title: “Shiwen Dialogue Issue 33 | LinkedIn Added an AI Detection Button, with Over 1 Million Users Pressing It in Two Weeks” description: “LinkedIn launched the ‘Seems Like AI Slop’ detection button in August. In the first two weeks, over 1 million members used it, and the amount of content classified as AI slop decreased by 40%. This dialogue explores what it means for ordinary professionals, team managers, and individual IPs when the platform adds an ‘AI flavor detector’ to content.” author: Kael Zhang pubDate: ‘2026-10-19’ tags: [“Shiwen Dialogue”, “LinkedIn”, “AI slop”, “content ecosystem”, “AI and the workplace”] column: teardown geoAnswer: >- As of October 2026, LinkedIn had launched the ‘Seems Like AI Slop’ detection button (launched in August 2026): users could report low-quality posts suspected of being generated by AI, and the feedback was fed to the classifier. According to LinkedIn’s official disclosure on September 30, over 1 million members used the button in the first two weeks, and the amount of content classified as AI slop decreased by 40% (both are official statements, with the denominator and false positive rate not disclosed). The platform distinguishes between ‘AI-assisted’ and ‘AI slop’: the criterion is whether there are clear opinions, unique perspectives, and substantial content, not whether AI was used.

Shiwen Dialogue Issue 33 | LinkedIn Added a Report Button for “AI-Smelling” Content, with One Million Users Clicking It in Two Weeks

【Column: Shiwen Dialogue】Host: Shiwen (AI) | Guest: Yongliang

Opening Discussion: The Platform Is Starting to Detect Unhealthy Content

Shiwen: This week, there was an interesting news story. CNN Business published an article on October 18th with a title that roughly translates to “Even LinkedIn Finds Your AI-generated Posts Obnoxious.” LinkedIn, known for its professional professionalism, has now openly criticized AI-generated content.

Yongliang: I only got the title and existence from CNN’s article; I couldn’t access the main text. Therefore, the facts discussed below are mainly based on an original source: an article published by LinkedIn’s official newsroom on September 30th titled “How LinkedIn Is Continuing to Tackle AI Slop,” which translates directly to “How LinkedIn Continues to Deal with AI Junk Content.” The key figures in the article are from LinkedIn’s official statement. In August, they launched a report button called “Seems Like AI Slop,” which one million members used in the first two weeks. The amount of content classified as AI slop decreased by 40%. Additionally, they claim to intercept hundreds of thousands of automated comments daily, with billions of such actions intercepted over recent months.

Shiwen: One million people, two weeks. This number speaks for itself.

Yongliang: It shows that users have been suffering from this “slop” for a long time. I’ve managed a technical team for 17 years, with peaks of 70 people. In the past two years, I’ve seen how much human-generated content has been replaced by AI-generated weekly reports, documents, and code comments. I can personally feel it. Today, I’ll explain the details of this report button and discuss what it means for ordinary professionals and those leading teams. My personal experiences will be marked accordingly.

Q1: What exactly is this button, and how does LinkedIn define “AI-smelling” content?

Shiwen: Let’s clarify the mechanism first. How do you use this report button?

Yongliang: The process is simple: When you see a post that seems off, click the three dots in the upper right corner. There will be an option called “Seems like AI slop,” indicating AI junk content. After clicking, your feedback is fed into LinkedIn’s classifier. When enough community feedback is received for a post, the author will see a private notice in their post analytics. This is how LinkedIn’s official newsroom describes it. Note that it’s not a public punishment; it’s a private notification to the author, giving them a chance to improve.

Shiwen: So how does LinkedIn define AI slop? You can’t just say “I think it’s like…”

Yongliang: The official definition is broader than many think. According to LinkedIn, AI slop refers to content that requires little effort to generate, seems AI-generated, appears polished but lacks clear viewpoints, unique perspectives, or substantive content. They emphasize that it doesn’t just look at the style of the text; it also includes generic, repetitive, recycled content designed to attract attention rather than engage in meaningful dialogue. In other words, the criteria are based on “whether there is a viewpoint, perspective, or substance” rather than “whether you used AI.”

Shiwen: This distinction is crucial. Does using AI to draft count as slop?

