AI-driven search tools have really become popular, and more people are beginning to use them in place of older search engines. These platforms harvest online material from news outlets to present immediate answers. They often remix that information without clear links to publishers.
Traditional services guided users directly to a website, benefiting the original creator with click-through traffic. AI tools, on the other hand, package the text into a chat-like response. This method blocks readers from ever visiting the original source, removing any chance for journalists to gain attention for their work.
Inaccuracies are the biggest problem when it comes to AI… These systems often give answers that sound correct but turn out to be wrong. Some in journalism fear such errors may harm the standing of news outlets, particularly when the chatbots mislabel sources or use made up web links.
How Was This Study Conducted?
A research group at the Tow Centre for Digital Journalism tested 8 AI services to see how well they handled stories. They pulled excerpts from 200 articles, each belonging to 20 different publishers. Then they asked the chatbots to find the article’s headline, publisher, date, and link.
The same quotes brought up the correct piece in the top three results on an internet search. This meant the question was not too obscure. Yet, six out of ten times, the AI search tools gave the wrong article, a mismatched publisher, or a broken link.
One platform gave a response to every query but made up facts in over half of its answers. Others refused to respond sometimes but still stumbled when they tried to pin down the right source. This pattern of high confidence and low accuracy was a running theme.
More from News
- Two Thirds of Brits Are Telling AI Chatbots Their Secrets But Don’t Trust The Companies Behind Them
- Video Game Companies Are Abandoning “Buy Our Console To Play This” – Here’s Why
- UK Gambling Operators Face New Deposit Limit Rules From 30 September
- How A Ransomware Gang Hijacked A Rival Gang’s Website – And Is Now Extorting Them For “2.333% Of Their Net Worth
- Meta Promised Unmatched Privacy For Muse – So How Did Researchers Crack It So Quickly?
- Digital Avatars Are Bringing Deceased Artists Back To The Stage – Where Do We Draw The Line?
- Can Europe Solve Its Critical Minerals Problem Fast Enough To Support Its Tech Ambitions?
- Google Has Agreed To A £260 Million Settlement: Who’s Eligible For Compensation?
Which Tools Stood Out?
Perplexity and its premium version had stronger records overall, although they were not free from errors. A sizable group of queries still produced mistakes, such as broken URLs. Meanwhile, Microsoft’s Copilot often declined to supply an answer, which led to fewer incorrect statements but still left many queries unsolved.
Two models under the Grok name delivered some of the highest error percentages. One version reached a 94% fail rate, causing alarm among those tracking the study. Gemini also struggled to attach the correct article, especially for political content, and declined to handle certain topics outright.
Research showed that blocking chatbots from crawling a site did not always hold them back. Perplexity’s premium tier, for instance, identified details from sources that said no to its crawler. This points to a complex environment, where a quest for direct traffic runs headlong into the design of large language models.
What Does This Mean For News Providers?
Publishers often depend on search referrals as one way to draw attention to their work. If chatbots feed the public flawed details without routing them back to the original outlet, the economic model of online journalism might suffer. Lower visitor counts can reduce advertising revenue and impact news gathering momentum.
Some organisations have partnered with AI companies in the hope that deals for direct content access will increase accuracy and proper linking. Early results have been uneven. Time magazine fared better than many outlets, but others saw only a small number of correct results in a long list of queries.
Representatives from certain media groups say they want better disclosure about how content is used and attributed. They point out that if these tools depend heavily on a publisher’s reputation, they should at least direct users back to the correct source. Otherwise, brand image may be at risk with no added readership.
