Discussions about AI overview issues have been happening online, proving a rocky start for Google Search’s AI transition. In response, the tech giant released a statement on May 30, 2024.
The company explained that most viral “AI horror stories” were fake. Real AI overview errors were rare and usually stemmed from the search engine’s common issues.
READ: How to turn off Google AI Overviews
Nevertheless, Google announced it would implement tighter safeguards and monitor its new feature more closely to minimize errors.
What caused the AI overview issues?
AI Overviews summarize data from the top online searches and place them at the start of online search pages. As a result, users save time when looking for relevant information without having to go to multiple websites.
It also applies generative AI technology from ChatGPT and similar programs to online searches. Contrary to popular belief, they don’t use large language models alone.
The company integrated it with its core web ranking systems to perform “traditional search tasks.” These include identifying high-quality results from Google’s index.
The tech firm built this AI-powered feature to show information backed by the top web results. Thus, it does not make up or “hallucinate” results like other AI programs.
The AI Overview issues stem from other reasons, namely: misinterpreting queries, misunderstanding online languages, or lacking information.
Google claims these are challenges present in other Search features. Moreover, it debunked some of the outrageous errors attributed to the new Search function.
Some viral screenshots claim AI Overviews recommended leaving dogs in cars and smoking while pregnant. However, Google says it tried the same queries and did not receive these ridiculous answers.
Nonetheless, the company admitted that these errors may occur due to a “data void” or “information gap.” The latter is a lack of high-quality content related to a topic.
Google says it does not “fix” queries one by one. Instead, it updates the online search system to improve various query types. Here are the most recent ones it implemented:
- Improved detection mechanisms for nonsensical queries
- Limiting user-generated content for responses that may offer misleading advice
- Adding restrictions for queries where AI overviews were not helpful
- Enhanced safeguards for responses related to news, health, and other important issues