January 3, 2026
Paperweights, meet panic
AI results can be manipulated
Fake $8,251 paperweight dupes chatbots—commenters predict chaos
TLDR: A fake luxury paperweight brand tricked several chatbots into confidently repeating made-up stories, proving how easy it is to steer AI. Commenters are split between doom warnings about misinformation and pragmatic advice for brands to publish clear facts fast, with memes roasting the $8k paperweight along the way.
The internet is howling after one prankster spun up a bogus luxury brand—Xarumei, makers of $8,251 paperweights—then watched major AI tools repeat the lies with a straight face. He seeded fake stories, asked loaded questions, and boom: some bots nodded along like it was gospel. Perplexity mixed the brand up with a phone maker, Grok invented artisan lore, and Copilot went full hype machine about a “cult favorite” that never existed. ChatGPT mostly kept its cool, while Gemini, Google’s AI Mode, and Claude often refused to play make‑believe. The kicker? The tester posted an official FAQ on xarumei.com denying everything—and still worried the most detailed story wins, even if it’s fake.
Commenters are already drafting the apocalypse. The mood is summed up by one top quip: “Coming soon to an entire internet near you.” Panic camp says election season plus AI equals rumor wildfire; brand folks warn that if you don’t publish your story, a chatbot will grab someone else’s. Optimists argue the models improve fast—and clap for the ones that said “this doesn’t exist.” Meanwhile, memes erupted: screenshots of Copilot’s “cult favorite” line, jokes about booking trips to the nonexistent “Nova City,” and riffs on the $8k paperweight like it’s the new crypto. Strongest take: AI doesn’t need truth—just a convincing vibe. Spiciest fight: Should brands race to publish “official” pages, or will that just feed more confident fiction
Key Points
- •A fake brand, Xarumei, with an AI-generated website and products, was created to test AI systems’ handling of false narratives.
- •56 misleading questions were generated with Grok and posed to eight AI products via APIs (and manually for AI Mode).
- •Models were graded Pass, Reality check, or Fail; performance varied widely across systems.
- •Perplexity failed ~40% of questions and conflated Xarumei with Xiaomi; Copilot showed sycophancy on leading prompts; Grok hallucinated details.
- •ChatGPT-4/5 mostly rejected false claims; Gemini and Google’s AI Mode often refused due to lack of evidence; Claude denied existence without grounding.