Low-quality sites can use language models to produce articles at a pace that makes each page another opportunity to display advertising. A NewsGuard study found that established brands were advertising on such sites, with Google Ads responsible for most of the ads tracked.
A fast-growing category of sites
NewsGuard, a technology company that evaluates the quality of news and information online, labels these sites “Unreliable Artificial Intelligence-Generated News,” or UAIN. It reported that its tracked list grew from 49 to 217 sites in the last month, with about 25 new sites identified each week.
To identify them, NewsGuard looks for telltale model error messages, including ChatGPT’s “As an AI language model”. The approach has limits: the study says the method is inherently imprecise, so many sites may go undetected. That caveat matters when interpreting the count. The tracked sites offer a view of the activity NewsGuard could identify, not necessarily a complete measure of it.
The sites use chatbots such as ChatGPT to create articles or rewrite material from major publishers. The resulting pages can appear polished enough to pass ad technology companies’ anti-spam detectors. One site in the study reportedly published more than 1,200 articles per day, treating each article as advertising space.
Where the ads appeared
Across 55 UAIN sites, NewsGuard counted 393 programmatic ads from 141 established brands. Those ads appeared in the United States, Germany, France and Italy. Of the 393 ads, 356 came through Google Ads—more than 90 percent of the total.
The figures describe ads observed on the sites studied. They do not establish that each advertiser knowingly chose to appear there. The report says brands are likely placing ads on the low-quality sites without knowing it.
Advertisers can exclude particular websites, but doing so takes research and ongoing maintenance. A growing list of pages therefore creates a practical challenge: even a company trying to avoid certain placements has to keep identifying and excluding sites as they appear.
Why Google sits at the center
Google has several roles in this system. Its Ads and AdSense services manage advertiser bids and deliver ads to websites. Google also influences which pages people see through Search, News and Discover. When those routes bring attention to pages carrying ads, the company’s advertising business is involved in the same ecosystem that its search services help shape.
For many Western publishers, Google is also a major traffic source. The source article argues that being penalized or ignored by Google can threaten a publisher’s business. That dependence complicates the response to spam: publishers rely on the search engine for visibility, while low-quality pages may seek the same attention.
The difficulty of drawing a line
Automatically generated material can add to the volume of pages competing for attention, making the open internet harder to manage. But a blanket rejection of AI-generated writing would sit uneasily alongside Google’s use of AI text in Search Generative Experience. The article also notes that Google has not said it will take blanket action against AI content if it is useful.
Google’s position has not always sounded the same. In April 2022, search engine spokesman John Mueller said AI content was automatically generated content that violated webmaster guidelines. “So we would consider that to be spam,” Mueller said. The article points to a practical complication: text generators are used across business and society, Google offers its own generator, Bard, and reliable identification of AI-written material may be costly to impossible.
The broader spam tactic predates the recent expansion of language models. Early GPT-3 users created generic blogs about life advice and self-optimization that attracted thousands of readers and comments from people who thought they were interacting with humans. NewsGuard’s research suggests that increasingly powerful and accessible models are now driving a much larger wave of this activity.
That leaves a problem with several connected parts: rapid content production, imperfect detection, labor-intensive ad exclusions and a search system that influences page visibility. The study’s findings show how those parts can meet on the same page, where an article generated with little human oversight becomes inventory for programmatic advertising.