You have read AI content. You know what it feels like. The sentences are grammatically correct, the structure is logical, and somehow the whole thing feels like it was written by a committee that has never met a human.
That is the slop problem. AI models default to patterns. They mirror what good writing looks like in their training data, which means they produce the average of all professional writing, stripped of personality, specificity, and opinion.
This article gives you 47 concrete rules to break those patterns. They are organized into six categories. Apply them in any order. The more you use, the less your content will sound like a machine wrote it.
Why AI text sounds fake
Six failure modes that make readers distrust your content
AI models optimize for plausibility, not truth or personality. They produce text that looks like good writing because it matches the statistical shape of good writing. The problem is shape without substance.
Ask any AI to write an intro about project management and you will get some version of In today's fast-paced business environment every single time. The model has learned that this phrase appears at the start of professional articles. It has not learned that the phrase is dead on arrival with any reader who has seen it 400 times.
The AI slop failure map
AI slop
what readers detect
Rules 1–10: voice and opinion
Generic AI text has no opinion. Fix that first.
Opinion is the single biggest differentiator between human and AI writing. AI avoids taking positions because it is trained to be agreeable. That produces content that technically covers a topic without saying anything about it.
Compare these two sentences: Email marketing can be effective for B2B companies. That is AI. Now: Email marketing beats LinkedIn ads for mid-market SaaS, and the data is not close. That is a human with a view. The second sentence makes you want to read the next one. The first makes you want to close the tab.
Make a claim in the first sentence
Start with your actual position, not a warm-up observation about the industry.
Name the thing you disagree with
Pick one common belief in your space and say plainly why you think it is wrong.
Use 'I' or 'we' when you mean it
First person signals a real perspective. Use it when you are drawing on experience.
Cut the hedge before the claim
Remove 'it could be argued that,' 'some might say,' and 'in many cases' before direct statements.
Rules 5–10: the rest of the voice category
Rule 5: Write the sentence you would say out loud. Read your draft aloud. If you would never say a sentence in conversation, rewrite it until you would.
Rule 6: Replace 'this shows that' with the conclusion itself. Instead of 'This shows that email is effective,' write 'Email works.'
Rule 7: Use a number when you have one. 'Most companies' is weaker than '7 out of 10 companies in our study.' If you do not have a number, say so plainly.
Rule 8: End a section with your opinion, not a summary. The last sentence of any section should say what you think, not repeat what you just said.
Rule 9: Name your audience precisely. 'Marketers' is too broad. 'B2B content managers at companies under 200 people' is a real person you can write for.
Rule 10: Avoid the word 'important' entirely. If something matters, explain why. The word 'important' does the explaining for you and does it badly.
Prompt: add a point of view to any AI draft
Claude / GPT-4Here is a draft I generated. I need you to rewrite it so it takes a clear position.
Rules:
- The first sentence must make a specific claim, not an observation.
- Remove all hedging language ('may,' 'could,' 'in many cases,' 'it depends').
- Add one concrete example from a real context (you can use a hypothetical, but make it specific with a company size, industry, and outcome).
- End with a sentence that says what the reader should do or think, not a summary of what was covered.
- Do not use the words: important, crucial, key, valuable, or significant.
Draft to rewrite:
[PASTE DRAFT HERE]Rules 11–20: sentence structure and rhythm
AI writes in one gear. Human writing shifts constantly.
Paste any 500-word AI output into a readability tool and you will find that 80 percent of sentences fall within two words of the same length. Human writers vary sentence length without thinking about it.
Short sentences land hard. Longer ones carry nuance and build toward a point before they stop. The mix is what creates pace. Without it, every paragraph reads at the same speed, the same weight, the same volume. That is what makes AI text feel like a drone.
The two-sentence test
Read any two consecutive sentences in your draft. If they are within three words of each other in length and both start with a noun, rewrite one of them. That single habit catches 60 percent of rhythm problems before they reach a reader.
Rules 17–20: the mechanical fixes
Rule 17: Replace transition adverbs with actual connective logic. Instead of 'Additionally, email has a high ROI,' write 'Email also pays off fast: the average B2B campaign returns $36 for every dollar spent.' The data does the connecting.
