The Content Scissors

We made 86 articles in one night.

Twenty-one AI agents running in parallel, each assigned a topic, each churning through research, structure, draft. By morning, another website had grown from 34 posts to 120. Human hours: 20 minutes to describe two prompts - one for AI driven discovery of hot topics fitting to the homepage with minimal overlap of already existing topics and one prompt to do the actual execution in parallel.

I want to be honest about that, because it’s the exact thing this article is about. Not as a success story. As evidence.

The Myth About Who Invented Slop

The prevailing narrative is that AI created the content quality problem. That before ChatGPT, the internet was a garden of careful, considered writing. That some threshold was crossed in 2023 and everything got worse.

That’s wrong.

Human slop has existed for as long as there’s been a commercial incentive to publish. Tabloids running unverified sensationalism. Clickbait masquerading as journalism. SEO farms churning out 500-word listicles with no original thought. Bad authors with content mills. Ghost-written celebrity books. Advertorial that looked like reporting.

R. Paulo Delgado, a freelance writer who recently announced he’s done with the profession, wrote about this more honestly than most: he wasn’t a great writer. He was a mediocre writer with strong domain knowledge. That niche existed because producing adequate tech content still required a human to sit down and do the work. That labor bottleneck was the quality floor. Not actual quality. Just friction.

AI didn’t invent slop. It removed the labor tax on producing it.

That’s a different thing — and a more important one to understand.

The Labor Tax and What Happened When It Disappeared

Before AI, even mediocre writing cost something. Four to eight hours per article. The effort constrained volume. A bad author could produce maybe 200 articles a year. A content mill with ten writers could do 2,000.

After AI, that constraint is gone. One person with a prompt can produce 10,000 articles a month without breaking a sweat. We produced 86 in a night. Neither required originality. Both required almost no time.

The supply curve didn’t just shift. It went vertical.

Ahrefs analyzed 900,000 newly created web pages in April 2025 and found that 74.2% of them contained AI-generated content. Separately, a study tracking 65,000 URLs showed that AI-generated articles briefly outnumbered human-written ones in November 2024. The two have been roughly equal since — not because AI production slowed down, but because the existing human baseline of content is still enormous.

I wrote about the quality paradox before: AI made the floor free without moving the ceiling. What I didn’t fully confront there was what happens to the demand side at the same time the supply side goes vertical.

The Scissors

Supply going up. Demand going down. Two blades moving in opposite directions.

That’s the thing nobody talks about loudly enough: while content producers were scaling production, the platform that delivered audiences — Google — was quietly becoming the final destination itself.

Google’s AI Overviews arrived at scale in May 2024. By May 2025, zero-click searches had jumped from 56% to 69%. Over 58% of all Google searches now end inside Google’s own interface — never reaching an external website. Even if you rank first. Even if you publish something genuinely good. The search engine that built the content economy by rewarding good content with traffic is now absorbing that traffic for itself.

The click-through rate for queries with AI Overviews has dropped 61% — from 1.76% to 0.61%. Business Insider saw a 55% drop in organic search traffic between April 2022 and April 2025, severe enough to cut 21% of their staff. Google traffic to global publishers fell by a third over 2025. Google upgraded AI Overviews to Gemini 3 in January 2026, now reaching over 2 billion monthly users, and most of them never leave.

The content scissors: supply vertical, demand horizontal

People who ask AI for answers don’t need articles. Platforms that used to monetize content creation by delivering audiences have absorbed that audience. The intermediary — the article, the blog, the content creator — is being bypassed on both ends.

Supply scaling infinitely. The audience that would read the supply: shrinking.

What the Demand Side Actually Looks Like Now

It gets more nuanced than “fewer clicks.”

Reader trust in AI-generated content is split in ways that matter. The Reuters Institute’s 2025 survey across six countries found that people think AI makes news cheaper and more up-to-date — but also less trustworthy (-19 net) and less transparent. People know something changed. They just can’t always locate where in the article it changed.

Here’s the uncomfortable data point: a 2025 study found that in blind tests, readers often preferred writing from AI trained on copyrighted books over expert human writers. For science fiction, actual AI-written narratives were rated more enjoyable than human-written ones. For news and informational content, human-authored pieces showed substantially higher trust and brand resonance — but only when authorship was disclosed.

The critical word is disclosed. When readers don’t know the source, the preference gaps shrink dramatically. People cannot reliably distinguish good AI writing from mediocre human writing. Which means the quality floor has shifted, but the ceiling question is open: not because readers don’t care about quality, but because they can’t see it until someone tells them to look.

Pablo Andreu, in the comments of the Delgado article that sparked this post, said something I keep turning over: “It’s not that they want slop but rather that they can’t tell the difference. Continued use of AI will only further dull their ability to discern.”

That’s the taste erosion thesis. And it might be right.

Three Scenarios for Where Reader Taste Goes

Scenario one: full adaptation. Readers accept AI-generated content as the new normal. Taste adjusts downward to match supply. Writing quality becomes like pop music production — technically proficient, formulaically competent, indistinguishable to most ears. The mass market settles here without noticing. Nobody mourns what they can no longer perceive.

