AI Generated News
Kindus has already discussed the issues of digital advertising fraud. A June 2023 report from NewsGuard implies that Artificial Intelligence is fuelling a boom in junk news sites.
Their business model is to create a very large number of websites and use them to host Google AdSense. This engine serves adverts based on its interpretation of the host’s content and pays out on impressions and click-throughs. AdSense only requires simple code insertion to go live and promises automatic payouts to the host. Payments can be as low as $2 or less per 1,000 page views. One of the simplest ways to optimise this is through hosting a great number of pages.
Evidence that some of these are being created through Artificial Intelligence comes through the content text itself. An example from a Brazilian job information site contained the text “Sorry, as an AI language model, I am not able to access external links or websites on my own.” Others included content clearly sourced and re-written from genuine publishers. NewsGuard used clues such as these to identify 217 obviously AI generated sites. Many others could pass through the net because of more believable text or the even the simplest of human editing of the AI content. A few sites such as NewsGPT make no secret of their use of AI and even encourage feedback on the quality of the articles. Others have been shown to promote fake and incorrect news although this could be as a result of insufficient human oversight rather than any deliberate spread of false information.
One alleged AI driven world news site was alleged to be creating an average of over 1,200 articles a day. Checking the site in July 2023 revealed 16,291 pages of headings each for 8 or 9 pages for technology alone. There were, however, no adverts on these pages. Either these have been blocked or the site has some other reason for hosting so much content all of which is probably available elsewhere. Another named example; a food website did include Google Ad placements on its home page and at least some of its 497 pages.
If these sites simply sat on the Internet and were never seen then the model would not work. They need to garner impressions. The authors will be doing all they can to crawl up the Google search ranks, pushing sites with more merit further down. They may also rely on SPAM emails and social media entries to encourage casual click throughs. None of this is obviously illegal but is hardly for the public good or even benefitting genuine business interests.
There are consequences to these activities beyond clogging up the Internet with duplicate, low quality content. The AdSense model depends on advertisers delivering their messages to the most appropriate targets. If these pages can fool the placing model algorithm then the advertiser is paying for placements that are unlikely to follow through. The payment calculation for the host is complex and there could be a consequence of reduced payments to genuine, high quality human-created hosting sites as the calculating engine will have difficulty distinguishing informed opinion from AI synthesis.
