Google Faces Backlash Over AI Search Errors as New Feature Misfires

Last week, Google unveiled its biggest change to search in years, showcasing new artificial intelligence capabilities designed to answer users’ questions as the company strives to catch up with rivals Microsoft and OpenAI.

However, the new technology has generated numerous falsehoods and errors — including recommending glue in pizza recipes and suggesting the ingestion of rocks for nutrients — causing a significant backlash and damaging Google’s reputation.

These inaccuracies in the feature, called AI Overview, have undermined trust in a search engine relied upon by over two billion people for authoritative information. While other AI chatbots also produce errors, the reaction highlighted the heightened scrutiny Google faces in safely incorporating AI into its search engine.

This launch continues a pattern of Google encountering issues with its AI features immediately after rollout. In February 2023, when Google announced Bard, a chatbot to compete with ChatGPT, it provided incorrect information about outer space, leading to a $100 billion drop in the company’s market value. This February, Google released Bard’s successor, Gemini, which could generate images and act as a voice-operated digital assistant. Users quickly found the system often refused to generate images of white people and inaccurately depicted historical figures.

Tech industry insiders have criticized Google for these missteps. However, financial analysts argue that Google needs to act swiftly to stay competitive, even if it means experiencing growing pains. “Google doesn’t have a choice right now,” said Thomas Monteiro, a Google analyst at Investing.com. “Companies need to move really fast, even if that includes skipping a few steps along the way. The user experience will just have to catch up.”

Lara Levin, a Google spokeswoman, stated that most AI Overview queries resulted in “high-quality information, with links to dig deeper on the web.” She added that many problematic examples were uncommon queries or doctored images that Google could not reproduce. The company plans to use these “isolated examples” to refine its system.

Since OpenAI released its ChatGPT chatbot in late 2022, Google has faced pressure to integrate AI into its popular apps. But taming large language models, which learn from vast amounts of web data — including false and satirical content — poses significant challenges.

Google announced AI Overview at its annual developer conference, I/O, last week, integrating its latest large language AI model, Gemini, into its search engine for the first time. AI Overview combines AI-generated statements with snippets from live links across the web, citing sources but unable to discern when those sources are incorrect.

The system was designed to handle more complex and specific questions than regular search, aiming to benefit the public by leveraging Gemini’s capabilities and simplifying information searches. However, users quickly highlighted errors, sharing screenshots of problematic results on social media platforms like X.

AI Overview advised some users to mix nontoxic glue into pizza sauce to prevent the cheese from sliding off, a fake recipe from an 11-year-old Reddit joke post. It also suggested ingesting rocks daily for vitamins and minerals, advice originating from a satirical post by The Onion.

Google search is critical to the company's success, and maintaining its reliability and usefulness is paramount. “The one property Google needs to keep relevant/trustworthy/useful is Google search,” wrote software engineer Gergely Orosz on X. “And yet, examples of how AI overviews are turning Google search into garbage are all over my timeline.”

Users also shared instances of AI Overview providing dangerous advice, such as using “chlorine bleach and white vinegar” to clean washing machines, which can create harmful chlorine gas. Additionally, it provided incorrect historical information, such as claiming 17 presidents were white and Barack Obama was the first Muslim president, and stating Andrew Jackson graduated from college in 2005.

Despite the backlash, Google remains committed to improving its AI capabilities and integrating them into its products, though it faces significant challenges in ensuring accuracy and safety.