20 Rules to Keep Your Writing From Sounding Like AI Slop
Table of Contents
Why You Need Rules, Not Just Instincts
Rules for Before You Write
Rules for Drafting With AI
Rules for Openings and Endings
Rules for Language and Voice
Rules for Formatting
Rules for the Final Pass
FAQ
Most AI slop isn't made by lazy people. It's made by busy people who meant to edit the draft and ran out of time. The fix isn't more willpower. It's a short list of rules you apply every time, so the draft can't reach "publish" without passing through them.
Here are 20 rules to keep your writing from sounding like AI slop, whether you draft with ChatGPT, Claude or nothing at all. Copy them into your style guide, share them with freelancers, or print them and stick them next to your monitor.
Why You Need Rules, Not Just Instincts
You can usually spot slop in someone else's writing within a sentence. Spotting it in your own is harder, because you already know what you meant. Rules close that gap. They also scale: a list you can hand to a team is worth more than a skill that lives in one editor's head.
Others have landed on the same idea. Wikipedia's volunteer editors maintain a long Signs of AI writing guide to help clean up AI-generated articles, and product writer Peter Yang published an open-source No AI Slop skill that strips more than 20 AI writing patterns from drafts. The rules below cover the same ground, plus the substance problems that pattern lists miss. (New to the term? Start with our complete guide to AI slop.)
Rules for Before You Write
Rule 1: Write your point in one sentence first
If you can't state what the piece argues in one sentence, a model can't either, and it'll fill the gap with generalities. Write the sentence before you open any tool.
Rule 2: Gather material a model doesn't have
Your data, customer quotes, screenshots, results, mistakes and opinions. This is the raw material that separates your piece from the thousand others on the same topic.
Rule 3: Name one reader
"Marketers" isn't a reader. "A content lead at a 50-person SaaS company who just lost half her organic traffic" is. Specific readers produce specific writing.
Rules for Drafting With AI
Rule 4: Prompt with your notes, not a topic
"Write a blog post about onboarding" produces slop. "Turn these notes from our three onboarding experiments into a draft, keep my numbers exactly" produces a starting point.
Rule 5: Write the key parts yourself
Let AI handle structure and connective tissue if you like. Write the opinion, the example and the conclusion in your own words. Those are the parts readers remember.
Rule 6: Never publish a first draft
Not yours, not the model's. Every draft gets at least one full editing pass by someone who knows the subject.
Rules for Openings and Endings
Rule 7: Cut the warm-up
Delete your first paragraph and see if anything's lost. With AI drafts, it almost never is. Open on the claim, the problem or the surprising fact.
Rule 8: End on something concrete
No recap, no "the future is bright," no "the time to act is now." End with a recommendation, a result or a question worth thinking about.
Rules for Language and Voice
Rule 9: Delete hollow emphasis
Words that make things sound important without saying why: "crucial," "pivotal," "plays a vital role," "stands as a testament." If something matters, show the consequence instead.
Rule 10: Stop correcting claims nobody made
"It's not just a tool; it's a mindset." "This isn't about speed. It's about trust." This corrective framing is one of the most overused AI patterns; Graphite's 2026 study of AI writing tells found some variants appear hundreds of times more often in AI text than in human writing. Say the positive claim directly.
Rule 11: Break the rule of three
Models group everything in threes. Real lists have two items, or five, or one that matters more than the rest. Use the number the content actually has.
Rule 12: Vary sentence length on purpose
Mix long sentences with short ones. Use a fragment now and then. Uniform rhythm is one of the first things both readers and detectors notice.
Rule 13: Write the way you'd say it
Use contractions. Pick the plain word ("use" not "utilize," "help" not "facilitate"). If you wouldn't say a sentence to a colleague, rewrite it.
Rule 14: Name names
Real tools, real companies, real people, real numbers, real dates. "A leading CRM" is slop. "HubSpot" is information.
Rule 15: Kill vague attributions
"Experts say" and "studies show" are red flags. Either link the study and name the expert, or make the claim yourself and own it.
Rule 16: Take a side
AI defaults to balance: here are the pros, here are the cons, it depends. Readers came for your judgment. Give it, and give your reason.
Rules for Formatting
Rule 17: Format for reading, not decoration
Bold a phrase only if a skimmer truly needs it. Skip emoji bullets and "Key Takeaways" boxes on short pieces. Formatting should help readers find things, not make thin content look busy.
Rule 18: Make every heading say something
"Introduction" and "Benefits" are labels. "Why our churn dropped after we killed the free trial" is a heading. A reader should be able to skim your headings and get the argument.
Rules for the Final Pass
Rule 19: Verify every fact and link
Models invent statistics, sources and quotes. Click every link. Search every number. If you can't find the original, cut it.
Rule 20: Read it aloud, then run a humanizer pass
Reading aloud catches rhythm problems and phrases no person would say. After that, run the text through a tool built to catch what you missed. StealthGPT's AI Humanizer rewrites AI-sounding passages so they read naturally without changing your meaning. For the full process on the language side, see our guide on how to humanize AI text and bypass every AI detector for free.
One bonus rule, if you only keep one: publish less. Every rule above gets easier when you're not trying to ship five pieces a day.
FAQ
What are the most important rules to prevent AI slop?
Decide your point before prompting, bring your own specific material, cut generic intros and conclusions, and verify every fact. Those four rules fix most slop before it's written.
Can I still use AI if I follow these rules?
Yes. The rules are about how you use AI, not whether you do. Using a model to structure your notes or speed up a draft is fine, as long as a knowledgeable person shapes the result.
What words make writing sound like AI slop?
Inflated emphasis words like "crucial" and "pivotal," filler phrases like "in today's digital landscape," and stock constructions like "it's not X, it's Y." The exact tells change as models update, so focus on patterns rather than a fixed word list.
Is there a tool that applies no AI slop writing rules automatically?
Pattern-removal skills and humanizers can handle much of the language side automatically. No tool can add your point of view or your data, so the substance rules still need a person.
How do I get a whole team to follow these rules?
Add them to your style guide, make the final-pass rules part of your review checklist, and set publishing targets based on quality rather than volume.
