The Smear Campaign Advertisers Already Paid For

2026-08-28

NewsGuard's researchers have now catalogued 3,749 websites publishing news and information generated almost entirely by AI systems, across sixteen languages, with no meaningful human editing. A year ago the count was roughly 3,000. It is growing by 300 to 500 new sites a month. These sites carry deliberately forgettable names — Times Business News, Business Post, that general shape — built to look like local or trade outlets rather than the automated content mills they are. And by NewsGuard's estimate, more than 140 well-known companies are running programmatic ads on them right now, without knowing it, contributing to roughly $13 billion wasted globally each year on advertising that funds exactly this kind of site.

That last part is the detail worth sitting with. This is not a fringe problem running on goodwill and a hobbyist's server bill. It is a self-financing industry, and the money keeping it running comes from the same companies most exposed to what it eventually publishes.

The attack no longer needs an attacker with skills

For most of the internet's history, publishing something false and damaging about a company required at least a little friction: someone willing to write it, a site with enough traffic to matter, some reason to think it would be believed. That friction was never a strong protection, but it was something. A person had to show up.

Automated content mills remove the person. Connect a generative AI system to a content management system, point it at a topic, and it will produce dozens of plausible, search-optimised articles a day without anyone reading them before they publish. Reputation-management firms that track this space are direct about what that enables: a disgruntled former employee, a competitor mid-negotiation, an activist group with a grievance, can now commission a wave of damaging, specific-sounding content about a named company for the cost of an API connection and a few hours of setup. No editor to talk them out of it. No publication with a reputation of its own to protect. Nothing that resembles the old cost of doing this kind of harm.

Real-looking is the whole design

What makes these sites effective isn't sophistication — it's plausibility. The output mimics the structure of ordinary journalism: a headline, a dateline, quotes attributed to named people, references to real events woven around the fabricated ones. A reader skimming a search result has no obvious reason to distinguish it from a wire story. And a reader is no longer the only audience that matters. Search systems, AI browsing agents, and chatbots answering questions about a company all draw from the same open web these sites are built to game. A fabricated story doesn't need to convince a human directly. It only needs to get indexed once, somewhere a system compiling an answer about a company might reach for it.

That is the mechanism that turns an SEO problem into a reputational one. Content built purely to capture ad impressions was always a nuisance. Content built to capture ad impressions while describing a named company in specific, damaging, invented terms is a different category of exposure entirely — and the tools to produce it at scale now cost next to nothing.

Takedown was built for a world with fewer, slower sites

The standard response — monitor for mentions, send a takedown notice, wait for the platform to act — assumes a fixed target: one article, one URL, one owner who can be identified and pressured. Content-farm output doesn't behave that way. It republishes under a new domain as fast as the old one gets flagged. There is often no clear owner to name in a notice, no editor to appeal to, no single page whose removal ends the problem. Reputation specialists who deal with this describe it as a hydra: cut off one head, and the economics that made it cheap to publish in the first place make it just as cheap to publish again somewhere else.

Static defences were built for a world where damaging content came from a handful of identifiable sources, publishing occasionally, with something to lose. Three thousand seven hundred and forty-nine sites, adding hundreds more a month, funded by companies who don't know they're paying for it, is not that world. It's an assembly line, and it runs whether or not anyone is watching it that week.

Which leaves an uncomfortable question for anyone responsible for how a company is actually perceived: if a fabricated story about your organisation surfaces today, disappears tomorrow, and resurfaces next month on a domain that didn't exist yesterday, what account of any of it will still be standing by the time someone needs to prove what was said, and when? Most companies find out the answer to that question is "none" — usually while they're already trying to explain a claim they can no longer locate, to someone who saw it once and remembers it clearly.