When can you expect your query to generate an AI-generated summary of the results? “AI overviews appear for complex queries,” says Google spokesperson Mallory De Leon. “You’ll find AI overviews in Google Search results when our systems determine that generative AI may be especially useful, for example, when you want to quickly understand information from a variety of sources.” During my initial testing, I felt that AI overviews appeared almost randomly for queries, with summaries appearing for simple questions as well as more complicated questions.
According to De Leon, AI Overview works with a customized version of Google’s Gemini model that is complemented by aspects of the company’s search system, such as the Knowledge graph which has billions of general facts.
One of my biggest concerns about this feature as it rolls out is the continued potential for AI hallucinations, more commonly known as lies. When you interact with Google’s Gemini chatbot, a disclaimer at the bottom says: “Gemini may display inaccurate information, including about people, so double-check its answers.” No such disclaimer is added to the end of the AI overview, which often simply says: “Generative AI is experimental.”
When asked why there is no mention of possible hallucinations in the AI overviews, De Leon emphasizes that Google still wants to offer high-quality search results and mentions that the company conducted adverse red team testing to discover possible weaknesses of the function.
“This implementation of generative AI is rooted in Search’s core quality and safety systems, with built-in guardrails to prevent harmful or low-quality information from appearing,” it says. “AI overviews are designed to highlight information that can be easily verified by the supporting information we present.”
Knowing this, you may want to click on the links on the website to check that the information is actually correct. Although it is difficult to imagine that many users, who are often looking for quick answers, will spend more time reading the source material of the answer generated by Google’s AI.
Liz Reid, Google’s head of Search, recently told my colleague Lauren Goode that AI overviews are expected to reach countries outside the United States before the end of 2024, so more than a billion of people will soon find this new feature. As someone whose job depends on readers clicking on links and spending time reading articles, of course I’m concerned about this change, and I am not alone.
Beyond publishers’ concerns, it’s also unclear what additional impacts could impact users from Google’s AI overviews. Yes, OpenAI’s ChatGPT and other AI tools are quite popular in Silicon Valley tech circles, but this feature will likely expose billions of people, who have never used a chatbot before, to AI-generated text. Although AI overviews are designed to save you time, they can lead to less reliable results.