Google AI Overview fails 90% of tests on AI image accuracy
A Finnish analysis reveals Google's AI summaries frequently provide false information when assessing AI-generated images.
Google's AI Overview continues to struggle with factual accuracy, specifically when identifying and assessing AI-generated images. New data suggests the tool remains unreliable for users attempting to verify the origin or nature of visual content.
An analysis conducted by Faktabaari, a Finnish partner of NORDIS, tested the AI's performance in April 2026. The study focused on the tool's ability to provide accurate information when connected to reverse image searches. In Finnish-language tests, the AI Overview provided completely correct information only once out of ten attempts, representing a 90% failure rate in these specific instances.
The Hallucination Problem
This failure is part of a broader, ongoing pattern of "hallucinations" within Google's AI Overviews. Since their wide rollout, these AI-generated summaries have faced consistent criticism for producing plausible-sounding but factually incorrect answers. While Google has attempted to refine the system, the Faktabaari results indicate that the tool still struggles to synthesize technical data from reverse image searches into reliable text summaries.
Implications for Trust
Persistent inaccuracies in AI-generated summaries are particularly problematic when they concern the nature of AI-generated content itself. When a search engine provides false information about whether an image is synthetic or real, it can mislead users on the technical capabilities of the software and the authenticity of the media they encounter. This undermines the general reliability of the search engine as a tool for fact-checking and verification in an era of increasing digital manipulation.
Looking Ahead
The NORDIS report highlights a critical gap between the deployment of AI summaries and their actual reliability in specialized tasks like image verification. As Google continues to integrate generative AI into the core search experience, the industry will be watching to see if the company can move beyond superficial fixes to solve the underlying hallucination issues. For now, the results from the Finnish tests suggest that users should remain skeptical of AI-generated claims regarding the provenance of images.