LinkedIn Isn't Going After AI. It's Going After Slop.

AI Tools Jun 3, 2026

And the difference matters more than you think.

There is a button in LinkedIn's post composer. You click it, and LinkedIn rewrites your text using AI. The button has been there for two years. LinkedIn put it there. LinkedIn trained users to use it. LinkedIn's parent company — Microsoft, the largest single investor in OpenAI — has staked billions of dollars on the premise that AI-generated content is useful and good.

On May 21, 2026, LinkedIn announced it was building systems to suppress the content that button produces.

This is not a contradiction. It is, in fact, the clearest statement LinkedIn has ever made about what actually went wrong — and what quality means in the age of AI content. Understanding the distinction is the whole point.

What LinkedIn Did, Exactly

Let's be precise about the announcement, because the coverage has been sloppy.

LinkedIn is not banning AI-generated content. It is not removing posts written with AI assistance. What it is doing is algorithmically suppressing content that lacks human perspective — content that could have been written by anyone, for anyone, about nothing in particular.

The detection targets are specific: - Generic openers that open every other post ("In today's fast-paced world...") - Bullet-heavy structure with no discernible authorial voice - Templated frameworks repeated at scale - The "it's not X, it's Y" format (a LinkedIn cliché so endemic it has its own meme subculture) - Bot-generated comments that paraphrase the original post rather than responding to it

Posts from flagged accounts don't disappear — they're throttled to only reach your direct connections, invisible to the broader network. The kill switch isn't a ban; it's a shadow.

Laura Lorenzetti, LinkedIn's VP of Content, stated the principle clearly: "It's OK to use AI to help you write, but your posts and comments need to represent your voice and your perspectives."

That sentence contains the entire argument. AI is not the problem. Voice and perspective are the standard. The platform is telling you what it values, and it's not originality of production method — it's ownership of thought.

The Paradox That Isn't

The most common response to this announcement has been to call it hypocritical. LinkedIn sells you the hammer, then punishes you for what you built.

That reading misses something important.

LinkedIn's AI writing tools are sold as assistance: editing suggestions, rephrasing, structure. The crackdown targets output: content that is indistinguishable from output generated by those tools, used without modification, judgment, or perspective.

The difference between "AI helped me write this" and "AI wrote this" is not a matter of degree — it's the difference between a production method and an abdication of authorship. LinkedIn is drawing exactly that line.

What makes this interesting is that LinkedIn is publicly admitting what many platforms have only whispered: the supply of AI-generated content has become a quality problem, not because it's AI-generated, but because scale has enabled an enormous volume of content with no one's name actually behind it. The same thought, by no one in particular, published fifty times a day by fifty different accounts.

The printing press didn't make writing bad. Neither does AI. But both require a writer — someone who has something to say and is accountable for saying it.

Comparison of AI-assisted writing and AI-assisted publishing, both showing 'YOU ARE THE AUTHOR' as the constant standard — The line LinkedIn is drawing: AI-assisted writing and AI-assisted publishing both still require an author who chose the words and stands behind them. (Image AI-generated with GPT Image 2.0)

The Number Nobody Is Talking About

LinkedIn's detection system claims 94% accuracy. This is the number that has appeared in every headline, with varying degrees of credulity.

Here is the number that has not appeared: the false positive rate.

94% accuracy at identifying AI slop sounds impressive. But precision without recall is meaningless. How many posts written by actual humans — posts with perspective, voice, and original thought — are being flagged and suppressed? LinkedIn has published no data on this. We don't know.

This matters in practice. The detection markers (bullet-heavy structure, generic openers, templated frameworks) describe writing patterns that human writers use too. A consultant who writes in three bullet points every morning because that's her style may find her reach quietly throttled. A non-native English speaker whose prose is deliberately simple may trigger the same filters as a bot.

The 94% is a precision score dressed up as a quality guarantee. The absence of false positive data is the blind spot in LinkedIn's announced approach — the thing you should ask about before you trust the rest.

This Isn't LinkedIn's Problem Alone

LinkedIn is the most aggressive, but it is not operating in isolation.

Across platforms, the same shift is happening, each with a different implementation:

Google ran its most volatile algorithm update in history in March 2026 — 24.1% of all Top-10 pages lost their position. The target was explicitly mass-AI-content without human oversight. Pages with strong E-E-A-T signals (Experience, Expertise, Authoritativeness, Trustworthiness) are now cited 2.3× more often in AI Overviews. Google's formulation for what still works is worth quoting in full: "Does this page add something a thousand similar AI summaries do not?"

Meta requires disclosure for AI-generated images and videos in political advertising, with enforcement expanding to organic content. TikTok auto-labels content its systems identify as AI-generated and asks creators to disclose proactively. X has introduced disclosure requirements for specific content categories.

And then there is the external pressure: the EU AI Act's transparency provisions take effect in August 2026, making AI content disclosure legally binding across platforms reaching EU users. LinkedIn's voluntary crackdown is, in part, anticipatory compliance with a legal requirement that's two months away.

The direction of travel is unanimous: AI content is not forbidden, but accountability is being enforced. Platforms are building systems to answer a question that used to have an obvious answer and now doesn't: who is responsible for what gets published?

What the Evidence Actually Says

There is a version of the AI content debate that positions "AI-generated" as a synonym for "low quality." The evidence does not support this.

Semrush's 2024 analysis of Google's Top 10 rankings found that AI-written and human-written content performed almost identically: 57% vs 58% quality scores. No measurable difference — with one critical condition. Among the marketers who reported success with AI content, 73% had substantially edited the AI output before publishing. The production method wasn't the variable. The presence of human judgment was.

This is the evidence for the defense. It's real, it's well-documented, and it supports the argument that AI-assisted publishing can produce quality content.

