In recent years, Google has changed the operation of the search engine by prioritizing displaying answers created by artificial intelligence at the top of the page.
The AI Overviews feature shows that Google is shifting from leading users to news sources to directly creating and providing content.
This change helps users access information faster, but also raises many debates about the accuracy, transparency and reliability of AI-generated answers.
A recent analysis by the startup in the field of artificial intelligence - Oumi (USA), shows that the answers from AI Overviews are about 85% accurate when using Gemini 2 technology and increased to 91% with the upgraded Gemini 3 version.
However, with more than 5,000 billion searches each year, this small error rate can still be equivalent to tens of millions of wrong answers per hour.
Notably, more than half of the correct answers were rated as "unfounded", meaning that the source links did not fully support the information provided. This makes it difficult for users to verify.
Oumi's analysis is based on the SimpleQA standard test, which is a popular tool for assessing the accuracy of AI systems.
However, Google refuted this result, saying that the evaluation method has a "serious loophole", because the test question set itself may also contain misinformation.
Besides the issue of accuracy, the data source is also a controversial factor.
Among the more than 5,000 sources cited, platforms such as Facebook and Reddit appeared with high frequency.
Notably, when AI gives the wrong answer, the citation rate from Facebook is even higher than when the answer is correct, raising questions about the quality of the reference source.
Another problem lies in the very way AI is evaluated. Companies like Oumi often use other AI systems to verify answers.
However, this method contains risks when "AI testing AI" can also make mistakes, making the evaluation results not completely absolute.
Despite many controversies, it is undeniable that AI Overviews are increasingly being improved and becoming an important part of the search experience.
However, experts warn that users should not completely rely on these answers, but need to maintain the habit of verifying information from many different sources.
