Google is trying to correct its mistake.

The biggest search platform is incorporating Reddit and forums into AI Search to increase the credibility of answers, but this still leaves users confused and misled by the information.

Google's AI Overview has a 90% accuracy rate, but that still means millions of times it's wrong. Photo: Mashable .

Google has just released a new update to its AI-powered search engine, making it easier for users to find information from sources they know and trust. One notable change is the "perspective preview" which draws on real-world discussion from social media, Reddit, and other web forums.

The AI ​​Overviews feature previously revealed several weaknesses, particularly regarding information quality and accuracy. A survey published in April by the NYT showed that the search tool only produced accurate results about 90% of the time.

Meanwhile, the platform received hundreds of millions of questions in a short period of time. According to Ars Technica , this means Google is generating millions of incorrect answers every hour.

Furthermore, several cases demonstrate the serious extent of information fabrication by AI Search. Canadian singer Ashley MacIsaac sued Google last week for being accused of sexual harassment. The Guardian also reported that the tool provides a lot of dangerous information about health, particularly mental health and eating disorders.

Google says this update aims to address the fact that users "increasingly seek advice from others" when searching for information online. Typically, real-life experiences carry more credibility than SEO-optimized web results.

AI Overview uses answers from the community. Photo: Google.

By integrating these features, Google is trying to encourage more users to try them out instead of searching through traditional web results. The company says it will also add context to these new AI-responsive links, such as the name of the content creator, account, or community.

These direct experiences will be filtered by AI and appear with titles like “Expert Advice” and other labels in search results. This allows users to access specialized forums or specific Reddit sections they want to explore further.

However, this update will somewhat exacerbate the "answer bubble" phenomenon. A study at the University of Illinois found that when AI synthesizes information and produces a single result, it may not cite sources correctly but still use a confident tone.

Therefore, AI will make the internet sound more reliable than reality. Users who don't do thorough research or critically evaluate information are more likely to accept false information.

On the other hand, the nature of information is being subtly altered. Users are receiving opinions that are generally accepted but not necessarily entirely accurate. Google has also not disclosed how it verifies expert advice or what criteria it uses to filter comments.

In 2024, Google's reliance on Reddit resulted in numerous nonsensical responses, such as the idea that you can eat rocks or make pizza from glue. Data from SERanking that year showed the company had to reduce its reliance on Reddit because its AI still couldn't distinguish between factual information and jokes or sarcasm.

According to The Verge, this move by Google further complicates the role of AI Overview. The tool's purpose now lies somewhere between directly answering questions or providing users with a variety of information sources. This is no different from any other conventional search platform.

However, the platform will also add context to its answers, including links and information about the author. Users need to carefully check and verify the information before using it.

Source: https://znews.vn/google-dang-co-sua-sai-post1649444.html