We read 968 LinkedIn posts to find out what makes people scroll past.

Eleven years of one creator's posts, each checked against the reactions and comments it actually got. Some of it confirmed what everyone says. Some of it didn't.

By the 9th House team · September 2026

It started with LinkedIn's slop button

In July, LinkedIn added a "Seems like AI slop" option to every post and said it would show generic, AI-sounding posts to fewer people outside the author's network. A lot of people who write on LinkedIn for a living felt that one. Their drafts get a quick pass through ChatGPT, and now that pass might be costing them the audience they're writing for.

So we wanted a simple thing: paste a post before you publish it and find out if it's going to get buried.

The first version we built was an AI detector. Then we found ten free ones that already existed, most of them launched in the same fortnight. That told us an AI score on its own wasn't going to help anybody.

You can't predict a viral post from the words

Before building anything else we read the research, and it's blunt. In large studies of social posts, the text alone explains almost none of the difference between a post that flies and one that doesn't. Who's posting, who follows them and when they post all matter far more. One creator built his own virality predictor and said it had been wrong every single time.

That one you can test. Hold the author steady, and their audience, following and habits stay roughly the same. What's left to compare is mostly the writing.

The test

We used a public set of 968 posts by one tech creator, spanning 2015 to 2026, with the reactions and comments each one got. For every post we asked our checker about 18 yes-or-no questions. Does it open with something specific? Does it ask for engagement? Does it take a position someone could argue with?

LinkedIn's reach has changed a lot over the years, so we compared every post only against others from the same year. Then we looked at which answers lined up with doing better or worse than that year's typical post.

The things that hurt were the ones you'd guess. A weak opening was the most consistent drag we found. Posts that opened on a truism, a greeting or "I'm excited to share" did worse on both reactions and comments. Hype words (game-changer, unlock, transformative) and asking for engagement ("Agree?", "comment YES") went the same way.

The thing that helped most was less obvious. Taking a stance people could argue with went with more reactions, and it had the strongest link to comments of anything we checked. Posts with real numbers in them did better too.

The part that surprised us

Two things we expected to hurt did the opposite. Posts that ended on a tidy moral, and posts built around familiar advice ("your network is your net worth"), got more reactions, not fewer.

We think that's because a neat lesson is easy to like. You nod, you tap the thumb, you move on. It just doesn't start a conversation.

We could have left them in as warnings anyway. It would have made the tool sound smarter. But we'd be telling you something our own data says is false. So See More lists them under "reads like AI", as a style note, and never says they cost you reach.

It did. A weak opening dragged in every stretch, including 2025 and 2026, after LinkedIn's changes. Arguable posts did better in all three. Tidy morals kept picking up likes in all three, which is why we believe the surprise and didn't write it off as a fluke.

The trap we almost fell into

We also wanted to tell you what's working, so you don't edit the good parts out. We had a list of eight "strengths" ready to go. Then we ran the same questions on 45 posts that we had an AI write from lazy prompts, the "write me a viral LinkedIn post about leadership" kind.

A "strong opening" showed up in 91% of the AI posts. Of course it did: AI hooks are written to hook. The same went for "one clear point" and "has a story". If we'd shipped those, See More would have praised the exact posts LinkedIn is trying to bury.

So we cut them. Two strengths survived, because they almost never show up in slop. One is sounding like a person talking: half the real posts, 2% of the AI ones. The other is taking a stance people can argue with. We kept "real numbers" too, but hide it when a post reads like AI, since AI makes numbers up.

How good is the AI check?

We tested it on 201 posts written before ChatGPT existed and 60 written by AI.

We're fine with that last one. LinkedIn itself says using AI is okay as long as the post sounds like you. If you used AI and it still reads like a person with something to say, there's nothing for us to flag.

What it can't do

• It can't tell you a post will go viral. Nobody can, from the text alone. It tells you what's likely holding a post back.
• It learned from one creator's audience. A tech audience may reward different things than finance or fitness. We're adding more creators as people share their analytics.
• The effects are small. Each one nudges a post up or down a bit. None of them decides it alone.
• It's tuned on English. We haven't tested Hinglish or other languages yet.

Every time you tap "Fair? yes / no" on a result, you're telling us where we got it wrong. That's how it gets better.

We keep checked posts for 30 days to improve the checker, then delete the text. We count visits and clicks on this site to see what's useful, with no cookies and nothing that follows you elsewhere. There's no account, and we don't sell anything you paste.