LinkedIn's slop classifiers cut views by 40%. The number is not about you.

On August 25, Moneywise (republished on Yahoo Finance) walked through two numbers LinkedIn's chief product officer, Hari Srinivasan, posted in mid-August: more than a million unique members had clicked the Seems Like AI Slop feedback option since it launched, and members were seeing roughly 40% fewer views on content LinkedIn classifies as slop. The two figures are often quoted together. Moneywise's reporting makes clear they should not be.

LinkedIn's corporate communications lead, Amanda Purvis, told Moneywise the 40% figure is specific to the content classifiers the company rolled out the same day it launched the button on July 30 — not to the member-flag button. The drop reflects LinkedIn's own low-quality detection, and Srinivasan did not break out how much of the decline came from reader flags. So when you see "40% fewer views," the cause is a LinkedIn-side model, and it has been live for roughly a month.

Why the classifier matters to anyone who posts for reach

If you post on LinkedIn to reach clients, businesses, or recruiters, a 40% reach haircut on a demographic slice of your audience is a large quiet change. The classifier is not a report from a single reader. It is a product decision about which posts LinkedIn's ranking system should surface, and it happened without a note to most authors. Srinivasan's post also showed a screenshot of a message an author can now receive: Some members told us this post seems like AI. Purvis said that notice is rolling out to members over the coming weeks.

That message is the point of maximum confusion for writers. Getting it does not prove a machine wrote your post, and not getting it does not protect you from the classifier. Both the flag button and the classifier look for the same signal — prose that reads generic, repetitive, and mass-produced — and that is a property of the finished text, not of the tool that produced a draft. A person can write bland LinkedIn content all by hand. An AI edit can sharpen a post a human already wrote.

Srinivasan separately noted that AI and slop are not the same thing, and that plenty of members refine their thoughts with AI. That is LinkedIn's stated position. The classifier does not know your intention. It sees patterns. The practical question for a Markdown writer is narrow: if you use AI anywhere in drafting and then convert the file to rich text for LinkedIn, how do you keep the finished post from carrying the patterns the classifier is trained to suppress?

Three concrete liabilities the 40% figure exposes

The reach figure makes three older risks feel immediate rather than theoretical.

1. The universal opener

A draft that opens with a broad statement about the changing world or the evolving workplace triggers the classifier's least ambiguous pattern. It is the same shape as the mass-generated comment spam LinkedIn says it blocks by the hundreds of thousands every day. The classifier cannot tell whether the writer meant it.

2. Numbers with no source

A post that writes "engagement is up 40%" with no path back to a real measurement reads as decoration. This matters more now because the most-cited LinkedIn stat of the week is itself a percentage — and the whole story turns on which percentage actually means what. If you quote the 40% figures, name the source and the date.

3. The edited-in-voice gap for non-native and formal writers

Moneywise asked LinkedIn what protects members who write in a second language or a formal register — exactly the writing AI detectors are known to misread. Purvis pointed back to the many-signals and anti-targeting safeguards. That is an acknowledgment that the classifiers can misread sincere formal prose. For those writers the pre-publish observation pass matters most, because concrete detail is the signal no detector pattern can fake.

What LinkedIn has not explained

The reporting is honest about the open questions. LinkedIn has not clarified whether the volume of AI-generated posts is falling — only that views on classified posts dropped. It has not defined where the line between slop and ordinary content sits in its systems, and it has not said how much feedback is enough before the author-facing message appears. Srinivasan said at launch that slop is hard to define and that the definition changes. Writers should expect the boundary to move.

A Markdown pre-publish pass before you convert

Because the classifier reads the finished rich-text post, the fix belongs in the source file before conversion. Keep the draft in Markdown, run four checks, and only then turn it into formatted text. If you use LinkedIn's composer, run the checks first; formatting early hides what the prose actually says.

1. Open with the dated event. Lead with the specific thing that happened — a launch, a number, a mistake. Delete any first paragraph that could be pasted into any industry.
2. Source every percentage. If you cite the 40% figure or the million-user count, write "(Srinivasan, LinkedIn, Aug 2026)" or link the post. Numbers without a source become decoration, and decoration is the classifier's home turf.
3. Add one artifact only you could have produced. Name the tool, the failed attempt, the customer comment, the internal metric. This is the single strongest defense against a pattern classifier.
4. Cut text you would not say to a colleague. Remove stacked adjectives, ceremonial transitions, and machine-shaped sentences. Read the draft aloud once.

Example 1: the classifier-flagged opener, rewritten

This is the shape the classifiers are trained to suppress — a hand-built but generic draft that only quotes the week's statistic:

## What the AI slop button means for your content

In today's fast-moving social landscape, authenticity is more important
than ever. LinkedIn reports 40% fewer views on AI-generated content,
so brands must prioritize genuine human connection to stay ahead of
the algorithm.

The rewrite names the source, the date, and what changed in the writer's own process:

## The 40% figure I keep seeing is not about your post

Hari Srinivasan (LinkedIn CPO, Aug 2026) says posts LinkedIn's own
classifiers tag as slop get roughly 40% fewer views. That drop is
from LinkedIn's detection model, not from reader flags. Moneywise
confirmed the distinction with LinkedIn's comms team.

We still run an AI outline on client posts. We stopped shipping the
AI's first draft after a test post from ChatGPT lost 70% of the reach
of a sister post I wrote from a phone screenshot and two real numbers.

The second version cites the exact source, adds a measured counter-example with a number, and names an internal process change. Whatever the classifier scores, there is now real evidence for a reader to evaluate.

Example 2: protecting a formal or second-language post

A precise but formal post can read as machine-written to a classifier. The same words become defensible once the writer anchors them to a verifiable artifact:

### The importance of transparency

Transparency is essential for building trust with stakeholders.
Organizations should communicate their decisions clearly and
consistently, ensuring that all parties remain informed and aligned.

Keep the register, add the artifact, and the meaning changes:

### What our Q2 vendor review actually published

We share the raw decision log with the two vendors we renewed and the
one we cut. The log names the review date, the SLA breach, and the
owner who called it. Renewing vendors tell us this is the most
transparent process they have seen from a mid-size buyer; the cut
vendor did not contest the record.

A classifier may still weigh the formal register. It has no pattern to lean on now, because the post carries a date, a process, a verbatim vendor reaction, and a specific outcome. That is the kind of content no detector confidently calls slop.

Keep the edit in Markdown, convert at the end

Both examples share a workflow. The source is a local Markdown file. The editing passes run on that file. Only the finished draft gets converted to rich text — and if you use md2rich, the conversion happens in your browser. md2rich does not upload the draft, keep a server-side copy, or send the text to a rewriting API. For a post that cites internal metrics or an unannounced product, that privacy boundary matters as much as the formatting.

The reach figure changes the arithmetic of LinkedIn posting. It does not change the writing task: pass the finished text past a skeptical reader, and make sure the argument survives without the author standing next to it. The classifier is just the newest, most automated skeptical reader on the feed.

FAQ