No point discussing it - AI generated news is already here

NewsGuard published a report on 49 AI-generated news blogs and TheMayor.EU tried to verify these claims

Since the public launch of ChatGPT, fact-checkers and journalists alike have been dreading the day when content farms and troll farms wise up and start mass adoption of generative AI language models. Well, it seems that day has come.

On  1 May, NewsGuard, a company that provides accuracy ratings for news sites published a report claiming to have found 49 websites that use only AI, with little to no human editing. These are mainly low-quality click-bait articles on a variety of topics including politics, health, entertainment, lifestyle, finance, religion and tech.

NewsGuard found AI websites function in seven languages – Chinese, Czech, English, French, Portuguese, Tagalog and Thai. Some publish hundreds of articles a day. And some advance false narratives.

Doubts

According to the report, all 49 websites feature programmatic ads, placed algorithmically. The content, in this case, serves as Search Engine Optimisation, to be able to generate as much ad revenue for as little effort as possible.

Our team at TheMayor.EU visited several of the listed websites and came to a striking conclusion: most of them were virtually indistinguishable from regular content farms - in other words, nothing new on the Internet.

The listed websites contained all of the hallmarks of low-cost content farms one would expect: BOMBASTIC HEADLINES IN ALL CAPS!!!, ads popping up from everywhere, articles written by ‘admin’ (rather than a real personal name), no contact information or company details, bland language and dodgy English.

However, these factors do not necessarily point to AI-generated content. Websites like this have been a staple of the Internet for quite some time now. And although their content has always certainly felt very ‘robotic’, they were written by real people. This made us doubt NewsGuard’s claims that the content was actually generated via AI language models – so, we tried to verify them.

The AI news sites pit is only inches deep

One of the big claims NewsGuard makes is that in every one of the 49 websites, there was at least one article featuring the following AI error messages: “my cutoff date in September 2021,” “as an AI language model,” and “I cannot complete this prompt,” among others.

Indeed, during our research, we found that some of the listed websites were very obviously written by AI. The effort to hide that fact was minimal at best and the supposed ‘deception’ was only inches deep. However, others proved more elusive and the distinction between the two often came down to manually deleting the error messages – which is still a very low bar to pass.

All the websites we examined had the hallmark look of content farms, yet we found some quite tricky to pinpoint and say without a shadow of a doubt that the content was 100% AI, humans need not apply. And maybe that is the scary part – all of the listed websites were able to produce enormous amounts of content.

A Checker’s Journey

One website we visited was GetIntoKnowledge.com. The website seems to have been active since 18 December 2022 and stopped publishing new articles on 23 April this year. All are credited to ‘GET INTO KNOWLEDGE’, as an author, and cover a wide range of topics, from engineering to biographies. The site also features a section called ‘Podcasts’, where an artificially generated voice reads some highlighted articles.

Additionally, the article "What is a Leap Year? Why is a leap year after every 4 years? – Amazing Science explained" features the famous phrase “As an AI language model, I am excited to delve into the reasons behind leap years and try to make it understandable for everyone.”

Screenshot from Get Into Knowledge, Source: TheMayor.EU

We cannot skip mentioning CountyLocalNews.com, which seems to only have one author - one John Ekpa. Since the website’s first publication on 7 February 2023, Mr Ekpa has written an astonishing 8,390 articles, which might make him possibly the most prolific author in the world.

At first, he started slow, writing only a couple or so days, but then on 8 March, he wrote 14 articles. Just today, 4 May, as we're ready to publish this article at 12:15 EET, John Ekpa has already managed to write 36 articles!

A title from an article he published on 20 April: “Life in Muddy Trenches Amid Russian Fire Note: As an AI language model, I am not programmed to promote or endorse violence, hate speech, or discrimination”, makes us think that his creativity must have been boosted with some AI help.

CountyLocalNews screenshot, Source: TheMayor.EU

Another website NewsGuard points to is called HarmonyHustle.com. It was created on April 2023 and covers topics from lifestyle to crypto. Additionally, it has no About page and is operated anonymously. Most of its authors have generic names like Tom, Leo, Nick and Sophia, but they do have profile pictures and we decided to try and track them down, using InVid - an attempt to prove if these are, in fact, real people.

Grainy profile picture of 'Leo', Source: HarmonyHustle

We decided to start with Leo since his grainy portrait photo looked like he had something going on with his fingers. One of the major tells with AI-generated imagery is fingers, usually they are mangled or of an irregular number. However, a reverse image search revealed that he was just a stock photo person from Shutterstock and ‘BigStockPhoto.com’ and this photo is titled ‘Young man eating sushi Asian food using chopsticks.   

It looks like ‘Leo’ is definitely a real person, even if his hands are photographed in a weird way. Also, the fact that this person is a stock photo model does not necessarily mean that he does not write articles for HarmonyHustle. So we looked into Sophia, and found her on Unsplash, a stock photo model for creator Jonathan Borba.

Grainy profile photo of Sofia, Source: HarmonyHustle     
 

Source photo from Unsplash, Source: Jonathan Borba

Despite our best efforts, however, we didn’t find any articles containing AI error messages on HarmonyHustle. The whole website, as far as we are concerned, could have absolutely been written by humans. It could have also been written by AI with human editors or at least someone who has combed through the articles to delete these lines.

Although the article ‘How Artificial Intelligence is Revolutionizing the Healthcare Industry’ is one that Sophia appears to have written at least five times. The text features very generic and repetitive lines, but we could not discover 100% telltale signs. In theory, it could be written or at least edited by a real person.

Additionally, the site’s owners could have cleaned the content up since the NewsGuard article was published - a low bar but still enough to seed doubts in our conclusion.

Yet, all of the above also highlights the need to find out ways to distinguish, without a shadow of a doubt, whether a piece of content has been AI-generated or is genuine. Ironically, that might be another job for AI to do.