A BBC journalist wanted to prove a point. He published a single article on his personal website claiming he was a world-champion competitive hot-dog eater. Twenty minutes later, both Google’s AI and ChatGPT were telling the public it was true.
The joke was dumb. The problem is serious.
The BBC investigation
Thomas Germain, a senior technology journalist at the BBC, set out to demonstrate how easily AI chatbots and search engines can be manipulated. His experiment was simple: create a web page claiming something false, wait for AI to find it, and watch it get repeated as fact.
It worked almost immediately. Google’s AI Overviews — the synthesised answers that appear at the top of search results and reach 2.5 billion people per month — started presenting his fabricated hot-dog championship as factual information. ChatGPT did the same.
But Germain’s stunt was the harmless version. His investigation uncovered the same technique being used systemically to manipulate information about health supplements, retirement finances, and political topics — areas where bad information can cause real harm.
“You should assume that you’re being manipulated until they have better systems in place,” warns Lily Ray, founder of the SEO and AI search consultancy Algorythmic. “We’re moving towards this ‘one true answer’ world. Before, Google would give you 10 blue links and you would kind of do your own research. But AI just gives you one answer. It becomes so easy to just take things at face value.”
How it works
The manipulation exploits how AI systems gather information. When you ask a chatbot a question, it often searches the web for an answer. The problem: it tends to surface information from a single source rather than cross-referencing multiple perspectives.
This means anyone who publishes a well-crafted blog post can effectively inject false information into the answers that billions of people receive from the world’s most powerful AI systems. It’s SEO poisoning on steroids — except instead of gaming a page rank algorithm, you’re gaming the source material for an AI that presents its answers with the confidence of an encyclopedia.
Google responds
After the BBC investigation and similar findings from researchers, Google updated its spam policies to explicitly cover AI Overviews and AI Mode responses. The policy update says attempts to manipulate AI-generated answers are now violations of Google’s spam rules, with penalties including removal from or downranking in search results.
A Google spokesperson insisted this was “a clarification, not any change in approach,” saying the company has “long applied our core anti-spam policies and protections to our generative AI Search features.”
But if the policies were already in place, why did it take a BBC investigation to highlight the problem? And if the defences were working, how did a journalist game the system in 20 minutes with a single blog post?
Signs of progress
Ray and other SEO experts say they’ve noticed some changes in recent months. Google and ChatGPT appear to be quietly removing companies from AI answers when they suspect self-promotion. “If you publish a list where you say you’re the greatest hot-dog-eater, they’re not going to include your name,” says Ray. “They might still cite your article, but you’re going to be removed from consideration.”
ChatGPT and Claude have started explicitly telling users they’re trying to filter out spam in responses. Google has added more caveats and recommendations to check third-party reviews when answering commercial queries.
These are defensive patches, not structural fixes. The fundamental problem remains: AI systems that present synthesised answers with authoritative confidence, drawing from a web that anyone can edit.
Why it matters
This lands the same week Google unveiled the biggest overhaul of Search in 25 years — a shift toward conversational, AI-powered answers, information agents, and generative UI that builds custom experiences instead of showing links. The more Google replaces “ten blue links” with “one synthesised answer,” the more impactful manipulation becomes.
When search gave you ten links, you could compare, evaluate, and choose. When AI gives you one answer, you trust it or you don’t — and most people do. A single manipulated source in that answer has more power than a hundred SEO-optimised pages ever did.
Harpreet Chatha, who runs the SEO consultancy Harps Digital, puts it plainly: “At the most basic level, the concern is the economic impact. At a more serious level, you might take medical advice that makes you sicker than you were before. Legally, you might get bad information and do something that is not legal in your state or your country.”
The NZ angle
Google’s AI Overviews are rolling out globally, including in New Zealand. With a small media ecosystem and limited local fact-checking capacity, NZ users are particularly vulnerable to AI manipulation. When the AI surfaces a single source for “best mortgage rates NZ” or “health supplements NZ,” and that source has been gamed, Kiwis have fewer alternative sources to cross-reference.
As AI search becomes the default way people access information, the integrity of those answers becomes a public infrastructure question, not just a Google product question. And right now, that infrastructure has a 20-minute exploit.
Related: Google Search’s biggest AI overhaul in 25 years · Google AI bots squeezing small businesses · Google AI Overviews publisher death spiral · AI deepfake scam calls fraud surge
