Google is Fixing its AI Overviews

Following months of testing in select markets, Google recently started rolling out AI Overviews in the US. It did not go well. Many users quickly noticed that the ChatGPT-like answers at the top of search results could give nonsensical answers, sometimes based on random posts from Reddit.

AI Overviews were supposed to give users “information you can trust in the blink of an eye,” Google said two weeks ago. Well, the company has now acknowledged that something was wrong with its AI Overviews, and it has also detailed recent changes to address the issues. However, Liz Reid, VP and Head of Google Search suggested that the bad buzz was somewhat overblown.

“In the last week, people on social media have shared some odd and erroneous overviews (along with a very large number of faked screenshots),” the exec wrote. “We know that people trust Google Search to provide accurate information, and they’ve never been shy about pointing out oddities or errors when they come across them — in our rankings or in other Search features. We hold ourselves to a high standard, as do our users, so we expect and appreciate the feedback, and take it seriously.”

The Head of Google Search went on to explain that the company’s AI Overviews work differently than ChatGPT and other chatbots: AI overviews combine a new Gemini model customized for Google Search with Google’s core web ranking systems, and that’s why the conversational answers include links to web content.

“This means that AI Overviews generally don’t “hallucinate” or make things up in the ways that other LLM products might,” Reid explained. “When AI Overviews get it wrong, it’s usually for other reasons: misinterpreting queries, misinterpreting a nuance of language on the web, or not having a lot of great information available.”

The exec claimed that the company’s tests revealed that the accuracy rate for AI overviews with “on par” with Google’s featured snippets, which also use AI systems to highlight information from web search listings. Google also tested AI overviews “extensively” before launch, and Reid admitted that “We’ve also seen nonsensical new searches, seemingly aimed at producing erroneous results.”

So, what went wrong exactly? According to the Google Search exec, AI Overviews could struggle to understand uncommon or nonsensical queries, as well as satirical content. Reid gave the example of the “How many rocks should I eat” query that went viral. Because there was a general lack of high-quality content regarding this strange topic, what happened is that an AI Overview linked to one of the only websites that answered this question, but with a satirical angle.

“In other examples, we saw AI Overviews that featured sarcastic or troll-y content from discussion forums,” the Google exec explained. “In a small number of cases, we have seen AI Overviews misinterpret language on webpages and present inaccurate information. We worked quickly to address these issues, either through improvements to our algorithms or through established processes to remove responses that don’t comply with our policies.”

Ultimately, the Google Search team recently made “more than a dozen technical improvements” to its systems to prevent AI Overviews from delivering nonsensical results. Google will now limit the use of user-generated content in its responses, and it also enhanced its quality protections for health-related queries.

The Google Search exec said that the team will continue to listen to feedback and improve AI Overviews over time, though it also admitted that mistakes could still happen. “At the scale of the web, with billions of queries coming in every day, there are bound to be some oddities and errors,” Reid said.