How AI-generated political ‘ghost creators’ gamed YouTube

AI isn’t just writing blog posts and product descriptions anymore. It’s now being used to mass-produce political content that looks real, feels personal, and reaches millions of people—without any genuine grassroots movement behind it.

A recent investigation uncovered a network of YouTube channels pushing sensational, anti-Democrat videos using AI-written scripts and low-paid actors hired from platforms like Fiverr. The goal wasn’t to win elections or push a clear ideology. It was simpler—and in some ways more worrying: game the algorithm, go viral, and make money.

What was actually happening behind these channels?

On the surface, the videos looked like typical political commentary: a person sitting at a kitchen table or in a casual home setting, talking straight to the camera about the latest scandal or outrage in Democratic-led cities and states.

Titles were pure clickbait, like “Mayor Mamdani panics as Taylor Swift triggers a $2.3 billion celebrity exodus” or “AOC freaks out as Riley Roberts caught running her entire operation.” These videos racked up hundreds of thousands of views each, with some channels collectively passing 45 million views and 90,000 comments.

To an ordinary viewer, they looked like independent, frustrated citizens making political YouTube content. In reality, they were part of a coordinated network.

Meet the ‘ghost creators’

The people on camera weren’t political activists or journalists. They were actors and spokespeople hired through gig platforms like Fiverr and Backstage, often paid as little as $26 per video.

They weren’t writing the scripts. Their job was simply to read them.

This is where the term “ghost creators” comes from—similar to ghost kitchens in food delivery apps. The face is real, the performance is real, but the identity and story behind the channel are essentially manufactured for engagement.

How AI powered the content factory

Behind the scenes, a YouTube growth hacking company was using AI to generate or heavily assist the writing of these scripts. Large language models were used to:

• Churn out repetitive, emotionally charged narratives

• Recycle the same themes, phrases, anecdotes, and even music across multiple channels

• Quickly adapt to trending topics and names (like Gavin Newsom, AOC, Bernie Sanders, or Zohran Mamdani)

The scripts were then handed to actors, who recorded videos that felt more human and less synthetic than fully AI-generated avatars. This approach had two big advantages:

• Harder to detect: Real human faces and voices are less likely to trigger automated moderation or AI-content filters.

• More believable: Viewers are more inclined to trust a person who looks like an everyday citizen than an obvious AI animation.

Gaming YouTube’s algorithm

The network was traced back to a young YouTube growth hacker running a business that promised to be a “strongest alternative to paid ads.” His strategy was simple: use AI and automation to spin up “ghost creator” accounts for brands and clients, then ride YouTube’s recommendation system for maximum reach.

The channels followed the same formula:

• Target highly polarizing, political topics

• Focus heavily on negative stories about Democrats and left-leaning politicians

• Use dramatic thumbnails and headlines designed to provoke outrage

• Recycle music, visuals, and narrative beats across multiple channels

Because political outrage content tends to drive comments, watch time, and shares, YouTube’s algorithm rewarded these videos with significant reach. One of the most successful channels outperformed major media outlets like the Washington Post and The New Yorker in total views over a short period.

Why the content skewed right-wing

Interestingly, the people running the network didn’t seem to care much about the actual politics. They reportedly experimented with different angles, including left-leaning content that criticized conservatives and even an anti-Trump persona.

Those experiments flopped.

What consistently performed well was content attacking Democrats, leftist policies, and progressive figures. That’s where the algorithm rewarded them—and that’s where they doubled down.

This raises uncomfortable questions:

• Is YouTube’s recommendation system inherently more favorable to right-wing outrage content?

• Or is the audience for that kind of content simply more engaged, more reactive, and more likely to feed the metrics the algorithm optimizes for?

The answer is probably a mix of both platform dynamics and audience behavior, but the outcome is clear: if you optimize purely for engagement, you’re likely to end up amplifying sensational, polarizing, and often misleading political narratives.

AI content detection accidentally uncovered the network

The operation might have stayed hidden longer if it weren’t for AI content tracking tools. A company monitoring AI-generated material online noticed these channels because they were using so much AI in their workflows that they triggered detection systems—even though the videos themselves featured real humans.

Investigators then dug deeper and found:

• Repeated use of the same scripts and story structures across multiple channels

• Shared music and editing styles

• Actor profiles on Fiverr and other platforms matching the faces in the videos

• An internal dashboard with YouTube analytics that had accidentally been indexed by search engines

This kind of forensic work mirrors broader concerns in the AI world about how to reliably detect synthetic or AI-assisted content. It also echoes other recent stories, like how a fake “world’s best AI model” exposed weaknesses in benchmarking and evaluation, as covered in this deep dive into AI benchmarks.

YouTube’s response

Once the story came to light, YouTube removed the accounts tied to the network. The company cited violations of its policies against spam and deceptive practices and confirmed that at least 20 channels were terminated.

But the takedown doesn’t fully solve the underlying issue. The same incentives still exist:

• AI makes it cheap and fast to generate endless political scripts.

• Actors are easy to hire at low cost.

• Algorithms still reward engagement, regardless of whether the content is authentic or coordinated.

This isn’t just about one network

The story fits into a broader pattern: AI is lowering the cost of producing convincing, emotionally charged content at scale, and growth hackers are quick to exploit that.

We’ve already seen AI-generated influencers, including politically themed personas like a fake “sexy MAGA nurse” created by a student simply trying to make money. The same pattern keeps emerging:

• Try multiple angles.

• See what the algorithm boosts.

• Double down on whatever drives the most outrage and clicks.

In this sense, AI doesn’t have a political agenda—but it supercharges whatever the engagement economy rewards. As other experts have noted, anxiety about where this leads is justified; even AI leaders like Sam Altman have acknowledged growing public concern, as discussed in this overview of AI anxieties.

The new face of political clickbait

The most unsettling part of this story is that it isn’t driven by a government psy-op or a sophisticated political operation. It’s driven by people who treat politics like a niche in the creator economy—no different from tech reviews or fitness tips.

They’re “engagement junkies,” focused on views and ad revenue, not on the real-world consequences of what they’re spreading. And thanks to AI, they can:

• Scale production dramatically with minimal effort

• Quickly pivot to whatever narrative is trending

• Mask coordination behind a sea of seemingly independent faces and channels

What this means for viewers

For everyday users, this makes the information environment much harder to navigate. A few practical takeaways:

• Be skeptical of “random” political channels: If a channel appears out of nowhere, posts only highly emotional political content, and has no clear personal history or transparency, it may be part of a network like this.

• Look for repetition: Identical talking points, music, and story structures across different channels can be a red flag.

• Check the incentives: If the content feels designed to make you angry more than to inform you, it’s probably optimized for engagement, not accuracy.

The bigger picture: AI, algorithms, and democracy

This case is a preview of where AI-assisted media is heading. As tools get better and cheaper, it becomes easier to flood platforms with realistic-looking content that has no real connection to the communities or opinions it claims to represent.

The combination of:

• AI-generated or assisted scripts

• Low-cost human actors

• Engagement-optimized algorithms

creates a powerful system for manufacturing political narratives at scale. It blurs the line between genuine grassroots sentiment and algorithmically engineered outrage.

As more people cut the cord, abandon traditional news, and get their information from platforms like YouTube and TikTok, this kind of AI-powered political clickbait isn’t just a curiosity—it’s a structural risk to how people understand politics and make decisions.

We’re still early in figuring out how to detect, regulate, and respond to this kind of content. But one thing is already clear: AI has given growth hackers a new playbook, and they’re wasting no time using it on our political feeds.