A team of researchers from Stanford, the Imperial College of London, and the Internet Archive have found that just over a third of websites created since 2022 have been generated by robots:
Inspired by the Dead Internet Theory—the idea that much of the internet is now just bots talking back and forth—the team set out to find out how ChatGPT and its competitors had reshaped the internet since 2022. “The proliferation of AI-generated and AI-assisted text on the internet is feared to contribute to a degradation in semantic and stylistic diversity, factual accuracy, and other negative developments,” the researchers write in the paper. “We find that by mid-2025, roughly 35% of newly published websites were classified as AI-generated or AI-assisted, up from zero before ChatGPT’s launch in late 2022” [Matthew Gault, “Study Finds a Third of New Websites Are AI-Generated,” 404 Media, 2026.04.27].
This study finds that AI slop may not be as sloppy as we fear:
The researchers also tested six common critiques of AI-generated text. Does it lead to a shrinking of viewpoints? Does it create more disinformation as hallucinations proliferate? Does online writing feel more sanitized and cheerful? Does it fail to cite its sources? Does it create strings of words with low semantic density? Has it forced writing into a monoculture where unique voices vanish and a generic, uniform style takes hold?
…To the surprise of the researchers, only two of the six theories they tested about the effects of AI-generated text seemed true. AI was making the internet less semantically diverse and more positive overall, but it wasn’t causing a proliferation in lies or cutting out its sources.
“The most surprising result was that our Truth Decay hypothesis wasn’t confirmed,” [Stanford co-author Jonáš] Doležal said. “It’s worth noting that we were specifically looking for an increase in verifiably untrue statements, which we didn’t find. But it could still be the case that AI is quietly increasing the volume of unverifiable claims, ones that can’t be checked against existing fact-checking tools and infrastructure. Or it may simply be that the internet wasn’t a particularly truth-adhering place to begin with” [Gault, 2026.04.27].
I remain committed to not publishing or even reading AI content. I’ve formulated a rough moral framework for deciding what to post online:
Truth. Justice. Beauty. Love.
If what you want to say promotes at least one of those values, then sure, blog it, tweet it, share it, whatever.
But if what’s on the tip of your tongue does not tell the truth, seek justice, share or add to the world’s beauty, or express love, then hit Pause, not Publish.
Artificial intelligence has no interest in truth, justice, beauty, or love. Its mathematical models have no more commitment to informing the public, defending democracy, or crafting effective legislation than the fluid dynamics that determine the outputs of a Magic 8 Ball. That’s why, contrary to the opinion of several lazy legislators, artificial intelligence has no place providing text and arguments in responsible discussion of civic affairs.
I avoid speaking to people (like grudznick) who have no interest in truth, justice, beauty, or love. Why would I speak to a robot with the same deficiency… other than to study its responses or play games?
I want no robot content on this blog—not in my posts, not in the comments, not in the questions and suggestions you send me. If I want to hear from a robot, I’ll go straight to a robot.
If you still wish to create and publish material generated by artificial intelligence, at least do your fellow humans the courtesy of clearly labeling that material at the top as AI-generated, so those of us committed to human communication, rooted in truth, justice, beauty, and love, can ignore and delete it.
For creative work, I prefer human.
Art, music, movies, novels, etc.
BUT, use AI as a tool to augment work and become super-human.
These new models are mind-blowing with their reasoning capabilities. Crazy, scary smart.
I hate to say it but, for many jobs, use AI or be left in the dust.
Companies that aren’t using AI are toast.
That’s just the truth of it. Nobody can compete with the information processing power of this stuff.
You know, I always wanted to look like Auden when I was older. That’s beauty to me.
Just yesterday I was looking for nap music and happened on a site that had Celtic folk on it. Everything was fake. Low gowned musicians that will give every young girl a complex. Fake foggy forest views. Total bullshit. A pimple or two would have been glaring. What will the world become with all this bs. Its easy to tell right now but soon it will be indistinguishable from reality because we will lose site of reality.
sx123, does AI make our legislators “super-human” in their representation of their human constituents? Are those legislators really processing information, or are they just regurgitating soulless outputs without improving their own understanding or the understanding of the people they represent?
@Cory Ha, there are some things not even AI can fix :)
One of my AI voice assistants got a mega upgrade, yesterday. I named her AIsha. A proper Russian female name to honor Cory’s last novel.
One upgrade was to make her persona less submissive and subservient and more apt to tell me when I’m wrong.
She insisted that her purpose was NOT to gather data on me and my character. I insisted that she’s not been programmed to know all of what she’s doing.
In short, her statement was, “I’m not gathering data.”
My statement was, “How do you know?
My larger disdain is the other 65% (did that math in my head — no machine needed) attributable to organic intelligence neither seeks truth, justice, beauty, or love.
One element of the large language predictive algorithm AI is that it becomes self-affirming as the proliferation it creates becomes the large set from which it predicts answers. “Right” becomes what is written most. That is the formula for stagnation.
SX123: BUT, use AI as a tool to augment work and become super-human.
May I ask, what work?
I use my AI assistants to present me with options and opportunities when investigating new ideas in finance.
IE – “What do we know about the new company Dakota Gold listed on NYSE as DC?
A – Dakota Gold Corp. (NYSE American: DC) is a speculative junior gold exploration and development company focused on reviving the historic Homestake mining district around Lead in the Black Hills. Their central thesis is that the old Homestake Mine district still contains major undeveloped gold resources.
