When ChatGPT launched in November 2022, millions of marketers and publishers around the world went into a frenzy. Suddenly creating short-form and long-form content became effortless. It didn’t matter whether you could write earlier or not. Now, you could just send a prompt to ChatGPT (or one of the many AI tools that came out after it), and voila! A piece of content was ready for publishing.
This party went on for a few months without any check until Google and the social media platforms started hitting back. In those days, publishers would push out thousands of AI-written pages every day, hoping a fraction of them would rank.
Most of that content would never be read by a human being. However, many of those got indexed, and the publishers hoped that some of those would continue to stick and bring business results.
Soon, Google and the other platforms this content was written for took action. The algorithms were strengthened, and low-quality AI pieces got rapidly devalued.
I’ve spent the last few years watching this play out across client work, and a pattern kept repeating itself, clear enough that it deserved a name.
As the quantity of content produced goes up, the thinking behind each piece goes down.
I call it the Law of Diminishing Quality.
The Law of Diminishing Quality states that as the volume of content a business produces increases, especially with AI, the amount of original thinking behind each piece decreases. Past a certain point, more output stops adding value and starts actively working against rankings, trust, and brand reputation.
– Anirban Kundu
The last three years have given us an unusually well-documented case study of exactly what happens when businesses ignore it.
Let’s take a deeper look at it.
The Math That Doesn’t Add Up
When AI writing tools became fast and cheap, the logic seemed obvious. More content means more visibility. More visibility means faster growth. Publish more, grow more.
Most people who followed that logic got the equation wrong.
More content only works if it is worth someone’s attention. When you create content without paying close attention to quality, you don’t get more marketing. You get more noise. Well-structured noise, grammatically correct noise, noise that reads fine on the surface. But it’s noise all the same.
This can get you short-term gains. A spike in impressions. A few more visitors. A dashboard that looks busier than it did last month.
But over time, both the platforms and your customers start to notice the difference. Algorithms are trained on engagement, and generic content simply doesn’t hold attention the way specific, thoughtful content does. Readers are trained the same way, just with better instincts.
The Mechanism Behind the Law
Here’s what’s actually happening underneath the pattern.
A team sets a content target. Ten blog posts a month. Fifty social posts. A hundred product descriptions. Early on, someone is thinking carefully about each one, because there’s time and attention to spare.
Then volume grows. AI makes it possible to produce ten times the output with the same headcount. But thinking doesn’t scale the way writing does. You can generate more words in less time. You cannot generate more original insight in less time, because insight comes from experience, not from a prompt.
The gap widens. Output goes up. Genuine thought per piece is going down. Content starts looking identical to everyone else’s because it’s drawing from the same pool of patterns AI models are trained on.
Nobody decides to publish worse content. It happens one shortcut at a time until the whole output becomes utterly forgettable.
What This Looked Like in the Real World
The last few years gave us several well-documented examples of exactly how this abuse played out and how expensive it turned out to be.
1. CNET
In late 2022, CNET began quietly publishing personal finance articles written by an internal AI tool, without clearly disclosing this to readers.
When outside journalists investigated, they found real errors, including a compound interest explainer that gave genuinely wrong financial advice and phrasing that appeared to be lifted from other sources without proper attribution.
CNET issued corrections to more than half of the affected articles. The damage didn’t stop there.
Wikipedia’s editors later voted to downgrade CNET as a source, specifically citing the AI-generated content period as unreliable, a rare and lasting credibility hit for a publication with nearly three decades of trust behind it.
Other properties owned by the same parent company, including Bankrate and CreditCards.com, were later found running similar undisclosed AI content programs around the same period.
2. Sports Illustrated.
In late 2023, Sports Illustrated was found to have published product review articles attributed to writers who didn’t exist, complete with AI-generated headshots and fabricated biographies. One fictitious author, “Drew Ortiz”, was described as an outdoors enthusiast, despite being nothing more than a stock AI-generated image.
The magazine’s parent company, The Arena Group, initially denied the content was AI-generated and blamed a third-party contractor, but the AI-generated author profiles quietly disappeared from the site once journalists started asking questions.
The fallout extended well beyond the specific articles: Sports Illustrated’s own union publicly said staff were “horrified”, and the company’s leadership saw a wave of executive departures in the months that followed.
A publication built for a century on the credibility of its bylines took a direct hit to the one thing it couldn’t afford to lose.
3. Grokipedia.
In late 2025, xAI launched Grokipedia, an AI-generated encyclopedia built entirely by its Grok language model, positioned as a rival to Wikipedia.
Its rise was extraordinary by any SEO standard: Google search clicks to the site went from just 19 in its first month to roughly 3.2 million by mid-January 2026, driven largely by rapid indexing of hundreds of thousands of AI-generated pages.
Then, in February 2026, its visibility collapsed almost as fast as it had risen. Multiple independent SEO analysts tracking the site, including Glenn Gabe, one of the most respected names in the field, confirmed the drop across Google’s organic results and its AI-powered search features alike.
Poynter’s forensic review of the content found part of the reason why: one Grokipedia article was a 96% textual match to its Wikipedia equivalent but carried zero citations, whereas Wikipedia’s version had twenty-two. The pattern has since been nicknamed “Mt AI” by Gabe, a sharp climb in visibility from scaled AI content, followed by an equally sharp collapse once search engines catch up to what’s actually behind the pages.
