Google Responds to Criticism of AI Overviews

In a new blog post, Google’s new head of Search, Liz Reid, addressed the backlash over AI Overviews. Google explained how AI Overviews work, their origins, and the improvements made and planned for the future.

Despite the criticism, Google stated that users “have higher satisfaction with their search results and are asking longer, more complex questions,” indicating that AI Overviews are here to stay. According to internal testing, the accuracy rate of AI Overviews is comparable to featured snippets, which also utilize AI.

Reid emphasized that AI Overviews generally do not “hallucinate” or fabricate information, unlike other large language models. Errors typically arise from misinterpreted queries or language nuances, or insufficient information available.

Google acknowledged that some users deliberately tried to produce odd results. They also noted that some examples were manipulated with fake screenshots. Nonetheless, Google continues to improve its models to handle a broader range of queries instead of fixing individual ones.

Regarding “data voids,” Google explained cases like the query “How many rocks should I eat?” which led to AI Overview generating results based on satirical content due to the lack of relevant information.

Improvements to AI Overviews

• Enhanced mechanisms to detect and filter nonsensical queries.
• Updated systems to limit user-generated content that might provide misleading advice.
• Added restrictions for triggering AI Overviews where they are less helpful.
• Strengthened safeguards for topics like news and health to ensure relevance and accuracy.

Google will continue refining AI Overviews and strengthening protections, incorporating user feedback.

Why It Matters

AI Overviews are becoming a permanent feature in Google Search, expected to improve over time. Users can anticipate ongoing enhancements and occasional inaccuracies, similar to the initial rollout of featured snippets.

This article available in Ukrainian.