On July 30, 2026, LinkedIn made two moves on the same day. It added an option for any reader to flag a post as content that appears to be written by AI, and it retired its own feature that rewrote your text for you. In its place it left one that proofreads without changing your voice. That is the line separating using AI to write faster from using AI to write instead of you, and it is now on record.
Before we go further, the clarification that will save you the scare the industry is about to sell you: nobody is going to fine you, nobody is going to suspend your account, and using AI to write is not banned anywhere.
What changed is smaller and more uncomfortable. Somebody in your network read you, thought "no person wrote this," and now has a button to tell the platform instead of telling you.
1. What actually happened, both parts
The flagging option is the move everyone covered. More than a million people used it in its first two weeks, according to LinkedIn's chief product officer, Hari Srinivasan.
The other move barely got covered, and it is the one that matters. The same day, LinkedIn retired "enhance your post," the feature that took your text and rewrote it. What it left behind is narrower on purpose: it fixes your grammar and spelling and leaves your voice alone.
That is not a product tweak. That is a company drawing a public line between AI as an editing aid and AI as a ghostwriter, and picking the first one.
Our take
When a company removes a feature it already built, one that already had happy users, that says more than any statement. They looked at the numbers in their own feed and decided the problem was not a shortage of content. It was a surplus of content with nobody inside it.
And along the way they settled an argument you have probably had with someone on your team this year.
2. What the button does, what it does not, and where to look
Precision matters here, because this week the story is going to get stretched in both directions.
What it does not do. It is not a reporting path for policy violations, and the platform was explicit about that. A single report does not decide anyone's distribution. When someone taps it, what changes most is what that person sees in their own feed.
What it does do, and almost nobody is covering this part. If enough people flag the same post, it does start affecting overall performance, and the platform tells you. The notice arrives privately, in that post's analytics. It is not public, it is not a penalty, and nobody in your network is going to mention it to you.
There is a third piece worth knowing. The same day it shipped the button, LinkedIn turned on automated classifiers that detect low-quality content and demote it in recommendations outside your network. So when the platform reports that content it classifies as low substance and low effort now gets 40% fewer views than it did a few weeks ago, that number mixes both mechanisms: reader signal and an automated system. It does not measure the button on its own.
Our take
Three caveats on that 40%, because it is being used badly.
It is a LinkedIn figure about LinkedIn, with no independent verification. It is a comparison against itself over time, not against good content. And it describes a whole category, not a penalty on a post of yours.
That said, the practical conclusion holds and it is the one we care about: the button did not invent a penalty. It invented a witness, and it gave that witness somewhere to leave a record.
What costs you money is the quiet case. If you run a dental practice, a law firm, or a clinic, reach was never your problem. Twelve people are: the colleague who sends you referrals, the patient who runs HR at a large company, the attorney you cross paths with once a year. When one of those twelve flags you alone, there is no notice because there is no volume. She stops seeing you, and you keep posting into a feed you are no longer in. It is the same underlying mechanic we work through in AI doesn't guess, it skips you: you do not get rejected, you get skipped, and skipping leaves no trace.
3. And it lands on the person who deserves it least
Now the part that does not sell courses.
The dentist, the attorney, the med spa owner did not start posting because they enjoy it. Somebody told them they had to show up, that if you do not post you do not exist. They have no time, they do not like writing, and they asked a tool for help. They did exactly what they were told to do.
The problem is not that they used AI. We use it every day and we say so by name.
The problem is that nobody explained the difference between using AI to write faster and using AI to write instead of you. The first one keeps your judgment in the text and only speeds up your hands. The second one takes your judgment out.
Our take
That distinction is the whole piece, so here it is in one line you are welcome to steal:
Our read, and we flag it as a read because there is no study behind it: your reader is not detecting artificial intelligence. Your reader is detecting absence. Nobody can spot a language model, but everybody can tell when there is no one home in a piece of writing.
And none of this calls for panic or for going back to writing longhand. It calls for a two-minute filter, once per piece, before you hit post.
4. The three tells
Three questions. Each one gets a yes or no, and each bad answer has a one-minute fix. None of them asks you to add statistics or go dig anything up.
Tell 1 - The first sentence announces the topic instead of saying something
"The importance of preventive dental care." "In today's world, digital presence is key." That is a topic, not a claim. A topic cannot be argued with, and what cannot be argued with does not get read.
The fix, one minute: delete the entire first sentence and read the second one. It usually says something. Most writing improves the moment you take off the on-ramp.
