Google AI Overviews Take Center Stage This Year: Moving Past the 'Glue Pizza' Debacle

Katie Notopoulos from Business Insider reflects on a quite peculiar incident from 2024, one that involved Google's AI Overviews and an unusual pizza topping suggestion—glue. It's now the end of 2025, and Google's AI has certainly come a long way since then. Despite having poked fun at the AI's awkwardness back then, its progress is undeniable today.

The Infamous Pizza Experiment

In the not-too-distant past, Google's AI Overviews made headlines for a bizarre and erroneous piece of advice: suggesting the use of glue to keep cheese from slipping off a pizza. The idea, seemingly borrowed from a comedic Reddit post, turned into a viral sensation. As a journalist known for chasing truth, I took it upon myself to experiment with this recommendation, creating—and tasting—what would end up being dubiously memorable culinary creation.

A Significant Transformation

Fast forward to this year, and our expectations of Google AI Overviews have substantially shifted. Once lampooned for their erroneous suggestions, these tools have impressively improved, to the point where both my editor and I find ourselves frequently relying on them instead of traditional search engines.

This shift reflects a broader trend: users opting to use AI-generated answers rather than driving traffic through traditional means, a development that has its pros and cons. While convenient, it raises questions about the dynamics of web traffic and the importance of content creators.

Growing Pains and Humorous Glitches

Originally launched in the spring of 2024, Google's AI Overviews quickly gained notoriety for producing strange and erroneous answers. My 'glue pizza' encounter wasn't an isolated incident, as the Overviews frequently fell short by interpreting nonsense queries as legitimate idioms. This feature's tendency to fabricate meanings for random or playful phrases became known as the 'You can't lick a badger twice' problem, a quirk that did not escape social media observations.

Testing Eccentric Idioms

Testing the limits of these Overviews became something of a game: typing whimsical sentences followed by 'meaning' into Google yielded AI-crafted explanations for completely fabricated expressions. Notable attempts included oddities like 'you can't fit a duck in a pencil' and 'the road is full of salsa,' which the AI dutifully provided interpretations for.

Towards Improved Accuracy

Today's AI systems are vastly better at identifying when a phrase is nonsensical. Just recently, I tested with a new phrase, 'you can't tell a yak not to dance,' and was met with a more measured response, acknowledging it as a non-traditional expression with potential poetic interpretations. This indicates a deepening sophistication in Google's AI tool, as it becomes more user-friendly and reliable.

Conclusion: A Futuristic Fine-Tuning

With familiarity, I’ve developed a sense of when AI Overviews will offer the information I need. While not infallible, its usefulness and accuracy have undeniably improved. It's encouraging to see technology evolve and optimize itself, hopefully leading to even fewer quirks and more reliable outputs in the future.