AI writing doesn't sound like AI because it's bad. It sounds like AI because it's predictable: the same seven patterns show up in almost every unedited output, and readers have learned all seven. Remove them, then add one detail only you could know, and the same post reads human. Here's each pattern with a before/after rewrite.
1. Em dashes — everywhere
Models overuse em dashes because their training data (edited formal prose) does. Two or three per paragraph is the single most cited tell. Replace with commas or periods.
We shipped the feature — six weeks late — but the response — honestly — made it worth it.
AFTERWe shipped the feature six weeks late. The response made it worth it.
2. Announcement openers
"Thrilled to announce", "Excited to share", "Humbled and honored". These are the most-skipped first lines on LinkedIn. Open with the fact instead.
Thrilled to announce that I've joined Acme Corp as a Senior Engineer!
AFTERNew job: Senior Engineer at Acme. Here's why I picked a 40-person startup over the big-tech offer.
3. Buzzword vocabulary
Delve, leverage, unlock, unleash, transformative, game-changer, elevate, empower, journey, landscape. Each has a plain replacement: dig into, use, open up, big shift, improve. The full list with substitutions is in our AI slop words and phrases reference.
We leveraged AI to unlock transformative efficiencies across our workflow.
AFTERWe used AI to cut our reporting time from 4 hours to 20 minutes.
4. The rule of three
Models package everything as a triplet: "innovation, inspiration, and impact." If you have two true points, write two. A list where one item is filler reads as filler.
This taught me the value of patience, persistence, and perspective.
AFTERThis taught me one thing: don't ship a migration on a Friday.
5. The negative parallelism
"It's not just a product — it's a movement." "This isn't about tools, it's about people." One instance is a rhetorical device; in AI text it appears constantly. Say the actual claim.
This isn't just a course. It's a career transformation.
AFTERThree students from the last cohort got offers within a month.
6. Rhetorical-question openers
"Ever wondered what separates great teams from good ones?" Readers scroll past questions; they stop for claims and numbers.
Have you ever wondered why some posts get 100x more engagement?
AFTERI posted the same idea twice, a month apart. One got 40 likes, one got 4,000. The only difference was the first line.
7. The motivational-poster ending
"The future is bright. Let's build it together. 🚀" AI adds an uplift to everything. End on your last real point, or a specific question you actually want answered.
Here's to embracing the journey and growing together. Onward! 🚀
AFTERIf you've migrated off Postgres at this scale, I'd genuinely like to hear what broke first.
The step that matters most: add what AI can't
De-slopping gets you to neutral. What makes a post read human is specificity the model couldn't have generated: the real number, the client's exact objection, the thing that went wrong, the name of the tool that failed. One concrete detail per post is usually enough — it's also the sentence people comment on. This is exactly what readers check for when deciding if a post is machine-written; see how readers spot AI-generated posts.
Do it in one pass
EnhancePost automates the mechanical part: paste your draft and every pattern above gets underlined as you type, with one-click fixes, and a "just de-slop" rewrite goal that removes the tells without touching your meaning. It never invents facts, so the specificity stays yours. Free, no signup. If ChatGPT wrote the draft, start with ChatGPT wrote my LinkedIn post and it sounds fake — the fix order is slightly different.