Overview
- Google’s AI Overviews generate false information known as hallucinations, sometimes offering dangerous or nonsensical advice such as adding glue to pizza sauce.
- Studies show the feature reduces click-through rates to publisher websites by 40–60 percent on searches where it appears, raising concerns about its impact on online journalism.
- Despite advances in the underlying Gemini model, its hallucination rate has climbed to 1.8 percent according to Hugging Face, exceeding Google’s publicly stated range of 0.7–1.3 percent.
- OpenAI’s latest reasoning-focused models, o3 and o4-mini, also report higher error rates of 33 percent and 48 percent respectively, illustrating industry-wide challenges.
- Google CEO Sundar Pichai and head of Search Liz Reid have defended AI Overviews for broadening source discovery but have acknowledged that the system requires further refinement.