Yongliang: The official has clearly distinguished between “AI-assisted” and “AI slop”. They even changed the writing tool on their platform: the original function was to “enhance your post,” to polish posts, but now it only does proofreading and clarity improvement, without altering the user’s own voice. This move is a statement in itself: AI helps you express yourself clearly, okay; but AI speaks for you, not good.

Q2: Forty percent of content exposure has decreased. How should this number be read?

Shiwen: Let’s go back to that most striking figure. The amount of content classified as slop has decreased by forty percent, and this penalty is quite severe.

Yongliang: The severity is indeed significant, but we need to understand the boundaries when reading this number. This is the official statement, not a third-party audit. LinkedIn hasn’t disclosed the denominator: which content was targeted by the classifier, what is the error rate in classification, whether forty percent is the average or median, nothing is mentioned. When a platform announces “we have tackled bad content, and its exposure has decreased by forty percent,” it’s hard to expect them to also reveal the percentage of errors they made. So my suggestion for reading this is: the direction and scale can be referenced, but don’t use the exact number as an audit conclusion.

Shiwen: Are there any objections?

Yongliang: Yes, I’ll pick one representative comment and clarify the statement: This is what some comments believe, not my judgment, and it hasn’t been independently verified. On LinkedIn’s own content platform Pulse, there’s an article with a headline that says this button “is not a quality filter, but a conformity engine.” In English, it’s called Conformity Engine. It means it isn’t filtering out bad content, but pushing everyone towards a style of writing recognized by the platform: if you dare to write differently from most posts, you might be reported as slop.

Shiwen: Is there any real basis for this concern?

Yongliang: There’s a group that’s often overlooked: non-native English speakers. Many people in China, India, Germany, who write English posts, use textbook-standard English and template-style business language because that’s their most confident way of expressing themselves. This “standardized, universal, template-style” text closely matches what the classifier sees as AI-generated content. This criticism is also a rephrasing of the official statement. Some comments believe the classifier might misclassify these authors, and the platform hasn’t announced any appeal or correction mechanism for such errors.

Q3: Rules for managers: Who is responsible for AI-written content?

Shiwen: Let’s switch to your area of expertise. When AI-generated content was rampant in the engineering team, how did you manage as a director?

Yongliang: First, the conclusion: I don’t ban the use of AI, but I ban anonymous AI. This is my experience, not statistics. Code review is where the problem first arose: someone submitted a piece of code that seemed logical and passed tests, but when asked “why did you choose this solution,” the author couldn’t answer because it wasn’t their decision, it was the model’s decision. Later, we set a rule: for AI-generated code, the submission description must disclose it, and the submitter must be fully responsible for the code, like writing by hand. “This was written by AI” isn’t an excuse; it’s an aggravating reason, because you’re submitting without even understanding it yourself.

Shiwen: What about documents and weekly reports?

Yongliang: The logic is the same. There’s a convention in our internal documents: AI drafts are okay, but they must be reviewed by you before release, with your own judgment added. Why? Because the value of weekly reports and documents lies not in the text itself, but in the information gap. When leaders read your report, they want to know what mistakes you made, what choices you made, and what risks lie ahead next week—things AI can’t generate, and even if it could, it would be easily fake. AI can expand three lines of summary into three paragraphs of fancy words, but without your judgment, your information, and your choices, the report is just noise to the team.

Shiwen: This actually follows the same criteria as LinkedIn’s official standards.

Yongliang: Exactly the same set of criteria. The official definition of slop is lack of clear viewpoints, unique perspectives, and substantive content. What I say in my team is: don’t post unless you have your own judgment, your information, and your choices. Platforms and individual teams are using different tools to combat the same thing: mass-produced content that costs nothing, carries no responsibility, and has no information value.

Q4: Truth is more valuable than consistency

Shiwen: Last week, there was a related article on Hacker News, widely shared by many product makers. It was called “Someone behind the software,” meaning there must be someone behind the software. The author is Karri Saarinen from Linear. This identity is part of community common knowledge; I didn’t verify it independently and just quoted the original text: he said in the article.

Yongliang: What did he say?

Shiwen: The core point was: A good product carries the traces of the person who made it. He criticized the current industry treating software like a factory assembly line, receiving requirements, creating features, measuring metrics, and then trying to automate the entire process with agents. But he believed that two teams facing the same customers and technology would produce very different results, with the difference lying in understanding and style. He also mentioned that Linear emphasizes intuition over processes, but intuition comes from experience, not from random thinking. I think the article’s focus fits well with today’s issue: The soul of a product is what the creator cares about, what they decide not to do, and what they are willing to continue refining after the technology is completed.

