Google’s AI Search Can Be Tricked by Fake Web Pages

Google’s AI Overviews can summarize the web. The problem is that the web can lie.

Google is trying to police a growing problem inside AI-powered Search after a BBC investigation found that misleading online content can influence answers from Google’s AI Overviews and other major chatbots. Its updated spam policies now explicitly include attempts to manipulate generative AI responses.

The search giant says the policy language only clarifies rules it already enforced. But as AI takes up more space in Search, the old spam problem starts to look different.

A silly test with a serious flaw

AI chatbots do not always answer from memory. In some cases, they search the live web, grab information from online sources, and use that material to build a response.

BBC reporter Thomas Germain tested how fragile that process can be by publishing a fake article on his personal website claiming he was a “world-champion competitive hot-dog eater.” By the next day, he wrote, AI tools like ChatGPT, Gemini, and Google’s AI Overviews were repeating the claim.

No one is likely to make a life decision based on a fake title. Germain’s point was that the same method does not have to stay silly. 

According to the BBC, similar tactics have appeared around medical supplements and retirement-related financial information, where a planted or biased page can do much more damage.

A ‘one true answer world’

Traditional search spam usually had one job: to get a questionable page in front of users and make them click.

AI search, however, changes the payoff. A weak or planted claim can be folded into a direct answer, where users see fewer competing sources before deciding what to trust.

A single answer can feel more settled than the information behind it really is. Lily Ray, founder of the SEO and AI search consultancy Algorythmic, told the BBC that users should assume they are “being manipulated” until better systems are in place. 

“We’re moving towards this ‘one true answer’ world,” Ray said. “Before, Google would give you 10 blue links and you would kind of do your own research. But AI just gives you one answer.”

Harpreet Chatha, who runs the SEO consultancy Harps Digital, warned that the concern can move quickly from economic damage to health or legal harm. “At a more serious level, you might take medical advice that makes you sicker than you were before,” Chatha said. “Legally, you might get bad information and do something that is not legal in your state or your country.”

Google is making the target bigger

The AI manipulation problem gets harder as Google gives AI more of the search experience to run.

Google is expanding Search with follow-up questions, generative results, and information agents that can monitor the web and send synthesized updates. Links are still there, but users are being steered toward answers that are assembled for them.

Cracking down on abuse gets harder when the answer layer keeps expanding. A manipulated page no longer has to persuade a reader on its own; it may only need to become part of the answer Google’s AI delivers first.

Google used I/O to show Gemini spreading across nearly every major workflow, from search agents to app generation.