The Microsoft-owned network is asking its 1.3 billion members to flag machine-written posts, three weeks after a detection firm estimated that 41 percent of long posts on the platform are fully AI-generated.
LinkedIn has handed its members a way to say out loud what many of them have been muttering for two years. On Thursday, July 30, the company confirmed a new reporting option tucked inside the three-dot menu at the top of any post. It reads: "Seems like AI slop."
Click it and LinkedIn replies with a single line: "Thanks for letting us know. Your feedback helps improve the feed." The post is then hidden from the user who flagged it.
The button was first spotted by 404 Media. Reporter Joseph Cox wrote that he scrolled for about six seconds before finding a candidate, which he identified by the gratuitous spacing between single sentences, emoji, bulleted takeaways and the "X is not Y" sentence construction that has become a signature of chatbot prose.
What stands out is the vocabulary. The option does not say "this looks AI-generated" or "this may be automated." It uses "slop," a blunt piece of internet slang for low-effort machine output, now written into the interface of a platform owned by one of OpenAI's largest backers.
LinkedIn's product chief separates AI from slop
Hari Srinivasan, LinkedIn's chief product officer, confirmed the rollout in a post on the platform after the 404 Media report.
"AI slop is a top priority for all of us," he wrote. "People come to LinkedIn to connect with real people and share their real perspectives, ideas, and expertise."
Srinivasan framed the button as a data-collection exercise rather than a moderation tool. "We are ramping the ability for members to tell us if they believe a post or comment seems like AI slop," he said. "Slop is hard to define and the definition changes; this lets us tune our models and make better feeds."
He also drew a boundary that will matter to anyone who has ever run a draft through a chatbot before hitting publish. "AI and slop are not the same thing," Srinivasan wrote, acknowledging that plenty of members use the technology to sharpen thoughts they came up with themselves.
That distinction shapes the enforcement design. Rather than penalising suspected AI use outright, LinkedIn will test a private notification inside a poster's own analytics dashboard when their writing may have come off as inauthentic. The signal goes to the author, not the audience.
The company is deleting the tool it built for this
The more striking move sits alongside the button. LinkedIn is retiring "enhance your post," the generative feature that rewrote members' drafts inside the composer, and replacing it with a proofreader that corrects the text without altering the writer's voice.
The platform spent years lowering the cost of producing a post. It is now discovering that the cost was doing useful work.
Other changes announced by Srinivasan include a series of new and improved classifiers built to identify whether a post is AI slop or low-quality content, which will pull flagged material out of the recommendations LinkedIn serves from outside a member's network. The button feeds directly into that system, supplying human labels that the classifiers train on.
On the automation side, the company says it now blocks hundreds of thousands of automated comment attempts every day, and has blocked millions of other automation attempts over the past two months. LinkedIn is also widening access to profile and page verification, and adding a control that lets members block comments from company pages they would rather not hear from.
The number that forced the issue: 41 percent
The urgency traces back to a study published on July 9.
AI detection company Pangram analysed roughly one million posts that users of its Chrome extension organically scrolled past across LinkedIn, Medium, X, Reddit and Substack over a two-month window. Because the extension scans passively while someone browses, it measures the slop people actually encounter rather than the slop sitting unread on abandoned SEO farms.
On LinkedIn, 41 percent of long-form content the extension's users saw was likely fully AI-generated. For short-form posts, the figure was 30 percent. Pangram classified anything between 50 and 250 words as short-form and anything above 250 words as long-form.
The pattern held across every platform in the sample: longer posts were consistently more likely to be machine-written. On X, a quarter of articles were fully AI-written and a further 23 percent showed AI assistance, meaning drafted, edited or rewritten by a model with some human involvement. Roughly one in ten longer Reddit and Substack posts registered as AI.
"AI content is a tax on readers' time," Pangram CEO Max Spero told 404 Media.
Pangram wrote in its accompanying blog post that people are overwhelmingly willing to use AI to speak on their behalf in professional settings tied to their real identity, and less willing to do so on casual and anonymous platforms. The finding is uncomfortable for a network whose entire premise is that the name on the post belongs to a real career.
Spero also cautioned that the figures represent a floor rather than a ceiling. Anyone who installs an AI detector in their browser is probably already muting or blocking the worst offenders, which would push the observed rate down. He put Pangram's false positive rate at roughly one in 10,000.
LinkedIn had already tried the quiet approach
Thursday's button is the second attempt, not the first.
