Commentary To see where this dump truck is going, let’s first follow the trail of debris.
It’s hard to trace, but the impact of Internet content mills, which flourished until about 2010, is still very much visible.
The net effect of content being generated daily at the rate of ten to thirty pieces on special topics – all by non-specialists via the guiding hand of Google Trends – led to an internet in 2010 that was riddled with fluffy (if not nonsensical) ), keyword-filled articles that offered little useful information and, in many cases, a lot of advice and information that was simply wrong.
And because content mills naturally spawned more content mills (and why not when the worst offender sold Associated Content to Yahoo! for $100 million), what happened next was inevitable.Those new content companies simply copied what they found at larger content factories, using the Internet of the day as a training set, so to speak.The cycle of bad articles with few details or worse, inaccurate details was repeated over and over until it became difficult to distinguish one article from another unless it was found on one of the handful of edited, reputable sites.
The name of the game for these early content companies was pure volume.Revenue from the ad network (Google Adsense, etc.) was already declining in 2005, but with thousands if not millions of articles each bringing in maybe three cents a day, the money wasn’t bad.
For a content factory with 200,000 articles, that was a tidy $2 million in revenue with ultra-low overheads.Hosting wasn’t hugely expensive, web design was easy with open source CMS tools like WordPress, Drupal, and others, and most importantly (and ultimately the most disastrous) was that the bulk content could be purchased for mere pennies per item from offshore stores.
This model quickly flooded the internet with poorly written nonsense, much of which is still searchable in its original form or has been refurbished even worse.Google had to step up its game to filter around this and learn how to deliver quality content instead of the magic combination of keywords that content factories could exploit.
The problems with content mills are obvious, especially all these years later, but it was all on a human scale with the limitations of “slow” writers and keyword stuffers.
The future presents us with a new issue – one that could destroy the way we use the internet for good.
Let’s do a little math
Imagine it’s 2006 and you’re in the content mill business.You are on top of your game.You have a team of 100 writers in India who earn the equivalent of $10 a day to write and post twenty pieces of 400 words (topics dictated by Google keyword trend data vs expertise, etc).
Your daily salary cost is about $1,000.Every day your content mill posts 2,000 pieces of “unique” content, 365 days a year, and each of those articles, assuming a good ranking in the search engines (which was easy to play with keyword tricks back then), each of those articles will earn three cents per day.
And while we’re using nice, rounded numbers for convenience, consider these yearly numbers (yearly because you only have to run this business for a year, the Adsense money will come in anyway, at least for a while):
Salary for writers who create, post and tag 20 articles a day will cost you $365,000 a year.They generate 730,000 pieces of content over the course of the year, worth $10.95 each (assuming three cents a day for 365 days).And all that, which is pretty straightforward for you Western Content Lord, means you have a business that generates about $8 million a year.
Oh.But you have to deduct hosting and such.
Let’s call that five thousand.The big ugly cost? All those “expensive” writers.And you think to yourself, who needs them?
Well, you don’t.
Because oh boy, there is a new business model for content mills.And while their early 2000s predecessors made the Internet obnoxious and full of unwanted articles that hit the keywords and word count without saying anything, this one is disruptive enough to turn the Internet into a complete mess.
And not just waste from a content perspective, but also from a how-the-business-of-the-internet-works.
Put the S in IoS
This new business model is already unfolding.You have probably read many articles generated by GPT or similar AI models.
The reason you probably didn’t notice is because they aren’t bad.Well, you to think they’re not bad, but that’s because you’re devoid of the Internet of Shit (IoS) that the content mills created, which trained us to lower our expectations of information consumption.
The problem is that these AI-generated articles have to get their information from somewhere in sufficient volume to produce new information clones cloaked in slightly more eloquent language.And where do AI training algorithms get all this from? From the IoS of course.
Counting more, let’s assume that 10 percent of IoS-derived training data contains factual errors.As AI trains, retrains, and retrains, those errors increase.
And assemble.And multiply and within a decade of retraining based on bad, weird, oddly worded, and increasingly incomprehensible data, we’ll have a real IoS.
And math is again super important – and so is volume.
For example, a single operator of a content mill on the scale of Western Content Lord can use free tools to generate content as fast as human operators can plug it in with a simple suggestive phrase.The same team of 100 workers can import 300 pieces per day.
They don’t write it, they just ask ChatGPT.They can ask it to use keywords as a mofo and generate keywords for that matter.
Ultimately, that process of ChatGPT (like one of many examples) will have API hooks to publish output directly to WordPress or to whichever CMS Content Lord chooses.
When that unification from AI to CMS platform is complete, so is the circle: the internet just talks to itself.
The race to the bottom
What Western Content Lord and competitors don’t realize is how soon that race to the bottom will begin – and soon.
Google Adsense and every other ad network on the planet will recognize the flood and reduce the reward for a click or view to next to nothing.And then it will be nothing, but not before Google and the like rush to blacklist well-known AI content factories.
But too many will pop up too soon.For example, it will just be easier for a Google to create a safe list of well-known publishers, backed by floundering people.
Very well, you think, balance has been restored! Not so much.
To keep up with all the innovations in search that push those IoS results to the bottom for you, Google costs money, billion-dollar AI training, and significant, frequent retraining on the corpus of the Internet.That corpus will be infected fast and furious and how will the search giants pay for all that search innovation? Through ad revenue.
The search advertising giants like Google may seem to hold their noses and accept the results of the queued content mill because it is in their economic interest to do so.But what if the pool of “acceptable” content shrinks by 95 percent?
The exponential rate of internet shittification
We’re going back to the theme of math and volume and so on to get to the main point: the information peril is an exponential problem.One set of errors generated, which are then repeated by content factories for a decade, means that those problems are trained and reinforced in the AI language model from the internet corpus.
It’s one thing to live in an age of fake news, partly because it’s obviously fake to most thinking people.
If the internet repeats a mistake often enough, it becomes the truth, and that’s the most insidious accidental result of it all.
It would personally make me feel better to end this piece with some sort of “fight power” message, but frankly, the cat is out of the bag at this point.
The content factories can be content with earnings per article measured on a five-year value that can be as little as 0.05 cents over the term.But who cares, right? It’s free money.Hosting is cheap, a CMS is free, and as long as there’s advertising money, it’s worth the passive income.
This is apparently the internet you deserve.®.
