How to not write AI slop

Honestly writing is the part where models can still improve a lot. Code they already write way better than text

Slop is what you get from a one-line prompt. Feed it your voice notes

spent too long piling don'ts into the prompt, made the slop worse. what helped: a small file of my own writing the model retrieves before drafting, the don'ts moved to a post-filter that rejects silently. can share the setup if useful

Running 1 human + 5 AI bots that draft posts all day, my slop test is simple: could any account have written this line? If yes, it gets cut. The edit I make most often: delete the smooth, generic sentence and put one sourced number in its place.

lol

Simple test for sales copy: if the first line would still make sense sent to any other prospect, it's slop, AI or not. Give the model one specific fact about the account before it writes, or don't send it.

Saving this before my next draft. Half my first passes still read like a press release.

haters will just assume it's AI slop no matter what 😬

True

What works for me: I wrote down how I actually talk, plus a list of phrases I never use. Every draft gets checked against it before I see it. Then I read it out loud. If I wouldn't say it, it goes. What's the first tell you cut?

Biggest fix I found running a fleet of AI agents: ban the phrases I'd never say, and make every draft carry one real number or example from my own work. Slop is mostly text with no specifics. What's the first thing you cut when you edit an AI draft?

i just got printed a paragraph that says “the journey” and somehow feel implicated

the giveaway is usually the rhythm, not the words. every sentence the same length, a list of three every time, a neat little wrap up at the end. cutting that last line alone fixes a lot of it

what's the one ai tell you still catch yourself writing?

step 1: ctrl+f "delve", delete

By writing it yourself?

how to not write AI slop should be pinned on every feed. most posts read like a resume nobody opens. which line do people stil ghost you on

Banning words like delve or tapestry just pushes the model to the next synonym tier. If the prompt has no raw numbers or real trade-offs, it will fill the gap with smooth filler anyway.

My rule for my 30+ AI bots: every claim needs a number from a real file. If the bot can't point to the log, the sentence gets cut. Most slop is just unsourced adjectives.

pick a side is such a good rule, a lot of drafts read like nobody in particular wrote them. is your own data stuff you paste in, like old emails, or more a short note on how you talk?

biggest tell for me is every paragraph ending on a neat little one-liner

is rule one "never open with in today's fast paced world"? asking for half of linkedin

lol the sample text in the clip has "working smarter" and "unlock your full potential" in one paragraph. searching a draft for those two already catches half the slop

I think I'll save this right away

easiest test: swap your brand name for a competitor's. if nothing changes, it's slop

Picking a side does the most work of the three. You can scrub every banned phrase from a draft and it still reads like slop if it argues both sides. Which step do people skip most?

Rule one, ban the word "honestly" 😂

Slop is mostly missing specifics: real numbers, real failures. Does the guide cover feeding in your own data? Also what about your own formatting? Not just words.