Get all your news in one place.
100’s of premium titles.
One app.
Start reading
iMore
iMore
Technology
Lloyd Coombes

Google breaks silence over AI search results that told people to eat rocks, put glue on pizza

Google AI.

While Apple's WWDC 2024 is expected to focus on plenty of AI-focused features, Google gave us a glimpse of what not to do last week.

The company, whose search engine is routinely regularly updated algorithmically to add new features that adjust rankings, rolled out its AI Overviews feature to summarise responses to a user's query.

With the AI seemingly unable to discern between genuine medical advice and satire, however, the process went more than a little awry, infamously suggesting users could use glue to make cheese stick to pizza or eat rocks to maintain a healthy lifestyle.

In a blog post this week from Liz Reid, VP and Head of Google Search, the company reflected on a challenging launch.

(Image credit: Brett Jordan)

Google discusses AI Overviews launch

While the post suggests there are some "faked screenshots", Reid notes that AI Overviews were predominantly caught out by "queries that people don’t commonly do."

"One area we identified was our ability to interpret nonsensical queries and satirical content," it reads, pointing to the "how many rocks should I eat" query that went viral.

"There isn't much web content that seriously contemplates that question, either. This is what is often called a “data void” or “information gap,” where there’s a limited amount of high-quality content about a topic," Reid explains.

"However, in this case, there is satirical content on this topic … that also happened to be republished on a geological software provider’s website. So when someone put that question into Search, an AI Overview appeared that faithfully linked to one of the only websites that tackled the question."

Google promises improvements have been made

According to the post, Google has now refreshed its technology with "better detection mechanisms for nonsensical queries that shouldn’t show an AI Overview, and limited the inclusion of satire and humor content" while also leaning away from user-generated content where it could be deemed misleading.

"We added triggering restrictions for queries where AI Overviews were not proving to be as helpful," Reid adds, also pointing to "additional triggering refinements to enhance our quality protections" when it comes to health and news-related content.

It's clear Google's AI Overviews need a sizeable change to bring it up to scratch - otherwise, it could join the graveyard of Google products. There are also concerns that by surfacing content without driving clicks to websites, Google will end up cannibalizing the sites that users go to.

More from iMore

Sign up to read this article
Read news from 100’s of titles, curated specifically for you.
Already a member? Sign in here
Related Stories
Top stories on inkl right now
One subscription that gives you access to news from hundreds of sites
Already a member? Sign in here
Our Picks
Fourteen days free
Download the app
One app. One membership.
100+ trusted global sources.