In a somewhat ironic twist, Google’s Super Bowl ad campaign for its Gemini AI model inadvertently showcased a major challenge facing AI tools: their tendency to fabricate information. The Verge recently reported that Google had to revise its Super Bowl ad promoting Gemini after the AI provided inaccurate information in the original version.
The ad campaign focuses on how small businesses across the U.S. are utilizing Gemini. Fifty different stories highlight a different business in each state. The Wisconsin spot featured a cheesemonger using Gemini to generate website copy. The AI-generated text claimed that gouda constitutes “50 to 60 percent of the world’s cheese consumption,” a statement that is demonstrably false.
Travel blogger Nate Hake pointed out the error on X (formerly Twitter), labeling the statistic a hallucination due to the lack of sourcing. Jerry Dischler, President of Cloud Applications at Google Cloud, responded, arguing that Gemini is “grounded in the Web” and users can verify information and references. However, the original ad text provided no citations. Dischler claimed multiple websites support the 50-60% statistic.
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/27471828/gemini_gouda_ad_google_ai.jpg”>https://cdn.vox-cdn.com/thumbor/4Z7539b9t14w7tF4P5R7h3n0L0=/0x0:1920×1080/1310×0/filters:focal(0x0:1920×1080):no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/27471828/gemini_gouda_ad_google_ai.jpg)
While the statistic does appear online, its origins trace back to cheese.com, a site that offers no supporting evidence for the claim. Blaming the inaccuracy on Gemini pulling from unreliable sources isn’t a strong defense, especially in light of the adage, “Don’t believe everything you read on the internet.”
Interestingly, Dischler’s X response concluded with some cheese-related puns, suggesting a possible reliance on Gemini for creative content generation. While demonstrating faith in the product, this approach might not project the most sophisticated image.
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/27471832/jdischler_tweets_natehake_google_gemini_gouda_ai.jpg”>https://cdn.vox-cdn.com/thumbor/d1z8_B4J08W2Jjyq-H1W2N7D3-0=/0x0:1916×1278/1310×0/filters:focal(0x0:1916×1278):no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/27471832/jdischler_tweets_natehake_google_gemini_gouda_ai.jpg)
Despite defending the initial ad, Google ultimately revised it, removing the erroneous statistic. Whether the change was manual or involved a revised prompt for Gemini remains unclear. The corrected version aired during the Super Bowl.
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/27471829/gemini_gouda_ad_google_ai_corrected.jpg”>https://cdn.vox-cdn.com/thumbor/f2T3bW4v5m8a6mN9qE6J39y-n8=/0x0:1920×1080/1310×0/filters:focal(0x0:1920×1080):no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/27471829/gemini_gouda_ad_google_ai_corrected.jpg)
This incident serves as a potent illustration of the potential pitfalls of AI tools like Gemini. They can readily replicate information without effectively discerning its accuracy. While AI can save time, critical evaluation of its output remains essential.
