Google has faced criticism for some of the inaccurate, amusing, and bizarre answers provided by its AI Overviews in search results.
What Are AI Overviews?
Earlier this month, Google actively expanded the rollout of AI Overviews, but the reception has been mixed. For instance, one user seeking advice on getting cheese to stick to their pizza was advised to use glue, a suggestion pulled from an old Reddit post. Another user was told to eat “one small rock per day,” a recommendation sourced from The Onion.
If you haven’t encountered such odd responses or can’t replicate viral searches, it’s because Google is actively working to eliminate inaccuracies. A company spokesperson stated that Google is taking “swift action” to improve its systems using these examples.
“The vast majority of AI Overviews provide high-quality information with links for further exploration,” the spokesperson noted. “Many of the errors have arisen from uncommon queries, and we’ve also seen some examples that were doctored or couldn’t be reproduced. We conducted extensive testing before launching this new feature, and as with other Search features, we value the feedback.”
The Purpose of AI Overviews
Seeing these AI-generated search results raises an important question: What is their actual purpose? Even if they were flawless, how would they improve upon traditional web search?
Google’s goal is to deliver answers directly to users without requiring them to sift through multiple web pages. The company claims that early tests of AI Overviews showed people using Search more and being more satisfied with the results.
The concept of moving away from the “10 blue links” in search results isn’t new, and while Google has already minimized their prominence, it might be too early to completely phase them out.
Consider a search for “what is MyAIGuideBook,” which provides a mostly accurate but overly padded summary, with traffic numbers seemingly sourced from a Yale career website. A search for “how do I get a story in MyAIGuideBook” pulls an outdated article on submitting guest columns, a practice no longer accepted.
Addressing the Errors
The issue isn’t just identifying more errors in AI Overviews, but recognizing that many mistakes will be mundane rather than spectacular. Although the Overviews include links to source pages, identifying which information comes from which source still requires significant clicking.
Google explains that many inaccuracies highlighted on social media occur in data voids—topics with little accurate information online. This highlights the need for a healthy, open web full of reliable information.
Ironically, AI could threaten the open web it relies on. There’s less motivation to write accurate articles or break big stories if readers only see AI-generated summaries.
Google claims that AI Overviews lead to more diverse website visits for complex questions and that links in Overviews get more clicks than traditional web listings for the same queries. While this is a hopeful sign, if it doesn’t hold true, technical improvements alone won’t compensate for the potential loss of vast amounts of web content.