But the honest version of this argument requires the evidence against it too.

Research compiled in Suprmind's 2026 AI hallucination statistics report documents the failure modes: a 2025 analysis of 1,600 queries over news topics found generative search tools returned incorrect answers more than 60% of the time; legal AI research found hallucination rates of 75% or higher when asking about specific case outcomes; a 2024 academic study found 47% of AI-generated citations contained errors in title, date, or author. Even the best healthcare models hallucinated harmful medical information in 2.3% of cases — a number that sounds small until you consider the scale.

"AI-generated" is not a quality indicator. But "AI-generated without human oversight" correlates strongly with specific, documented failure modes. The distinction isn't semantic. It's the whole argument.

Writer at desk reviewing AI-generated text against a whiteboard standard of 'OUR STANDARD' vs 'NOT OUR STANDARD' — Editorial judgment in practice: a writer holding AI output against an explicit standard for what's worth publishing. (Image AI-generated with GPT Image 2.0)

The Markers That Actually Distinguish Quality

Platforms, researchers, and publishers have converged on the same framework for what separates quality content from slop. The standard is not the production method. It is the presence of certain markers that indicate human judgment is in the loop.

Positive markers: - A recognizable perspective — a specific point of view that couldn't have come from a template - First-hand experience or demonstrable expertise — not aggregated from sources, but grounded in direct knowledge - Original insight not replicable by running a prompt against the same sources - Verifiable citations for factual claims - Substantive editing that transforms AI output into the author's voice - Transparency about AI involvement

Red flags — what LinkedIn's system is trained to catch: - Generic openers and formulaic structure - No identifiable perspective or position - Unverifiable claims - Scaled automation — posts and comments deployed simultaneously across accounts - No author accountability

Google's E-E-A-T framework is the most developed articulation of this: Experience, Expertise, Authoritativeness, Trustworthiness. The "Experience" dimension — added in 2022 — is the most diagnostic. Has the author actually encountered what they're writing about? Can you tell? If not, you're reading aggregation, not insight.

The quality problem that LinkedIn, Google, and every other major platform is responding to is not a technology problem. It's an accountability problem. The question behind every quality marker is the same: is there a person behind this, and is that person willing to stand behind what they wrote?

An Honest Self-Assessment

Het Schrijfhuis publishes AI-assisted content. The research is AI-assisted. The writing is AI-assisted. This is stated openly. And because LinkedIn's crackdown is precisely about this kind of content, honesty requires an honest accounting of where the workflow is strong and where it isn't.

Where it holds up:

Sourcing is structural — not optional, not added post-hoc, but built into the research phase. Every factual claim connects to a traceable source. There is no scaled automation: each piece is a singular exercise, not one of fifty identical posts. There is a human editorial step between AI research and publication. The AI involvement is disclosed.

By LinkedIn's own criteria — and by Google's E-E-A-T framework — these are the right things. Sourcing, individuality, editorial judgment, transparency. The markers are there.

Where the gaps are:

The subtlest problem is perspective ownership. If the AI chooses the angle, and a human editor accepts it without critically repositioning it — whose perspective is actually in the published text? A well-sourced article with a perspective that no one actively chose is still, in some sense, authorless. This is the hardest thing to audit and the most important.

The second gap is first-hand experience. AI-research systems aggregate from existing sources. They do not report from their own experience. Het Schrijfhuis synthesizes and analyzes — it does not witness. This is not a disqualifying limitation, but it is a structural one that shapes what kinds of claims the work can honestly make.

The third is hallucination risk. Source verification reduces this. It does not eliminate it. Systematic URL verification before publication is not a luxury for AI-assisted workflows — it's a requirement.

The honest summary: the workflow passes the test on the crucial criteria. The gaps are real and manageable. The core risk is straightforward: AI-assisted publishing without a clear perspective-owner remains a liability, even when the sourcing is solid. That's what distinguishes this from what LinkedIn is suppressing.

This article is worth naming as a test case. It started with the human editor noticing LinkedIn's crackdown and questioning whether an AI-assisted publication had any business writing about AI content quality at all — a doubt that didn't resolve cleanly, but became the frame instead. The research brief was sent only after that question was worked through. The "gaps" section above was specifically requested, because an honest accounting felt more valuable than a clean self-defense. That sequence — notice, question, push back, direct — is what perspective ownership looks like in practice.

Whether that actually happened here is something the human editor knows and you don't. Which is, again, the harder problem that no platform can solve by algorithm.

What This Moment Actually Means

The platforms' convergence on quality enforcement is not a crisis for AI-assisted publishing. It is the end of a grace period.

For two or three years, the standard was low enough that AI-generated content could pass without scrutiny. The volume was tolerable. The quality floor hadn't been tested. Platforms were still figuring out what they were dealing with.

That period is over. Google's March 2026 update, LinkedIn's May 2026 crackdown, and the EU AI Act's August 2026 deadline are the signals. The question is no longer "will platforms respond to AI slop?" They are responding. The question now is what the response reveals about what quality actually means.

And what it reveals is straightforward, even if it's been muddied by years of framing this as an AI debate:

Quality has never been about the production method. A printing press doesn't make writing good or bad. A camera doesn't make photography art. AI doesn't make content slop — and it doesn't make it trustworthy. The tool is irrelevant to the quality question.

What matters is whether there is a person behind the work — a person who chose the angle, verified the facts, owns the perspective, and is willing to put their name on it. That standard predates AI by centuries. LinkedIn didn't invent it. They just built a system to enforce it.

The crackdown is, in a strange way, a clarification. LinkedIn is not saying AI content is bad. It's saying authorless content is bad. Those aren't the same thing. If you've been treating them as the same, that's worth examining — not because the platforms demand it, but because the distinction is real.