Google is making some changes to its AI overview after its artificial intelligence-powered search feature returned what the company called “weird and incorrect” responses to people's online searches.
The AI Summary feature was announced last month at Google's annual developer conference, I/O. Now, when users use Google Search to find information on a particular topic, they'll see boxes of AI-generated text at the top of their search results, annotated with links to external websites. Traditional search results will appear below the AI Summary feature, marking a major change in the way Google presents information.
According to a blog post from Google vice president Liz Reid, the AI Summary results are generated using the company's large-scale language model (LLM) Gemini and are designed for cases where you “want to get both a quick summary of a topic and links to learn more.”
Google tech expert Alex Joseph told ABC Audio that AI Overviews can answer more complex questions than a traditional Google search.
“In AI Overview [Google] “What we can really do is synthesize a lot of information and get the answers we're looking for very quickly,” Joseph said.
Joseph said AI Overviews streamlines the process by summarizing information and providing users with concise answers, rather than presenting them with pages of links to dig through.
“It reduces friction and eliminates the need to click through to several different websites, which is often quite tedious when you want information quickly,” said Chris Stokel Walker, technology journalist and author of “AI: A Concise History of Artificial Intelligence and Its Long-Term Future.”
But Stokel Walker said the new feature will make it harder for people using Google Search to verify the accuracy of the information they're reading.
“Because of the dominance of Google search over the last 20 years, we've become accustomed to the results we get for search terms being mostly correct,” he told ABC Audio. “And if you suddenly take that away, as Google is proposing, and just shoehorn answers generated by generative AI directly into your search results page, then you have absolutely no way to identify and analyse that information and work out whether it's true or not.”
There are other concerns about the new feature, including that generative AI technology from Google and other companies has faced criticism for “hallucinating,” or generating information that is unreliable and inaccurate.
For example, in the weeks since AI Overview was publicly released, it advised people using Google Search to eat at least one small rock a day, and told one user to mix glue into tomato sauce to make cheese stick to pizza—both very bad ideas, of course. It also told users that Andrew Jackson, the seventh president of the United States, who died in 1845, graduated from college in 2005.
Stokel Walker said the benefits of AI Overview ultimately come down to a trade-off between convenience and cost: “You no longer have to click through five or six, maybe even pages of search results to find the right answer, but that answer might also be wrong or not the answer you actually want to know,” she said.
“We've always been clear about the limitations of LLM, namely that hallucinations sometimes occur,” Google's Alex Joseph said, adding that that's why they also cite the website that AI Overviews uses to generate answers.
“That's one of the reasons why we present all of the information comprehensively,” Joseph said. “These are quick shortcuts to get the information quickly, but then we also have areas where you can review and verify.”
Joseph also said that not all queries are best served by AI summaries, saying, “We only show AI summaries for queries where we are confident that it will be helpful and will actually improve the experience.”
Following the unusual reactions reported by some social media users, Google announced that it had made “more than a dozen technical improvements” to its AI summaries. According to a blog post by Liz Reid, these include limiting the inclusion of user-generated content and satirical or humorous web pages in the data used to create the AI summaries. Reid said that Google has “launched additional trigger improvements to strengthen quality protections” for health content, and that “[s] We will not show AI summaries for hard news topics where freshness and factuality are important.”
The blog post also states that “AI Overviews typically does not 'hallucinate' or fabricate facts like other LLM products,” and that incorrect answers are the result of “misinterpreting the query, misinterpreting nuances in language on the web, or not having much good information available.”
Apart from concerns about accuracy, Stokel Walker said Google's prioritization of AI summaries over traditional search results could impact revenue and change the way it does business on the web.
“Websites create content and try to make it attractive to Google, Google displays that content in search results, people then click on the website, they see ads on the website, and publishers get revenue to put new content on their websites,” Stokel Walker explained.
But replacing the top of Google search results pages with AI-generated content could result in fewer website visitors and less advertising revenue, Stokel Walker said.
It's an ironic situation, according to Stokel Walker, because Gemini, the law master's program Google uses to create its AI overviews, relies on the websites that currently rank highly.
“These websites still need to exist and have a way of generating revenue, otherwise there would be no basis for these AI-generated search results,” she said.
In a statement to ABC News, Google said that its tests actually showed the opposite: that links included in the AI overview received more clicks than if the pages were displayed normally in search results. Google also said it “remains committed” to sending valuable traffic to publishers and creators.
Regardless of how concerns about AI Overviews are ultimately resolved, it's just one of many features the company has planned for its technology product line.
“I think it's concerning that Google is moving this so quickly,” said C. Scott Brown, a tech journalist at the website Android Authority.
Google has announced plans for additional features similar to AI Overviews that aim to answer questions about specific web pages or YouTube videos, and Brown said these will come against a backdrop of increased competition.
“And the reason they do so is because they feel they have to: Google has to keep up with companies, particularly companies like OpenAI, that are developing generative AI technologies that threaten Google's core business of providing information to people through Google Search and therefore serving the ads that allow Google to make billions of dollars,” Brown said.
“Google sees this as a threat and they can't just sit back and figure out how to do this cleanly and correctly,” Brown added. “They just have to do it.”