Scenario two: bifurcation. The market splits cleanly. Most readers consume AI-generated content, often without knowing it. A small premium market survives for human-crafted work — specialist newsletters, technical deep-dives, long-form journalism — behind paywalls or in niche communities where audiences already trust the author. Substack’s survivor pool is this market. It’s real. It’s small.

Scenario three: taste renaissance. The flood of mediocrity creates demand for its opposite. People, overwhelmed by sameness, start actively seeking writing with a specific voice, a genuine perspective, a recognizable human intelligence at the keyboard. This is the optimistic scenario. It’s historically plausible — every content explosion has eventually created premium pockets. The printing press gave us both pamphlet farms and Montaigne. Mass media gave us tabloids and the New Yorker.

My honest read: scenario two is already happening. Scenario one is the mass market floor — it’s where most of the internet is heading and most readers won’t notice. Scenario three will happen, but slowly, for a small segment, and probably too late for most professional writers who couldn’t wait it out.

The paywall problem makes this worse. Premium content survives by charging for access. But the effort needed to find it and the cost to unlock it are both higher than the zero-cost AI alternative. The masses won’t cross that barrier for most topics. For entertainment, for news, for how-to content — the friction is simply too high. The premium market will be smaller than the people who care about quality would like to believe.

Can You Earn a Living from Content Anymore?

Delgado couldn’t. His honest conclusion: most companies don’t want high-end writing. They want slop.

The economics are stark. If a company can produce 50 adequate articles a month for the cost of an AI subscription, why pay a human writer for 10 better ones? The classic argument — “better quality drives more traffic” — collapses when the platform that delivers traffic stops delivering it to anyone.

The mass content market — the SEO article farm, the informational blog, the how-to industry — is being compressed from both sides simultaneously. Supply is free. Distribution is captured. The margin for the human in the middle has effectively disappeared.

What survives? The specialist. The writer whose value isn’t the writing but the knowledge it encodes. A cardiologist writing about a surgical technique that surprised her. An engineer writing about the architectural failure that cost three months and $200k. Someone who has built and broken 18 AI agents in production and can tell you exactly where they fail and why.

The writing is a delivery mechanism for something that can’t be faked: real, specific, irreplaceable experience.

The uncomfortable truth for most professional writers is that “professional writer” was always two things bundled together: craft (the skill of putting sentences together) and knowledge (having something worth saying). AI separated the bundle. Craft is now cheap. Knowledge — genuine, specific, hard-won knowledge — is still scarce.

Delgado was a mediocre writer with strong tech knowledge. AI ate his craft bundle first because it was the thinner protection. His knowledge bundle was real but narrow — specific enough to differentiate him from a general content mill, not specific enough to differentiate him from a well-prompted model with access to current documentation.

The writers who survive are the ones where the knowledge bundle is irreplaceable. Not “I know about React” — AI knows about React. “I’ve seen what happens to React teams at scale when they ignore the server component migration, and I watched it happen twice” — that’s harder to replicate.

The Honest Admission About 86 Articles

Those 86 articles we made overnight? They’re not this.

They’re real — not garbage, not empty padding. They cover topics clearly, organized well, useful for a specific audience. But they’re not built from personal experience. They’re built from synthesis: organizing existing information into readable form on topics that people search for. Good enough. Adequate. Functional.

We know exactly what they are. We didn’t pretend otherwise. The website they went to is an experiment in whether AI-scaled content can reach niche audiences at all, given the demand collapse we’ve been describing. The results will tell us something honest about whether the supply side of the scissors can still find any audience even with the demand side contracting.

The experiment is worth running. But it isn’t a model to replicate for every site.

Where That Leaves Us

The audience here — CodeWithAgents.de — is different. Engineering leaders evaluating AI adoption for their teams. Senior developers considering whether agentic workflows are worth the architectural cost. People who have read enough to know the difference between someone who actually built something and someone who prompted their way through a summary.

They will notice if I generate this post. Not because they run AI detection tools — they don’t. Because the specific detail won’t be there. The specific frustration from session 9 when the full pipeline first ran end-to-end. The specific number: 18 agents, then skills replacing agents, then the architecture flipping again. The colleague who asked one question that broke three months of confident architecture. That texture is irreplaceable because it isn’t composable from training data. It’s mine.

AI closes a specific gap for me: I’m a competent writer, not a great one. I think clearly but my first drafts are rough around the edges. The post you’re reading has been shaped by AI — structurally, rhythmically, for flow and clarity. But the ideas, the experiences, the specific technical details that make it worth reading — those didn’t come from a prompt. They came from two years of building something real and paying close attention to what happened.

That’s the only content model that survives the scissors. Not more content. Not better-prompted content. Content where the value lives in irreplaceable knowledge, with AI as the production infrastructure to get that knowledge into a shape someone skeptical will actually finish reading.

The vision statement for this site says it plainly: no AI slop, no 10-tips listicles, quality over quantity. A good post once a month beats four mediocre ones. That principle isn’t idealism anymore. With the scissors closing, it’s the only economically defensible position left — because the mediocre middle is exactly what’s being eliminated.

The content scissors are closing. Supply goes vertical, demand goes horizontal, and the gap between them grows every month. The mass market for adequate human writing is already gone. What’s left is the market for the thing only you can say — shaped well enough that someone skeptical enough to matter will keep reading past the second paragraph.

The world has enough content.

It has a permanent shortage of specific, hard-won truth.