Tell 2 - There is not one sentence a colleague of yours could push back on
This is the hardest one to spot because it disguises itself as a virtue. Writing where nothing can be wrong feels professional, careful, safe. Which is exactly why it says nothing.
If a peer in your field read your whole post and could not find a single line to argue with, you did not write an opinion. You described a consensus, and the consensus has already been published a million times.
The fix, one minute: add one sentence that takes a position, not one that adds a fact. "Most cleanings that get canceled get canceled over price, and that math is almost always wrong." Somebody can argue with that. That is why it works.
Tell 3 - It could run unchanged next month, and next year
The fastest one. Ask yourself whether your post could be published as-is in March without changing a word.
If the answer is yes, nobody who was actually there wrote it. Writing with an owner is anchored to something: a conversation from this week, a question you got three times this month, something that changed in your field. Writing with no owner floats, which is why it could belong to anyone, on any date.
The fix, one minute: anchor the opening line to a moment. "Three patients asked me this last week." "Since the rule changed in July, this is the question that keeps coming in." It is not a statistic of yours, it is a date of yours, and it cannot be copied.
Our take
The three tells are not an AI detector and we do not want them used as one. They detect writing with no owner, which is a bigger and much older category: plenty of brochures written by hand in 2015 fail all three.
They are also the same three we run on our own work before it goes out with our name on it. If this post failed one of the three, we would not publish it.
5. What to do about what you already published
The short answer: nothing.
Do not go delete old posts and do not rewrite them in bulk. We are direct about what we cannot prove here: there is no study showing that rewriting an old post gives you reach back, or that deleting one clears a flag. Anyone telling you otherwise is selling a service.
The one backward-looking thing worth doing takes thirty seconds: open the analytics on your recent posts and check whether there is a notice. You now know that notice exists and where it lives. If there is none, you do not have a volume problem and you can leave the past alone.
The same eye pointed the other way, at text somebody else wrote about you, is in how to spot an AI-written review.
6. The platform just backed you up in writing
Back to that first move, because it is the best news in all of this and almost nobody covered it.
For two years the marketing industry sold you speed. More posts, more channels, more volume, with AI as the way to get there without hiring anyone. If you ever suspected that could not end well and felt old-fashioned for suspecting it, the largest professional content platform just agreed with you, and it paid for the position by removing a feature that already worked.
It pulled the one that rewrites. It kept the one that proofreads without changing your voice. Translated, that is exactly the distinction in section 3, said by the party holding the data for the entire feed.
Which is where the underlying work fits. Content does not compete on volume, it competes on being the answer somebody cannot find anywhere else, and that is the whole frame in the map of the five disciplines. Speed was never the advantage. It was just easier to sell.
The one thing to do this week
Run the three tells on the next post you were about to publish. Before you publish it.
One piece, two minutes. Not your last five, not your whole site, not an audit. The next one.
If you feel slightly nervous publishing it, you are doing this right. That is what it feels like when a piece of writing has an owner.
Frequently asked questions
To close
What ended is not using AI. What ended is nobody noticing, and the platform made it official by removing the feature that wrote for you.
The fix is not writing more, or writing longhand, or hiring anyone. It is two minutes per piece and three questions you can ask yourself today without anyone's permission. We know time is the thing you do not have when you are seeing clients all day, which is why the filter takes two minutes and not two hours.
For you, who made it all the way down here: thank you. Spending five minutes on how to write better is exactly the kind of decision that keeps your writing from sounding like a machine.
Keep reading
- The map of the five disciplines - the full frame for where content fits
- How to spot an AI-written review - the same eye, pointed at what others write about you
- Getting cited by ChatGPT doesn't pay the bills - what happens when they do see you and still nobody calls
- AI doesn't guess, it skips you
- This week's news recap
- The original announcement, July 30, 2026 - the full coverage
Sources
- LinkedIn adds an option to flag content that appears AI-generated, retires its "enhance your post" feature that rewrote text, and replaces it with one that proofreads without changing your voice - announced July 30, 2026 by Hari Srinivasan, chief product officer - includes the automated classifiers launched the same day and the private notice in post analytics when reports reach volume - https://techcrunch.com/2026/07/30/linkedin-adds-a-button-to-report-ai-generated-slop/
- More than a million uses in the first two weeks, that the option is not a policy-violation reporting path, that a single report mainly affects the reporter's own feed, that overall performance is only affected at volume, and that content classified as low substance now sees 40% fewer views than a few weeks earlier - figures the platform reported about itself, with no independent verification - https://www.socialmediatoday.com/news/linkedin-says-1m-people-have-reported-ai-slop/828465/
