Urgent.News

What's breaking now, across thousands of outlets.

AI

Sibos 2026: Google Cloud reveals banks are moving from AI copilots to autonomous workflows

Held at The National Hotel in Miami during Sibos 2026, Google Cloud hosted a breakfast roundtable that focussed on the future of purpose-built AI for financial institutions, welcoming Georgina Bulkeley, director of financial services solutions, Google Cloud; Kristin Reinke, VP of finance, Google; Jason Allen, assistant treasurer, Google; Sarthak Pattanaik, chief data and AI officer, BNY; Joanne…

This comprehensive audit of Google AI Overviews, conducted by researchers from Washington University in St. Louis, reveals significant gaps between AI-generated answers and the sources they cite. Since its introduction in 2019, Google's AI Overviews have changed the way users discover information online, providing a single AI-generated summary instead of a list of sources. According to Professor Jacob Montgomery, Google now has editorial control over what users read and know, an unprecedented shift.

The study examined 55,393 Google searches over a 40-day period, revealing that AI Overviews appeared in 13.7% of all queries. However, the appearance rate varied greatly depending on the user's search phrasing. For question-based queries, AI Overviews appeared in 65% of cases, compared to only 9.5% for other types of searches. Longer search phrases also increased the likelihood of an AI Overview (38.7% for six-word queries or more).

Topic of the search played a role as well. Overviews were most frequent in hobby and leisure searches (46.1%) and science queries (39.9%), but appeared least frequently in political (7.5%) and law and government queries (9.6%). The cited domains in AI Overviews tended to have higher credibility scores than the first-page results, but 29.8% of them were not found among the first-page results, indicating Google's unique source selection process.

While most cited sources were credible, 11% of the overviews contained claims not supported by the cited text, suggesting that even trusted sources do not guarantee accurate summaries. Only 41.9% of the overviews were considered fully grounded, meaning every claim was supported by the cited text. The remaining 58.1% contained unsupported claims, highlighting the need for users to verify information even when it appears trustworthy.

The study also examined the financial implications for publishers. More than half of the cited webpages displayed advertising, and Google AI Overviews can prevent users from clicking through to these publishers, resulting in lost advertising revenue. In 2.2% of cases, a Google-sponsored ad appeared above the AI Overview, potentially preserving Google's ad revenue while reducing the incentive for users to visit the publishers whose content informed the answer.

Written by urgent.news from Phys.org's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

This story

This is one outlet's version. Read the fullest account.

Read the original at finextra.com →

More in AI

AI agents spark an identity cleanup

Identity and access management is getting a second look as AI agents move from pilots into production. The rush to deploy them is exposing decades of stitched-together identity tools, integration…

More from Tuesday 29 September →