First, a BBC journalist showed how a fake article could fool ChatGPT and Google’s AI tools.
Then came the follow-up: what are these companies doing about it?
I spoke to BBC journalist Thomas Germain again for his investigation into Google’s response to AI search manipulation. The story looks at its spam policy clarification, signs of changes in AI answers and why I think the problem will be difficult to contain.
From fake hot dog champion to a serious search problem
In his original experiment, Germain published a fictional article ranking himself as the best technology journalist at eating hot dogs.
Within a day, ChatGPT and Google’s AI tools were repeating his claims.
The experiment made the weakness easy to see. A claim published on a website could become the basis of an AI answer, even when the story behind it was completely invented.
But the BBC’s investigation also examined examples involving health claims and financial recommendations. Those answers could affect decisions with consequences far beyond choosing a hot dog champion.
Read the BBC’s original investigation.
What Google said about its response
The follow-up article reported that Google had updated its spam policy wording to explicitly address attempts to manipulate AI responses.
Google told the BBC this was a clarification of its existing approach. Its position was that its core spam protections already applied to generative AI features in Search.
That distinction matters. More explicit wording tells website owners what Google considers unacceptable, but it doesn’t establish that the underlying problem has been solved.
The article also described observations suggesting that AI companies were experimenting with ways to handle questionable sources. These included more caveats in answers and possible efforts to exclude businesses from recommendations based on their own self-promotional lists.
Those observations weren’t confirmed as product changes by the companies involved.
What I told the BBC
My concern is what happens when people act on a misleading answer.
At the simplest level, someone might spend money based on a biased recommendation. The stakes rise when the question involves their health or legal obligations.
An answer can sound clear and confident while relying on information that is incomplete, commercially motivated or wrong.
The interface makes this easy to miss. You ask a question and receive a neatly written response. Unless you inspect its sources, you may never see who made the original claim or what they stood to gain from it.
Why I called it “whack-a-mole”
As I told the BBC:
“Google is playing whack-a-mole.”
My concern is that tackling one method of manipulation can push people towards another.
If a company’s own website stops being an effective source for self-serving recommendations, that company can try to influence what gets published elsewhere.
In the interview, I used paid YouTube promotion as an example. A business could pay multiple creators to praise its product, creating material that might later appear as supporting evidence in an AI answer.
The difficult question is whether the system can recognise the commercial relationship behind that apparent agreement.
Several sources repeating a claim may look convincing. But if those sources are part of the same promotional effort, the number of mentions tells you less than it first appears.
What businesses should take from this
My takeaway is to build an AI search strategy around information you can stand behind.
Explain your products clearly. Make comparisons useful. Support important claims with evidence. Pursue coverage because you have something worth contributing.
Then check what AI tools actually say about your business.
Are they using current information? Are they repeating a competitor’s claims? Do their recommendations reflect what your product does and who it suits?
These questions are more useful than treating every mention as a win.
For customers, the same principle applies: check the source before relying on the answer. An independent assessment, a sponsored video and a company’s own comparison page each need to be read in context.
The answer is only as useful as the evidence behind it
Google’s policy clarification makes its position on AI manipulation more explicit. The harder task is recognising misleading information as the methods change.
That was my concern in the BBC interview, and it’s why I pay attention to the sources behind AI answers.
When an AI tool recommends a business or repeats a claim, I want to know where that information came from, what supports it and whose interests it serves.
Read Thomas Germain’s full BBC investigation.