28 September 2026 · 5 min read

For a long time the worst thing that could happen to a post was that nobody read it. That is no longer the worst thing.

Since 30 July 2026, every post on LinkedIn carries a small option in its menu. A reader can press it to say the post seems like AI slop. The report is private. You never learn who pressed it. You are told, quietly, in your own analytics, that somebody did.

What LinkedIn did, and what it stopped doing.

Hari Srinivasan, LinkedIn’s chief product officer, announced the button on 30 July 2026. In the same announcement the company retired its own writing feature, the one that offered to enhance your post, and replaced it with a proofreading tool that corrects the writing without changing the voice.

Read those two together. The platform that spent two years offering to improve your sentences has withdrawn the offer. What it kept fixes commas.

Three weeks later, in an update reported on 24 August 2026, Srinivasan said the button had been used more than a million times. That is a very large number of people saying they can tell.

The number the careful people should read twice.

In the same update, Srinivasan said that people sharing AI written content are seeing around 40 percent fewer views than they were before the feature launched.

Hold that figure loosely. It came from the platform, in a LinkedIn post rather than a published study, with no methodology attached. It is a direction, not a measurement.

He also said no single piece of feedback is used to adjust how widely a post travels. That sounds like a contradiction and is not. One report does nothing to you. A pattern of reports, feeding classifiers that then run on everything, does something to everyone who writes the same way.

We want members to get feedback from real humans on what sounds authentic, not just have an AI detector review it and get it wrong.

That is Srinivasan, quoted on 28 August 2026, on why the judgment went to readers rather than to a detector. The thing now scoring your writing is a person’s instinct, at speed, scaled to a million presses.

How much of the feed this actually describes.

Pangram, a detection company, scanned 1,002,627 posts across several large platforms between April and June 2026 and published the results on 9 July 2026. On LinkedIn, 40.5 percent of long form posts, anything over fifty words, came back flagged as fully AI generated. Reddit was 4.4 percent, X 10 percent.

A vendor measuring with its own detector deserves suspicion, but the gap between LinkedIn and every other platform is too wide to be a rounding problem.

Enforcement moved too. LinkedIn’s European Digital Services Act disclosure for the first half of 2026, reported on 1 September, records a 46 percent rise in detected inauthentic activity against the previous six months.

Why this matters more if you are senior.

An unknown account that gets flagged loses some reach. A known one loses something harder to rebuild.

The old risk was invisibility, and we have already looked at how badly volume repays that fear. The new risk is a reader who decides in one second that you did not write it. That decision costs nothing, arrives without a conversation, and attaches to your name rather than the post. It compounds, because the same reader is the one who will later meet a summary of you assembled from it.

LinkedIn research published in early September 2026, covering more than 18,000 professionals and 6,000 marketers, found 42 percent naming expert voices as the most effective way to reach younger buyers. Credible people are the scarce thing now. Fluent sentences are not.

Four checks before anything goes out.

The useful question is not whether you used a model. It is whether the finished piece contains anything a model could not have supplied. A draft that fails more than one of these is not ready, however well it reads.

1. One number that belongs to you.

Not an industry statistic. A figure from inside your own work, with the period attached. What the renewal rate did after you changed the onboarding. A model can produce a plausible number. It cannot produce yours.

2. One moment with a date on it.

A specific decision, when you made it, and what it cost. Not the lesson. The scene. Generic writing floats free of time, and readers notice that before they can explain it.

3. One position with a price.

Something a reasonable colleague would argue with. If nobody in your industry would dispute a single sentence, you have written a summary of the consensus, which is what these systems are trained on.

4. One sentence that would be false in someone else’s mouth.

Take the strongest line in the draft and put a competitor’s name on it. If it still works, delete it. You have written something true of the category, not of you.

What to do with the tool that is left.

LinkedIn kept proofreading and threw away rewriting. That is a reasonable instruction for anyone working with a model.

  • Use it after the thinking, not instead of it. Structure, fix, tighten, check.
  • Never let it supply the substance. The number, the date and the position come from you.
  • Read the draft aloud. Any sentence that could sit unchanged under another name should go.
  • Check your analytics for the private notice. If it appears, treat it as data, not an accusation.

None of this requires posting more. It requires that each thing you publish carry evidence of a particular person having thought about it.

LinkedElla writes and positions LinkedIn presence for senior executives, founders and investors who already have a full job and need what people find about them to carry the weight of the work they actually do.

Questions and answers.

Does using AI to write get my post penalised?

LinkedIn has not said that. What it said on 24 August 2026 is that people sharing AI written content are seeing around 40 percent fewer views than before the button launched, and that no single report changes distribution. The signal being acted on is reader reaction, not tool use.

How do I know if my posts have been reported?

LinkedIn began testing a private notice in the author’s own analytics when members flag a post as inauthentic or heavily AI written. It is visible only to you and does not name who reported it.

Is 40.5 percent of LinkedIn really written by AI?

That is Pangram’s figure, from a scan of 1,002,627 posts between April and June 2026, published 9 July 2026, counting only posts over fifty words. A detection vendor measuring with its own detector deserves care. The direction holds across the other evidence.

Should an executive stop using AI for LinkedIn entirely?

No. LinkedIn kept a proofreading tool and removed the rewriting one, which is the right division. Use it to correct and tighten. Do not use it to decide what you think.

What single change makes the most difference this week?

Add one number from your own work, with a date, to the next thing you publish.