In May, LinkedIn published a policy update titled "Keeping conversations real on LinkedIn," written by Laura Lorenzetti, vice president and executive editor of LinkedIn Global Editorial. The company said it would reduce the reach of content that appears to be generated by AI and lacks clear perspective. It also unbundled the AI writing assistant from the post button, so the feature no longer sat in the path of every member composing an update.
That crackdown suppressed slop from recommendations without removing it. Two months later, Pangram's data suggested the volume had not meaningfully receded, and 404 Media's follow-up reporting put the number in front of a much larger audience.
Asked for comment in July, a LinkedIn spokesperson said: "Professionals come to LinkedIn to hear from real people and their unique insights and perspectives. We actively work to reduce low quality, automated or generic content, and while AI can be used to beat the blank page problem, our focus is on surfacing professional conversations that help people advance their careers."
Substack moved first, with a detector rather than a button
LinkedIn is arriving late to a shift already underway across online publishing.
On July 21, Substack launched an AI detection feature built on a Pangram integration. Readers can scan posts, Notes, replies and comments longer than 100 words and receive an estimate of how much was written by hand versus with AI assistance. The feature is live on the web and iOS, with Android support promised later. Writers can run the scan on drafts before publishing, add an optional "How I make this" statement describing their process, and report results they believe are wrong.
Substack CEO Chris Best announced the partnership in a post titled "Against Claudefishing," his coinage for the gap between what a reader assumes about authorship and what actually produced the text.
"When content made by no one takes over parts of the internet that are supposed to be human, it pollutes the commons and makes it hard to discover and hear human voices," Best wrote. "When readers have to wonder if what they're reading is real, it undermines trust in authorship and threatens the livelihood of writers, including those who use AI tools thoughtfully to produce work they believe in."
Best has been explicit about which platform he does not want Substack to resemble, naming LinkedIn directly when he announced the tool.
Pangram itself has become the infrastructure layer for this fight. On July 29 the New York company announced a $9 million round led by Menlo Ventures, with participation from Haystack, ScOp Venture Capital, Script Capital and Cadenza, bringing its total raised to close to $13 million. Alongside the funding it released Pangram 4, a text detection model it says exceeds 99 percent accuracy on AI-assisted and mixed human-AI writing while resisting "humanizer" programs designed to launder machine output, plus a research preview of an image detector benchmarked internally at 99.5 percent.
The web the button is trying to hold back
The pressure on feeds is arriving from a much larger shift in who is using the internet.
On June 3, Cloudflare CEO Matthew Prince reported that automated requests had overtaken human ones on the company's network for the first time. Cloudflare Radar put the split at 57.4 percent bot traffic against 42.6 percent human. Prince had previously forecast the crossover for late 2027.
"Welp, that happened faster than I predicted," he wrote on X.
Newer platforms have already been crushed by it. Digg, revived by founder Kevin Rose with Reddit co-founder Alexis Ohanian, opened its public beta on January 14 and shut down its app roughly two months later while laying off staff. CEO Justin Mezzell called it a "hard reset" and pointed to automated spam that overwhelmed the voting and moderation systems the product depended on.
Reddit, meanwhile, published a post in July stating that spam, bot activity and inauthentic content are top of mind for people who love the site, and has been running an ad campaign built around the line that its users are human.
What a crowd-sourced label can and cannot do
The mechanism has a known weakness. What members flag is not AI authorship but the appearance of it, and the two overlap imperfectly. Formulaic structure, hollow inspirational arcs, heavy formatting and non-native English phrasing all read as machine-written to a scrolling reader. Since LinkedIn is explicitly using these reports to tune classifiers, whatever bias sits inside the reporting behaviour gets absorbed into the model.
There is also the question of what happens after the click. LinkedIn hides the post from the person who flagged it and adds the signal to its training data. The company has not said whether a human reviews flagged content, whether repeat offenders face any distribution penalty, or how many flags a post needs before its reach changes.
And none of it touches the incentive. LinkedIn's distribution still rewards volume and consistency, which is precisely the pressure that makes automated drafting rational for anyone building a professional brand. A report button raises the cost of one bad post by a fraction. The payoff for posting five times a week is unchanged.
Srinivasan's own framing conceded the harder part of the problem before the feature even shipped. Slop is hard to define, he said, and the definition keeps moving. LinkedIn has now outsourced that definition to 1.3 billion people and will find out what they mean by it.
Comments
Join the discussion and share your perspective.