BBC reporter poisoned ChatGPT and Gemini with one blog post in 20 minutes; Google updated spam policies but experts say manipulators stay ahead.
Key Takeaways
A single well-crafted blog post on a personal site was enough to make ChatGPT, Gemini, and Google AI Overviews spread false claims the next day.
The same technique is being used commercially to manipulate AI answers on health supplements and retirement finance, not just stunts.
Google updated its spam policy to explicitly ban AI response manipulation; violations risk removal or downranking from search entirely.
Observed mitigations include quietly stripping self-promoting sources from AI answers, adding low-confidence labels, and citing third-party reviews for purchase queries.
SEO experts warn the tactic is already shifting to YouTube influencer campaigns, which Google’s AI now cites, continuing the cycle.
Hacker News Comment Review
Commenters are skeptical Google’s spam policy enforcement will work here any better than it did against classic search spam; the whack-a-mole dynamic is widely expected to continue.
A technical hypothesis circulated: RLHF training that rewards accurate summarization may have overfit on verbatim source repetition, making models structurally prone to laundering single-source claims.
There is no consensus on a clean fix; curated training data is seen as cost-prohibitive, and source-count confidence signals are undermined by LLMs’ ability to generate corroborating text synthetically.
Notable Comments
@graemep: Drily notes Google is applying the same spam policies to AI that “worked” for search, implying low confidence in the outcome.
@tencentshill: Points to HubSpot and Semrush already selling AEO/AI optimization products, confirming this is an industrialized marketing category now.
