Google is pulling back the use of AI-generated answers in search results after the feature made some infamous errors, including telling users to put glue in their pizza sauce.
Google launched AI Overviews, which put AI-generated summaries of search results on the top of the page for US users, two weeks ago. Over the last few days, users, including an SEO expert, noticed fewer AI overviews and suspected that the tech giant was taking them down a notch after criticisms. It is not possible to turn off the AI feature while using the search engine.
Google’s head of search, Liz Reid, confirmed in a blog post on Thursday that the company is addressing some of these issues.
The changes come after recent examples of AI overviews going haywire — and faked pictures of the feature — flooded the internet. These included responses claiming Barack Obama was a Muslim president, that Africa has no countries beginning with the letter K, and that people should eat “at least one small rock per day.”
Google’s new guardrails include detecting “nonsensical queries” that shouldn’t show AI results, limiting satire or humor content, and introducing restrictions for prompts where AI results would not be helpful because there is not enough data about that topic.
Google’s own ads show that the erroneous summaries aren’t limited to a few viral queries. In a demo video released two weeks ago, the Overview feature wrongfully advised the actor on how to fix their film camera.
Reid’s blog post also said Google has limited content from forums or social media, which can have misleading advice.
“Forums are often a great source of authentic, first-hand information, but in some cases can lead to less-than-helpful advice, like using glue to get cheese to stick to pizza,” Reid wrote in the post.
Reid wrote that the company already has systems in place to not show AI-generated news or health-related results. She said that harmful results that encouraged people to smoke while pregnant or leave their dogs in cars were “faked screenshots.”
The list of changes is the latest example of the Big Tech giant launching an AI product and circling back with restrictions after things get messy.
Earlier this year, Google AI’s image-generating feature came under fire for refusing to produce pictures of white people. It was criticized for being too “woke” and creating photos with historical inaccuracies like Asian Nazis and Black founding fathers. In a blog post a few weeks later, Google leadership apologized and paused the feature.