LinkedIn's AI slop button is working. That is the part worth worrying about.

On 30 July 2026 LinkedIn shipped a button. It sits in the three-dot menu on any post, next to save and copy link, and it says "seems like AI slop". Three weeks later the company started publishing results, and they are large.

This is a content-moderation story that is also, quietly, an availability story. If a meaningful share of how people reach you runs through LinkedIn's feed, then a change to what that feed will carry is an outage you cannot page on — and there is no component on any status page that turns yellow when it happens.

Everything below is from LinkedIn's own statements as reported at the time, linked at the bottom. Where a number is the company's own, it is labelled as one.

What shipped

Four things went out together, and they are more coherent as a set than any one is alone:

The report button. Any member can flag any post as AI slop. LinkedIn's chief product officer described it as a signal that helps refine the models identifying low-quality content in the recommendation algorithm — so this is not primarily a takedown mechanism. It is a labelling pipeline. Members are annotating a training set.

The "enhance your post" feature was removed. LinkedIn's own AI rewriter — the one that took your draft and returned a polished version — is gone, replaced by a proofreading tool that corrects writing without altering the author's voice. The line being drawn is between AI that edits a human and AI that replaces one, and it is notable mostly because LinkedIn shipped the rewriter itself, not long ago, as a headline feature.

Private inauthenticity flags. LinkedIn will privately tell you in your own dashboard when readers seem to think your content reads as AI-written. Framed as coaching rather than punishment: improve the writing, rather than lose the reach without being told why.

Comment spam defences. Hundreds of thousands of automated comment attempts blocked daily, plus new classifiers aimed at slop in suggested content — the out-of-network recommendations, which is where a feed's quality problem is most visible and least controlled.

The results LinkedIn published

Reported on 21 August, roughly three weeks after launch:

• The button has been used more than one million times.
• Members are seeing, in the CPO's words, "40% less views on what we classify as AI slop from just a few weeks ago."

Both numbers are real and both are worth reading carefully.

A million reports in three weeks is a genuine signal about demand. Nobody had to be persuaded to use this. It is also, unavoidably, a million judgements made by people who cannot actually tell — there is no reliable way to identify AI-written prose by reading it, which is the finding that has survived every attempt to build a detector. What the button collects is not "this was written by AI". It is "this reads like the thing I have learned to dislike". Those correlate, and they are not the same, and the gap between them is where the false positives live.

The 40% figure has a subtler problem: the classifier defines the bucket and also measures the outcome. "40% less views on what we classify as AI slop" is a statement about a set whose boundaries LinkedIn draws. Widen the classifier and the number moves without anything improving. Narrow it and the same. It is a legitimate internal metric — it tells LinkedIn its intervention is reaching the thing it aimed at — and it cannot tell you whether the feed got better, whether good posts got caught, or whether anyone's experience changed. Only LinkedIn can see the denominator.

LinkedIn also says it has "built safeguards to help prevent individual feedback from being used to unfairly target other members", and that reach is determined by "many signals". Both are the right things to say. Neither is checkable from outside, which is the recurring condition of every platform metric anyone will ever quote at you.

Why this belongs on a status site

Here is the connection, and it is not a stretch.

Availability is usually framed as "does the request return". But if you run a business whose pipeline starts with people seeing your posts, the thing you actually depend on is distribution, and distribution can go to zero while every server is healthy. A 40% reduction in views for a classified bucket is, for anyone inside that bucket, indistinguishable from an outage: the thing that used to work stopped working, revenue moves, and nothing anywhere says why.

The difference is that an outage ends and gets a postmortem. This does not. There is no incident, no timestamp, no resolution note — and no status page anywhere that reports it, because from LinkedIn's side nothing is wrong. The system is working as designed. It is simply working differently than it did last month.

This is the same category of failure as a silent rate limit or a shadow ban, and it is the reason the honest answer to "is LinkedIn down?" is so often "no, and that is not your question". A green dot answers a narrow question about whether servers respond. It says nothing about whether the platform is still doing for you what it did in July.

And LinkedIn is a particularly poor place to look for that answer even when something is wrong. As covered in the piece on LinkedIn's incident record, its status page filed August's outage already resolved, backdated eighteen hours, at an impact level that leaves the indicator green. A provider that publishes infrastructure failures that reluctantly is not going to publish algorithmic ones at all.

What to actually do about it

Stop treating one platform's feed as infrastructure. The standard advice about single-provider dependency applies exactly as written here. If a classifier change can take 40% of your views, that is a single point of failure with no SLA, no status page and no support path. The mitigation is the boring one: an email list, or anywhere else the relationship is not intermediated by a ranking model.

Instrument your own reach. LinkedIn's numbers describe LinkedIn's classifier. Yours are the only ones that describe you. Impressions per post, tracked over time, is the monitoring you actually need — and it is the only way you will notice a step change in the week it happens rather than the quarter after.

Read the private flag as a real signal. If the dashboard starts telling you your content reads as inauthentic, that is the most direct feedback any platform gives about distribution, and it arrives before the reach does. It costs nothing to act on.

Recognise which failures have no page to check. The instinct when something stops working is to check status. That instinct is correct for servers and useless here. Knowing which of your dependencies can fail silently — and which of those will never generate an incident — is most of what operational awareness means outside of infrastructure.

You can watch LinkedIn's live status here, alongside the other social platforms this site monitors. Just be clear with yourself about which failures that page can see, and which it structurally cannot.

---

Sources:

• LinkedIn adds a button to report AI-generated 'slop' — TechCrunch, 30 July 2026
• LinkedIn says its AI slop button is working — Engadget, 21 August 2026, source of the one-million and 40% figures
• LinkedIn offers the option to report AI slop — Social Media Today
• LinkedIn status on isuptime