Almost 50 news websites are “AI generated” – how to recognize them

So-called content farms are nothing new: websites that mass-produce news articles that aim to generate advertising revenue with clickbait headlines.

For these websites, text-generating AI technology seems to be a godsend. The US organization Newsguardwhich specializes in false information on the Internet, has so far identified 49 sites in which artificial intelligence is used for text creation in a new study.

The actual number is undoubtedly much higher, as Newsguard initially limited itself to seven languages: English, Chinese, Czech, French, Portuguese, Tagalog and Thai. The 49 pages found were in the format of regular news websites, but had content that was fully or partially generated by AI.

This is a worrying development for Newsguard: “As numerous and increasingly powerful AI tools become available to the public in recent months, concerns that they are creating entire news outlets – what was once merely the speculation of media scholars – have now become a reality,” the organization writes on its website.

The sites checked by Newsguard have harmless and unobtrusive names that suggest a reputable publisher behind the scenes, such as Market News Reports or Biz Breaking News. Often, however, an imprint or other information about the publisher is missing.

Of 29 operators contacted by the organization, only two have confirmed that they use artificial intelligence. 17 would not even have responded to the inquiries.

However, it is often already apparent from the article whether it comes from an AI. Often a crucial part of the text would be a message from the respective chatbot that it could not implement the prompt.

The British Newspaper The Guardians tested several of the sites himself and found, among other things, an obituary for the lively US President Joe Biden on Celebritiesdeaths.com. A statement that he had passed away peacefully in his sleep and that his vice president, Kamala Harris, would now take over, was followed by a note that “misleading content was against OpenAI’s policies.”

The Guardian also reports grammatically odd – but SEO-friendly – headlines, constant repetition in the texts and overly simplistic phrasing in supposedly scientific texts. However, it is not always so obvious to recognize AI-generated content.

In addition to the missing imprint, Newsguard points to a few features that you can look out for. So some of the pages would generate hundreds of articles per day, which is hardly possible in a regular editorial office. Also, many of the websites would be riddled with advertisements since the revenue from them is the sole reason for their existence.

This use of artificial intelligence is still in its infancy. As The Guardian notes, future AIs will learn from texts generated by others of their kind.