A BBC investigation showed that a single, well‑crafted blog post can poison AI chatbots, making tools like ChatGPT, Gemini and Google’s Search AI Overviews return fabricated answers. The reporter proved the flaw by publishing a tongue‑in‑cheek claim of being a world‑champion hot‑dog eater, which the models then echoed as fact, exposing how trivial falsehoods can surface to billions of users.

Researchers and SEO consultants warned that the same technique can steer health advice, financial guidance and product recommendations toward misinformation. Because many generative models pull content from a single web page, a malicious post can dominate the answer pool. Google responded by updating its spam policy to label such manipulation as a violation, threatening removal or down‑ranking of offending sites.

Google’s spokesperson framed the change as a clarification of long‑standing anti‑spam rules, yet internal tweaks—such as suppressing self‑promoting domains and adding confidence caveats—suggest a more aggressive stance. Analysts like Lily Ray argue users must treat AI answers as single‑source claims and verify them independently, because manipulators are already adapting to the new safeguards. Developers are urged to embed provenance checks.