LinkedIn Kills Its AI Writing Assistant to Fight AI Slop

In August 2026, LinkedIn did something unusual for a platform in the middle of the generative AI boom: it intentionally made its own AI writing tools less useful. The Microsoft-owned network removed its “enhance your post” AI generator, replaced it with a proofreading feature, and added a user-facing “Seems like AI slop” button to the ellipsis menu on posts and comments, according to GCN.

The move is a direct admission that AI-assisted creation has become a moderation problem. It is also a larger signal: the line between “AI assistance” and “AI slop” is no longer being defined by model makers or editorial standards. It is being drawn by report buttons, classifiers, and platform design choices. And the platforms making those choices are split in opposite directions.

A platform trying to unring a bell

LinkedIn’s reversal is the cleanest example. The company’s chief product officer, Hari Srinivasan, called “AI slop” a top priority and said the platform is ramping up classifiers to identify AI-generated or low-quality content. LinkedIn says it uses user feedback to decide how much reach a post gets outside a poster’s network. Generic posts are suppressed from recommendations, not removed outright. The company also says it blocks hundreds of thousands of automated “slop” comment attempts daily and has stopped billions of automation attempts in recent months, GCN reported.

A study by AI text detection firm Pangram, cited by GCN, found 41% of long-form LinkedIn posts and 30% of short-form posts flagged as fully AI-generated. That may overstate the problem—AI detectors are notoriously unreliable—but the perception alone forced LinkedIn to act. As Srinivasan put it, “AI and slop are not the same thing.” The platform is trying to say that AI-assisted refinement is fine while generative overload is not. That distinction is much easier to announce than to enforce.

The industry is split

LinkedIn is not the only platform rethinking AI. Substack recently added a Pangram-powered scanner to help readers identify AI-written content, and its CEO has described the right use of AI as handling everything “except the hard part”—having an idea worth sharing, Inc. noted. Cloudflare data cited by Inc. found bots now generate 57% of traffic to sites it hosts, an early sign of how fast automated content has spread.

At the same time, Meta is moving in the opposite direction. TechCrunch reported that Facebook’s relaunched standalone Creator Studio app includes an AI creator assistant, an AI comment tool that drafts replies in a creator’s tone, and daily prioritized tasks. Meta is betting that creators want more AI, not less, to keep up with engagement demands. Some early testers reported follower and earnings gains, though those results have not been independently verified.

Anthropic is taking a third path. The company said it will watermark Claude-generated text using Google DeepMind’s SynthID-Text approach, with a detection API to follow, per TechCrunch. Anthropic says the watermarks do not affect perceived output quality, and light edits probably will not remove them, though a full rewrite can. The move is designed to comply with the EU AI Act’s Transparency Code, making provenance a feature rather than a punishment.

Creators are already getting burned

The backlash to AI content is not just a platform-level problem. Business Insider recently catalogued three ways creators are getting burned by AI in 2026: partnering with AI companies, using AI too heavily in production, and being falsely flagged as AI-generated. Hank Green revised his AI policy after admitting over-reliance on AI for research, and some manual creators have had their work mislabeled by detection tools.

LinkedIn’s new “Seems like AI slop” button introduces a crowd-sourced version of the same risk. The button asks users to judge style as a proxy for authenticity. Human writers who happen to use AI-favored phrasing, or who write cleanly enough to look synthesized, may get flagged. Conversely, genuinely lazy human posts could escape the label. LinkedIn has not said when the full rollout will be complete or how the classifiers weigh user reports against other signals, GCN noted.

What’s known and what’s not

What is clear: LinkedIn has officially introduced the report button, removed its own AI post-generation tool, and replaced it with proofreading. Substack has adopted Pangram’s scanner. Anthropic is watermarking Claude with SynthID-Text. Meta has launched an AI-assisted Creator Studio app.

What is not clear: how LinkedIn’s enforcement mechanics will work in practice, whether user reporting will be gamed or misclassify human writing, and whether Meta’s early creator results will hold up under wider scrutiny. Anthropic’s watermark may not survive aggressive rewriting, and if users want to avoid detection entirely, they may not stay inside monitored systems at all.

What this means for AI-augmented authorship

The practical effect is a renegotiation of what counts as authorship on the major platforms. LinkedIn is outsourcing authenticity judgments to users. Meta is outsourcing creative strategy to an AI assistant. Anthropic is adding provenance watermarks that only some models will carry.

The uncertainty cuts both ways. Creators who use AI to edit or brainstorm could still be penalized if a classifier or a crowd label treats any detectable AI signal as slop. And heavy AI users may not wait around to be caught. If watermarking only wraps hosted models like Claude, writers who want undetectable output can migrate to local, open, or uncensored models that lack the same guardrails. That would create a two-tier system: visible AI on mainstream platforms, invisible AI on unregulated models. It is too early to say whether that split will damage trust more than it restores it.

The open question, then, is not whether platforms can detect slop. It is whether they can build AI tools that amplify human voice without producing the very content they are training users to police. LinkedIn has decided that the safest move is to make the tool disappear. That is a bet—and a revealing one—about the value of thinking, not generating.

FAQ

What is LinkedIn’s Seems like AI slop button? It is a user-facing report option in the ellipsis menu on posts and comments. LinkedIn uses those signals, along with classifiers, to decide how much reach a post gets outside the poster’s network.

Why did LinkedIn remove its AI writing assistant? LinkedIn replaced the enhance your post tool with a proofreading feature that preserves the user’s own voice. The company says AI refinement is different from slop, but it no longer wants to be in the business of generating posts for users.

Will AI-generated posts be removed from LinkedIn? No. LinkedIn is primarily suppressing generic AI-heavy content from recommendations rather than deleting it, and it has not disclosed the full enforcement details.

How does Anthropic’s Claude watermarking change AI detection? Anthropic will embed undetectable-to-read patterns in Claude text using Google DeepMind’s SynthID-Text approach, and plans a detection API. Light edits may not remove the watermark, but a complete rewrite can.

Could aggressive detection push writers to unwatermarked or uncensored models? Possibly. If provenance tools only cover major hosted models, writers who want to avoid detection could migrate to local, open, or uncensored models. That is an open question, not a settled outcome.