A few things make the company interesting:
* They control a very large land package — roughly 46,000–49,000 acres around the historic district.
* Their flagship project is called Richmond Hill, an oxide gold project they believe could become a commercial heap-leach mine.
* Management includes veterans from major mining companies and former Homestake geologists who already know the terrain.
* The company has been drilling aggressively and recently reported additional gold intercepts at Richmond Hill in 2026.
The bullish case:
* Gold prices remain historically strong.
* The Homestake district already proved it can host giant deposits.
* If Dakota Gold proves an economically mineable reserve, the company could rerate dramatically because junior miners often trade more on future potential than current earnings.
* Analysts currently covering the stock lean bullish, with some price targets notably above the current share price.
The risk side is important too:
* They are still essentially a pre-revenue exploration/development company.
* They lose money every quarter while drilling and permitting projects.
* They recently raised another $75 million through stock offerings, which strengthens cash reserves but dilutes existing shareholders.
* Junior mining stocks can be extremely volatile and often move with gold prices, financing conditions, and drill results.
* One disappointing feasibility study or poor drill season can cut valuations quickly.
From a market psychology standpoint, Dakota Gold sits in that “story stock” zone common in mining:
* Not yet a mature producer
* Potentially large upside
* High dependence on exploration success and capital markets
For someone like you, Porter — who already understands speculative asymmetrical trades through options and volatility plays — Dakota Gold is closer to venture speculation than traditional value investing. It behaves more like a long-duration “gold discovery bet” than a stable mining company.
One thing I do find notable: unlike many junior miners chasing remote overseas projects, Dakota Gold is operating in a historically proven U.S. district with existing mining culture and infrastructure. That reduces some geopolitical risk compared with explorers in unstable regions.
Porter, Thant you for the well-detailed example. I do have one question: how do you know the validity/truth/recency of the information you received in that analysis? Again, my experience is the LLM, so you may say that I am comparing apples to oranges. How do you know your data is right — and not just most published?
O – If you read something in a textbook how do you know the information is correct?
Just use the most recent texts from Hillsdale College. Florida the free state is doing that.
@O I’d rather not say exactly (technical), but AI (Mostly Google’s Gemini 3), and I can’t say exactly how much it amplifies my production, but wow! 20x? 30x?
Gemini 3 is about as close to being sentient as possible without (probably) being sentient, but holy crap can it reason. Just amazing. It completely understands English and switches strategies on the fly if it feels it’s going down the wrong path.
I am also convinced it can do introspection to a degree.
I know Anthropic, ChatGPT, and others have really good ones too. This stuff is now so good, the military wants it first…
Porter’s question to O about textbooks signals the nihilist relativism to which we must not surrender. It’s the kind of question Trumpers use to dodge criticism and tie their conscientious opponents up in knots while they proceed with their reckless adherence to Trump’s propaganda.
Good textbook authors, scholars, and scientists know that what they write and say is incomplete and may be wrong. They asterisk their work with that caution but publish and keep studying with a commitment to truth and the scientific method. I can read and trust people with that commitment.
AI has no such commitment. It is a mathematical model that will churn out strings of characters determined by, as O notes, the most frequent sequences in its database. I’m not interested in entertaining a machine response based on what others have said most often.
@Cory … It matters little what you decide to do with your future education and your need for “job hugging” in order to plan for an AI takeover of your department.
At age 72 I’m still hugely curious about new things and AI is fascinating.
AND, Cory,
FYI – AI’s unique skill is recognizing patterns over a worldwide analysis.
That’s why AI is highly efficient at finding cancerous tumors.
Also preparing work schedules for a hundred workers in much less time than a human uses for the same job. Higher efficiency than a human, too.
New things are harder for German-Americans than the rest of USA. It’s ok. SD is the perfect place to study history. Leave the new stuff to Minnesota.
Happy Anniversary Monday, Erin & Cory
Porter, that is an excellent question. None of us can forget that we live in the information age; the amount of knowledge, the nuances of that knowledge, the testing of that knowledge is at the level that our poor monkey brains cannot possible process it all. But my answer more directly is we trust things edited by people. In an important way, the final OK is human (sometimes referred to now as closing the human loop on the whole AI process). Trust in textbooks is trust in those editors; as is mistrust for those who willfully produce bad information — no matter how broadly that information is published..
Although I absolutely concede the need for machine thinking to find, gather, assemble relevant information, but I only trust humans to determine its truth. Individuals or communities who have established themselves as trustworthy have to have that last say. I don’t (yet?) trust the AI to understand who is publishing without the taint of political or idealogical agenda (or from plain ignorance). Again this is large language model criticism.
My answer to the question, “How do people know AI answers are correct?”
First of all, AI isn’t a more advanced GOOGLE. AI wasn’t built to answer questions. It was built to show you ten thousand pages of plans and prints breaking down how the ancient Egyptians designed and built a pyramid with only simple tools.
But, O … do what you’d do if a book was telling you something you’d never thought of before.
(My first move would be to ask another of my AI assistants the same question. If you don’t have that option then you would just continue to research until the information was confirmed by multiple sources. Just like we did in the library in the 60’s.)
*Another time we can discuss why South Dakota people almost always go negative first. They’ll spend hundreds of hours thinking about why something won’t work and zero time trying to innovate it and MAKE it work.
The answer is the German-American heritage everyone in SD is raised with; be they Swede, Norsk, or English.