Rule 18: Break any sentence with more than two commas into two sentences. If you need three commas to hold a sentence together, the sentence is doing too much. Split it.
Rule 19: Use a colon to set up a payoff, not a list. 'There is one thing that kills B2B content faster than anything else: no opinion.' That structure creates forward momentum. A colon followed by a bullet list kills it.
Rule 20: Read the last word of every paragraph. If the last word is a weak noun like 'approach,' 'process,' or 'strategy,' rewrite the sentence so it ends on something specific. Endings carry weight. Use them.
Prompt: fix rhythm in any AI draft
Claude / GPT-4Rewrite this draft to fix the sentence rhythm. Follow these rules exactly: - No two consecutive sentences can be within 3 words of each other in length. - At least 1 in every 5 sentences must be under 8 words. - Remove all sentences that only bridge paragraphs without adding new information. - Replace every 'However,' 'Furthermore,' and 'Additionally,' opener with a different sentence structure. - Do not change the meaning or the facts. Only change the structure and length. Draft: [PASTE DRAFT HERE]
Rules 21–30: specificity and detail
Vague writing is the default. Specific writing is the fix.
AI content is vague because vague content is hard to argue with. A model trained to avoid controversy produces sentences that are technically true and practically useless. 'Many companies struggle with content strategy' is true. It tells you nothing.
Specificity is the fastest way to signal that a human wrote something. Humans have experiences. Experiences have details. Details are what readers remember.
Replace 'many' with a number
If you wrote 'many companies,' find the actual number or say you do not have one.
Name the company, person, or product
Generic examples teach nothing. Specific examples stick.
Give the failure mode a name
If you are describing a problem, name it. Named problems are easier to recognize and fix.
Describe the before and after
Any claim about improvement needs a before state and an after state with numbers.
Rules 25–30: more specificity rules
Rule 25: Describe the exact step, not the category of step. 'Optimize your headline' is a category. 'Put the primary keyword in the first 6 words of your headline' is a step.
Rule 26: Use a time frame. 'Results improve over time' means nothing. 'Most teams see measurable lift in organic traffic within 90 days' means something.
Rule 27: Say what does not work. AI avoids negative claims. Humans know what fails. Naming failures is more credible than only naming successes.
Rule 28: Attribute quotes to real people with real titles. 'Experts say' is not a source. 'Rand Fishkin, founder of SparkToro, wrote in 2023 that...' is a source.
Rule 29: Use a unit of measurement. 'A long article' is vague. '2,400-word article' is specific. 'A fast page' is vague. 'A page that loads in under 1.8 seconds' is specific.
Rule 30: Replace 'etc.' with the actual next item. 'Etc.' signals that you ran out of examples. Either finish the list or cut it to the examples that matter.
Rules 31–38: word choice and vocabulary
AI inflates vocabulary to sound authoritative. Cut it down.
AI models have learned that professional writing uses certain words. Words like 'leverage,' 'utilize,' 'robust,' and 'holistic.' These words appear in the training data as markers of credibility. The model reproduces them because they pattern-match to good writing.
They do not make writing better. They make it longer and harder to read. The fix is almost always a shorter, plainer word.
Rules 35–38: the vocabulary rules
Rule 35: Read every adjective and ask if it earns its place. 'Comprehensive guide' is a filler adjective. Either the guide is comprehensive and the content proves it, or the word is a lie. Cut it.
Rule 36: Replace nominalizations with verbs. 'Make a decision' is a nominalization of 'decide.' 'Provide assistance' is a nominalization of 'help.' Find the verb and use it directly.
Rule 37: Cut filler intensifiers. 'Very,' 'really,' 'quite,' 'rather,' and 'extremely' add length without adding meaning. 'Very effective' is weaker than 'effective.' If you need an intensifier, find a stronger word instead.
Rule 38: Use the word your reader uses, not the word your industry prefers. If your reader says 'blog post,' do not write 'content asset.' If they say 'meeting,' do not write 'synchronous collaboration session.' Write in their language, not yours.
