Whether you like it or not, the internet has entered its AI-generated content era.
With the advent of powerful AI text generators like OpenAI’s ChatGPT, the process of optimizing content to be found on Google, known as search engine optimization (SEO), has been turned upside down.
That’s in large part due to these tools’ ability to churn out content at much faster rates than human writers — and at a fraction of the cost.
However, some entrepreneurs are hellbent on making a buck by repurposing existing content, laundering it through an AI algorithm, and passing it off as their own.
“We pulled off an SEO heist that stole 3.6 million total traffic from a competitor,” bragged Jake Ward, founder of a UK-based SEO content marketing agency called Content Growth, in a recent X thread.
The ruse was as straightforward as it was ethically dubious. By stealing a competitor’s sitemap, a file that tells search engines like Google how a website’s content is organized, Ward turned “their list of URLs into article titles” and generated “1,800 articles from those titles at scale using AI.”
Ward’s controversial X thread drew plenty of ire among users.
“I can’t believe you are bragging about this,” one user replied.
“And you’re proud of this?” another user wrote. “Pumping garbage to get to the top of the trash heap? What about trying to make actually good and useful content… oh that’s right, it’s hard.”
It certainly smacks of plagiarism, but unfortunately the law still has a lot of catching up to do. For one, actually proving that AI was used to repurpose content is far from straightforward. AI detectors simply aren’t equipped to reliably distinguish between text that was generated by an algorithm and passages that were penned by a human.
Even ChatGPT maker OpenAI admitted earlier this year that educators are out of luck when it comes to checking their pupils’ work for plagiarism.
The AI text generator Ward used for his “heist” called Byword, a company he cofounded himself earlier this year that brags openly on its website about an optional feature that allows customers to avoid “AI detection.”
“Enabling this feature will instruct Byword to write in a way that’s significantly more difficult for detectors to pick up on,” the website reads. “Byword does this by varying word and sentence structure in a way that differs from other AI content generators, making Byword’s content difficult to detect.”
Byword is built on OpenAI’s large language model GPT-4 and offers a variety of pricing tiers to its clients, which range from $5 an article to $2,499 a month for “unlimited articles” and a “dedicated server.”
The goal is to reduce the amount of effort it takes to generate content and “spend less time on publishing, and more time on strategy,” according to the company’s website.
But whether the content it spits out can pass the smell test remains to be seen.
We tried the tool for ourselves, generating a 1,600-word blog post based on the keywords “first exoplanet discovered.” Besides waxing poetic about how “advancements in astronomical research have revolutionized our understanding of the universe,” the blog fails to actually mention the first exoplanet to have been discovered until about three-quarters into the text.
The generated blog also features a hallmark structure of SEO-friendly headers and subheaders that unnecessarily break up the blog’s flow and seemingly solely exist to game Google’s algorithm, rather than guide the reader.
A separate request to generate an article on the keywords “melting eggs” — a prompt that has already led to plenty of hilarity on Google’s own AI-based search — led to a familiar word salad.
The 1,800-word blog titled “The Art of Melting Eggs: A Culinary Delight” goes into excruciating and nonsensical detail on how to best melt eggs for a “luxurious breakfast.”
“Unlike traditional methods of cooking eggs, melting eggs are cooked over very low heat for an extended period of time,” the blog reads. It even invites the user to dream of “waking up on a lazy Sunday morning” and treating themselves to some molten eggs.
A 1,500-word blog on “investing in Tesla” made zero mentions of CEO Elon Musk. Given his many outbursts that have sent the EV maker’s valuation on a rollercoaster ride, that seems like a pretty glaring omission.
We’ve come across plenty of instances of high profile publishers using text generators with less-than-stellar results. Earlier this year, BuzzFeed landed in hot water after being caught publishing entire AI-generated articles featuring awkward language and copy-and-pasted phrases.
Other less careful content farms simply forgot to delete the five-word phrase “as an AI language model,” which ChatGPT often uses in its answers.
Some companies are taking the trend to its logical conclusion by mocking up entirely AI-generated writers masquerading as humans. Most recently, Futurism found that Sports Illustrated was publishing allegedly AI-generated articles by authors who were themselves entirely AI-generated.
But with his “heist,” Ward isn’t even putting on any pretenses, and is treating content creation as nothing more than a numbers game.
His experiment paints a dire picture of the current online media landscape, with companies racing to find new ways of having their content rank on Google search results, slavering at display ad revenue or affiliate sales.
Given Ward’s success — at least judging by his own metrics — laundering existing content through AI generators isn’t just effective, but is incredibly easy to do.
Whether any human who may come across the content at some point actually benefits from all of this seems entirely beside the point.
In other words, this isn’t content aimed at human readers — it’s a deceitful ploy to trick search engines into wasting people’s time.
More on AI content: Sports Illustrated Published Articles by Fake, AI-Generated Writers