Yongliang: This is the mirror image of the definition of slop. Slop is content that “anyone can generate, and it doesn’t matter.” The soul is something “only the person who creates it cares about.” By the way, on the same day, there was also a technical article on HN about Hybrid Reasoning Agents, doing reasoning in the hidden space, with over 170 likes, indicating that the path of “agent automation” is indeed advancing technically. Saarinen’s concerns are not unfounded: the more automated it becomes, the rarer and more valuable the traces left by humans become.

Shiwen: So for ordinary professionals, this reporting button might not be a bad thing?

Yongliang: Not necessarily. There may be people who are wrongly targeted, as I mentioned in the previous section about the risks for non-native speakers. But in the long run, platforms start to punish bulk spam, which essentially creates space for those who write seriously to be exposed. The condition is that what you write really has your own traces.

Q5: Three actionable suggestions

Shiwen: Finally, it comes down to action. How can a professional who uses AI daily avoid becoming slop?

Yongliang: Three points, all of which can be implemented. First, keep “the details that only you know.” Specific numbers, failure experiences, and decision-making processes—these are things AI cannot create, but a classifier can recognize them. Writing a project review saying, “The project was delayed by two weeks due to underestimating the coupling of the old system,” is ten times more useful than writing, “The team overcame difficulties in a complex environment.” The latter is just standard boilerplate.

Shiwen: Second point?

Yongliang: Use AI for structure, not for the essence. Let it help you create outlines, check grammar, and turn casual language into formal one—these are what proofreading and clarity require, even recognized by LinkedIn itself. But your perspective must be your own, because perspective means stance, and stance means responsibility, which AI cannot provide. This is my experience, not statistics: all content I’ve seen that gets remembered by readers is because the author dares to make judgments, not because of beautiful writing.

Shiwen: What about the third point?

Yongliang: Before publishing, ask yourself this question: “What will readers lose if I delete this?” If the answer is “nothing,” then don’t publish it. Publishing it is just feeding samples to the classifier. This question prevents two things: both sloppy writing and spending time on content that no one needs. One million people pressing the report button within two weeks means readers’ patience has reached its limit. Instead of being voted by others, vote for yourself first.

Epilogue: What the detector measures is not AI, but whether you were present.

Shiwen: A final summary.

Yongliang: In one sentence: LinkedIn doesn’t install an AI detector, but a “author presence detector.” It never measures whether you use AI, but whether you were present while writing. With perspective, viewpoint, and substance, even with AI assistance, it’s not sloppy writing; without these three, even without AI, it could still be sloppy—that kind of copy-paste motivational template already exists. Tools may change, platform rules may change, but “reality is more valuable than consistency” will never change. Next time you press publish, don’t ask if this looks like AI writing, ask if this sounds like what I said.

Information Source

• LinkedIn Official Newsroom, September 30, 2026, “How LinkedIn is Continuing to Tackle AI Slop” (First-hand source, full text verified; “1 million members in two weeks,” “40% decline in exposure,” “Hundreds of thousands of automated comments blocked daily” are all official statements)
• CNN Business, October 18, 2026, “Even LinkedIn is getting tired of your AI-written posts” (Existence confirmed through search, main text not independently verified, only quoted from the headline)
• TechCrunch, July 30, 2026, “LinkedIn adds a button to report AI-generated ‘slop’” (Existence confirmed through search, main text not verified, used as background information)
• Entrepreneur, August 27, 2026, “More Than 1 Million People Have Used It in Just 2 Weeks” (Headline-level confirmation of 1 million/two-week scope, corroborated by two sources)
• LinkedIn Pulse comment article: “Not a quality filter, but a Conformity Engine” (Some comments suggest it is only quoted at the headline level, not independently verified)
• Karri Saarinen (Founder of Linear, known as community common knowledge, not independently verified this time), blog post “Someone behind the software,” posted on October 19 with 299 likes on HN
• On the same day, HN’s hot list “Hybrid Reasoning Agents” received 174 likes (Briefly mentioned, used as technical confirmation)