4. SE Ranking’s controlled experiment.
SE Ranking, an SEO software company, ran a direct, first-party test of this exact question.
They published 2,000 unedited AI-generated articles across 20 brand-new websites and tracked the results for sixteen months. The early numbers looked promising.
Within the first two and a half months, impressions climbed from 122,000 to over 526,000, and clicks more than tripled.
Then, around the three-month mark, performance reversed sharply: the share of pages ranking in the top 100 search results collapsed from 28% to just 3%. The pages remained technically indexed but became functionally invisible.
Sixteen months in, there was no recovery. The company’s own conclusion was direct: content that had been edited and refined by a human team continued to perform, while the fully unedited AI content saw no meaningful traffic for the rest of the year.
5. Mass-produced SEO content, more broadly.
When Google rolled out its March 2024 core update, the sites hit hardest shared a common profile: heavy reliance on AI-generated content, often accounting for 90% or more of a site’s pages, combined with weak or absent author credentials, generic stock imagery in place of original photos, and product reviews written with no evidence the reviewer had ever used the product.
Travel sites were especially exposed, with several losing more than 90% of their search visibility in that single update.
One documented case, tracked directly through Ahrefs data by an SEO analyst who manually reviewed dozens of affected sites during the rollout. It involved a large programmatic content site that had built its traffic entirely on AI-generated pages, reaching an estimated 10 million monthly organic visitors at its peak before suffering a sharp, sustained decline once the update took effect.
Gaming sites that had expanded into unrelated categories like recipes, publishing shallow, templated answers to every conceivable search query, saw some of their pages deindexed entirely.
The common thread across all three cases isn’t that AI was used. It’s that AI was used to produce volume in place of judgment, and in each case, the businesses involved paid for it in trust, traffic, or both.
How Google Fought Back
Google’s response evolved in stages, and the shape of that evolution tells you where things are heading.
Early on, Google’s public position was that it doesn’t matter whether content is written by a human or a machine, what matters is whether it’s genuinely helpful. That’s still technically the policy. But in practice, Google had to build systems capable of telling the difference between AI-assisted content backed by real expertise and mass-produced AI content built purely to occupy search real estate.
The Helpful Content Update, first rolled out in 2022 and steadily expanded through 2023, was the first serious version of that system.
Then, in March 2024, Google folded helpful content signals directly into its core ranking algorithm, making the evaluation continuous instead of a periodic event you could wait out.
Google has since reported roughly a 45% reduction in low-quality, unoriginal content showing up in search results since that integration.
The lesson isn’t “don’t use AI.” Google has said repeatedly that automation itself isn’t the issue, and plenty of AI-assisted content ranks well when it’s genuinely useful.
The narrower, sharper lesson is this: content produced primarily to occupy search space, without real expertise or original value behind it, is now actively and systematically identified and demoted.
What used to be a slow, occasional correction is now a constant, ongoing filter, and it isn’t getting more forgiving.
A Simple Test Worth Stealing
At Inovaticus, we apply one test to every piece of content before it goes out for a client.
We ask, ‘Does this contain something that only this specific company, with its specific experience, its specific clients, and its specific way of seeing its industry, could have produced?’
If the answer is no, if the piece reads like something that could have been written about any client in any market, we rewrite it.
This test is deliberately simple. You don’t need a content audit tool or a scoring rubric to use it. You need to be honest about whether what you’re publishing could have come from a competitor with a few words swapped out.
Most content fails this test quietly. Nobody notices in the moment, because it still looks fine. It’s only over months that the gap becomes visible: in traffic that plateaus, in leads that dry up, and in a brand that starts to feel interchangeable with everyone else in its category.
How to Get Over the Law of Diminishing Quality
None of this means you should publish less. It means you should publish deliberately.
- Anchor every piece to something only you know. A client result, a mistake you made and learnt from, and a pattern you’ve seen repeatedly that outsiders wouldn’t notice. If a piece doesn’t contain at least one thing a competitor couldn’t have written, it’s a candidate for the pile that’s quietly hurting your site, no matter how well it reads.
- Separate drafting speed from thinking speed. AI can compress how long it takes to write a paragraph. It cannot compress how long it takes to decide what’s actually worth saying. Protect that second part. Spend the time you saved on drafting back on deciding what the piece should argue, not on producing three more pieces just like it.
- Track outcomes, not output. A content calendar that’s always full feels productive. It tells you almost nothing about whether the content is working. Track how many pieces generated a real lead, a real reply, or a real next step. Most teams caught in the Law of Diminishing Quality are optimizing for the wrong number.
- Let some content stay unpublished. Not every draft deserves to go live just because it exists. If it reads like something anyone in your industry could have written, it’s better left as a discarded draft than published as a page that quietly drags down trust in everything around it.
What Actually Survives
The internet has more content in it today than at any point before. Most of it will be ignored, and increasingly, most of it will be actively filtered out by the same platforms that used to reward volume.
CNET, Sports Illustrated, and the wave of AI content farms hit by Google’s 2024 update all learned the same lesson at different scales: the moment content stops being about the reader and starts being purely about occupying space, someone eventually notices, whether that someone is an algorithm, a journalist, or a customer.
What survives that filter, and what a reader remembers after they close the tab, is the same thing it’s always been: something that could only have come from one specific source. AI changed how fast you can get there. It didn’t change what “there” actually is.
So here’s a question worth sitting with, the same one we ask before every piece of content goes out: what’s the last thing you published that only you could have written?
Leave a Reply