The ability to search artificial intelligence in 60 % of the time is incorrect information

A new study conducted by the Tow Digital Journalism Center at Columbia University shows that artificial intelligence models equipped with Internet search capabilities provide a worrying amount of inaccurate information. Researchers examined 8 different models, including ChatGpt Search and Google Gemini, and found that these tools were answering more than 60 % of the questions wrong.

According to the Columbia Journalism Review, for example, the Perplexity AI, which was the most accurate tool in these experiments, still incorrectly offered 37 % of its responses. On the other hand, the Grok 3 search capability, owned by Ilan Musk, had the worst performance at 94 %.

The study showed that unlike traditional search engines that drive users to reputable sources, artificial intelligence -based search tools process and rebuild information and cut traffic from the main sources. Instead of refusing to respond if there is no valid information, these tools often provide incorrect information that seems significant.

In addition, the researchers found that some monetary models even performed worse than the free versions; For example, the monetary version of “Perplexity Pro”, which costs $ 20 a month, gave up more confidence than its free version.

Another important problem was to prevent the protocols from accessing the content of publishers; For example, the free version of “Perplexity” was able to extract articles from “National Geographic”, while the publisher blocked access to perplexity web creeping.

The accuracy of search capabilities with artificial intelligence is a big challenge

In addition, even when these tools cite their data sources, most users are directing users to redeveloped versions on platforms such as Yahoo News, rather than presenting the original link. In some cases, these models even produced fake links that guided users to the existential pages.

Despite the significant advances in artificial intelligence technology, search -capability models still have serious problems in providing information accuracy. Publishers face many challenges; Because they either have to allow their content to be used without direct traffic to their website or completely abandoned the technology. This can have significant impacts on the media industry and how users have access to reputable news.