or ‘Submit your own’ feature on what should be a static reference site
- A domain that’s a near-duplicate of yours with a prefix or suffix tweak (the Koenig clone simply added ‘the’ to the front)
- A ‘Site Credits’ or footer attribution to an agency you’ve never heard of
Step 2: Confirm and document
Before you file anything, screenshot everything. URL, timestamps, copy of the duplicated text, side-by-side with your original. Standard DMCA filings require you to identify the infringing URLs, provide URLs of your original work, assert copyright ownership, and include a signed good-faith statement – missing any one of these is the most common reason takedowns get rejected.
Step 3: Find the host, not just the search ranking
This is the step most guides skip. Run the cloned domain through a WHOIS lookup. You’re after two pieces of information: the hosting provider and the domain registrar. These are who can pull the site down. Google can only hide it.
Step 4: File the takedown – in this order
- Hosting provider first (they can remove the site outright)
- Domain registrar second (in case the host is slow)
- Google & Bing DMCA forms last, to de-index the URLs from search
Order matters. As the Red Points DMCA guide (2026) explains: a search engine DMCA takedown de-indexes infringing URLs from Google or Bing search results – it does not remove content from the internet. Skip the host and the site lives on, just slightly harder to find. Google typically responds within a few weeks; Bing within 5-10 days.
Pro tip: When you file with Google, attach the URL of the book listing on Simon & Schuster or Amazon as your proof of ownership, not just the original site. A publisher record is harder for an automated reviewer to dispute than a self-hosted page.
Edge cases nobody warns you about
Three traps to know before you start.
The clone can outrank you because it’s newer
Koenig’s original lives on a Tumblr-era setup. The clone is a polished agency build on a fresh domain – Google reads that as a higher-quality result and ranks it accordingly. Being first online means nothing if the squatter has better Core Web Vitals and a newer technical foundation.
The AI generator on the clone makes this worse. Every ‘Submit a Sorrow’ submission becomes a new indexed page. Fresh content, daily – exactly what Google’s freshness signals reward. The original site, static for years, looks stale by comparison.
Copyscape’s free tier has a shared rate limit
If your free scan returns ‘limit reached’ on day one, it isn’t because you scanned too much. The Copyscape FAQ states the message can appear ‘even though you have not performed many scans on a certain site – this simply means another user has already exceeded the limit.’ A journalist running the same check before you can lock you out for the day. Workaround: paste a unique sentence into Copyscape’s text-search box instead of the URL field, or upgrade to Premium (around $0.05 per search as of mid-2026 – no shared cap).
The ‘AI-generated’ label is not a legal shield
The Qontour site argues, implicitly, that the GPT-4-generated entries are ‘new.’ That argument doesn’t help them with the 311 verbatim original entries. But it could complicate the takedown of the generator feature itself if you’re trying to argue stylistic derivation. Focus your DMCA on the verbatim text first – that’s the airtight claim.
Why this case matters beyond one book
The recipe Qontour used is reproducible by anyone with a weekend and $20 of API credits. Find a beloved indie creator. Scrape the canonical text. Generate cover art with a diffusion model. Bolt a GPT-4 toy onto the front page. Ship a slicker site than the original. Rank. Collect Amazon affiliate clicks.
The defense is boring but it works: distinctive phrases you can search for, alerts via Copysentry or a simple Google Alert, and a takedown habit that targets the host before the search engine.
FAQ
Is the cloned Obscure Sorrows site still up?
As of the original Waxy.org report (June 2026), yes. Status may have changed – check the URL before assuming current state.
Could I use Copyscape to protect a book that isn’t on the web yet?
Yes, but you need Premium. The paid tier removes the shared rate-limit problem of the free version and lets you monitor pages on a set schedule. If a competitor or AI training scraper already has a copy of your manuscript floating around before publication, Copysentry’s scheduled scans can surface it before launch day – which is when the SEO race actually starts.
Does using AI art on a site automatically make it plagiarism?
No. AI-generated images on their own aren’t infringing. The Koenig case is plagiarism because the underlying text – every definition, every essay, every etymology – was copied verbatim from a copyrighted book. The DALL-E images are evidence of bad faith and a tell that the site isn’t authoritative, but the legal claim rests on the words.
Your move
Open a new tab. Copy your most distinctive sentence – the weirder, the better. Wrap it in quotes. Search Google. If anything shows up that isn’t yours, you already know what step 2 is.