The Orwell test
George Orwell's rule: if you can cut a word, cut it. Apply this to every AI draft. Target a 15% word count reduction on the first pass. The content will be clearer and faster to read. Readers do not notice when writing is concise. They notice when it is not.
Rules 39–43: structure and formatting
AI formats for scanners. Format for readers who actually want to learn.
AI-generated articles have a formatting signature. Three bullet points under every heading. Subheadings every 200 words. Bold text on random phrases. Conclusions that start with 'In conclusion.'
These patterns exist because AI has learned that formatted content ranks well and gets positive feedback. The problem is that formatting for scanners produces content that no one actually reads. You want readers, not scanners.
Use bullet points only for actual lists
Bullets are for items that have no natural prose order. If the items connect logically, write them as sentences.
Put the subheading where the topic changes, not every 200 words
AI adds subheadings on a timer. Add them when the subject genuinely shifts.
Bold the claim, not the topic
AI bolds topic words. Bold the sentence that makes the strongest point instead.
Delete the conclusion section
Conclusions that summarize what you just said waste the reader's time. End on the last actionable point.
Rule 43: the intro trap
Rule 43: Cut the first paragraph of any AI draft. AI models front-load context because they have been trained to warm up the reader. The first paragraph of an AI draft is almost always a restatement of the title. Delete it. Start with the second paragraph. The article will be stronger 90% of the time.
Test this yourself. Take any AI-generated article. Read paragraph two first. If it works as an opener, the first paragraph was filler. It almost always is.
Rules 44–47: the editing pass
The rules that catch everything the other rules missed.
The first five categories are about writing. This one is about reading. You cannot edit AI output at the speed you wrote it. Slow down. Read it as a skeptical reader who has seen a lot of AI content and has no patience for it.
These four rules are a checklist for that final pass.
The 'so what' test
After every paragraph, ask: so what? If you cannot answer in one sentence, the paragraph is not pulling its weight. Cut it or rewrite it until the answer is obvious.
The 'who says this' test
Read every claim and ask who is making it. If the answer is 'no one in particular,' the claim is AI filler. Attach it to a source, a study, or your own direct experience.
The 'would I send this' test
Imagine sending the article to one specific person you respect in your field. Would you be comfortable with them reading it? If you feel any hesitation, find the sentence causing it and fix it.
The 'first and last sentence' test
Read only the first sentence and the last sentence of the article. They should together tell the reader what you think and what they should do. If they do not, rewrite both.
Final edit checklist: run this before you publish
Prompt: full anti-slop rewrite
Claude / GPT-4Rewrite this article draft to remove all AI slop patterns. Apply every rule below:
Voice:
- First sentence must be a specific claim, not an observation.
- Remove all hedging language.
- Add first-person framing ('we tested,' 'I found') where the content draws on experience.
- End every section with an opinion, not a summary.
Rhythm:
- No two consecutive sentences within 3 words of each other in length.
- At least 1 sentence per 5 must be under 8 words.
- No paragraph openers starting with 'However,' 'Furthermore,' or 'Additionally.'
Specificity:
- Replace every 'many companies' with a number or a named company.
- Every claim of improvement needs a before and after with numbers.
- Remove 'etc.' and finish every list.
Vocabulary:
- Replace: utilize with use, leverage with use, facilitate with help, demonstrate with show.
- Cut all adjectives that are not proven by the content around them.
- Remove: important, crucial, key, valuable, significant, robust, holistic.
Structure:
- Delete the first paragraph.
- Delete the conclusion section.
- Bold only the strongest claim in each section, not topic words.
Draft:
[PASTE DRAFT HERE]What these rules change in practice
15%
Average word count reduction
▲ After applying vocabulary rules
2.4x
Higher time on page
▲ For content with strong editorial voice vs. generic AI output
3 min
Time to apply the checklist
▲ On a 1,000-word draft
47
Rules total
▲ Across 6 categories
Do not apply all 47 rules at once on your first draft
Pick one category and apply it consistently for a week. Voice rules first, then rhythm, then specificity. Trying to fix everything at once produces paralysis, not better content. Build the habits one category at a time.
Download the 47-rule anti-slop checklist
A one-page PDF with all 47 rules organized by category. Print it and keep it next to your screen for every AI editing session.
