Fraudsters Built 200 AI Slop Sites and Left Their Prompts in the JavaScript

If you're going to build an industrial-scale AI content fraud operation, there's one thing you probably shouldn't do: leave your entire playbook in the page source code where anyone with a browser can read it.

The operators behind AutoBait did exactly that.

200 Sites, Millions of Impressions, Zero Proofreading

DoubleVerify's Fraud Lab — researchers Arik Nagornov, Merav Geles, and Lia Bader — uncovered a coordinated network of more than 200 AI-generated websites designed to look like independent lifestyle blogs. Each site pumped out clickbait slideshows optimized for one purpose: harvesting programmatic ad revenue from legitimate brands that had no idea where their budgets were going.

The scale was methodical. Each page featured roughly 56 slides. Each slide carried up to 8 ad banners. The banners refreshed every few seconds for maximum impressions. A small team could operate hundreds of sites simultaneously with minimal oversight. The network generated tens of millions of impressions.

Production cost per slide: four cents. Total cost per page: less than $2.25. Revenue per page: multiples of that, courtesy of the programmatic ad supply chain that stands between a brand's media budget and wherever the ads actually end up.

The Prompts They Forgot to Hide

Here's where it gets entertaining. The AutoBait operators embedded their AI content-generation instructions directly in the site's JavaScript — visible to anyone who opened developer tools or viewed the page source.

The prompts were a masterclass in manufactured engagement. The LLM was instructed to frontload the first 5-10 slides with "the most sensational or shocking points." Headlines had to be 4-7 words and "ultra-literal." Each slide paragraph ran 50-70 words, calibrated to trigger "fear, anger, shock, relief."

The image prompts were even more deliberate. The AI was told to generate photos that appear "casually taken on a smartphone by a real person — unfiltered, unstaged, and emotionally authentic." The instructions explicitly stated images should "NOT look artificial, stylized, or generated by AI."

In other words, the operators knew exactly what they were building — synthetic content engineered to be indistinguishable from real media — and they documented the entire process in a file that any browser could read. It's the content-fraud equivalent of leaving the heist blueprints taped to the bank's front door.

Why B2B Advertisers Should Pay Attention

If you're running programmatic campaigns — and most B2B marketing teams are, even if they don't think of it that way — AutoBait is a case study in where your money might actually be going.

The problem isn't the existence of AI-generated content. That genie left the bottle two years ago. The problem is the economics. At $2.25 per page and thousands of pages per month, a single operator can create an entire media network that looks legitimate enough to pass through programmatic ad exchanges. The sites appear in ad inventory alongside real publishers. The bidding algorithms see impressions and click-through potential. The brands see reach numbers in their dashboards. Nobody sees the slideshow about "10 Foods That Are Slowly Destroying Your Gut" until someone actually clicks through.

For B2B teams spending on display and content syndication, the implication is straightforward: if you're not actively monitoring where your ads render, you're probably subsidizing at least a few AutoBaits. The programmatic supply chain is long enough that most buyers never see the final destination.

The Bigger Picture

DoubleVerify named this category "made for advertising" sites — MFA for short. AutoBait is the most brazen example so far, but it's the economics, not the sloppiness, that should concern enterprise marketers. Four cents a slide. Eight ads per slide. Refresh every few seconds. The math works even at trivially small audiences, and AI has made the content production cost approach zero.

The exposed prompts are almost a gift. They show, in the operators' own words, exactly how synthetic content is engineered to manipulate engagement metrics. That's useful information for any team trying to figure out why their programmatic campaigns look great on paper and accomplish nothing in practice.

Most fraud operations are invisible. This one accidentally published its own user manual.