The Anti-AI-Slop Editing Checklist

Count how many times you use exactly three items in a list, a sentence, or a subheading sequence. If it happens more than twice in one piece, you have an AI pattern. Collapse two of those threes into two-item pairs or expand one to four.

Read your opening paragraph and your closing paragraph. If they echo each other almost word for word, AI wrote both. Rewrite the close to add something the intro did not promise.

Find the first paragraph. Does it explain what the article is about before saying anything useful? That is AI warming up. Delete it. Start on the first useful sentence.

Search for any sentence that previews what the reader is about to read. Delete it. Your subheadings already do that job. The preview sentence is filler.

If every section in your article has exactly three bullet points or exactly two paragraphs, that is structural uniformity that signals AI. Break the pattern. Let one section be longer because it earns it.

Use Find to search for the em-dash character (the long dash). Delete every one. Replace with a comma, a full stop, or rewrite the sentence. There is no legitimate use of an em-dash in AI-edited content because AI uses them as a verbal pause it never earned.

Search for the en-dash (the medium dash). Replace with 'to' for ranges (5 to 10 minutes) and with a comma or full stop everywhere else.

Find every set of parentheses. Ask whether the bracketed text is doing real work or hedging a claim. If it is hedging, either commit to the claim without the bracket or cut the claim entirely.

Count your bolded phrases. If more than one phrase per 200 words is bolded, AI over-emphasised for visual effect. Bold only the single most important phrase per section and nothing else.

Check whether your document is using curly (smart) quotes or straight quotes consistently. More importantly, check that any quotation in the text is a real quotation from a real source. AI fabricates quotes. If you cannot verify the original source, remove the quotation marks and paraphrase instead.

Search for this phrase and delete it. Whatever follows is either worth saying or it is not. The phrase adds nothing.

Same instruction. Delete the phrase. Start the sentence with the important thing.

Find every 'may,' 'might,' and 'could' in the piece. For each one, decide: is this genuine uncertainty (keep it) or hedging a claim you know to be true (delete and state the claim directly)? AI defaults to these words even when the claim is solid.

These phrases appear when AI cannot name a source. Replace with a specific person, a specific study, or your own direct observation. If you have neither, cut the claim.

Find phrases where two qualifiers sit next to each other: 'often tends to,' 'can sometimes be,' 'might potentially.' Cut one qualifier from every pair. Keep the stronger or more specific one.

These are vague softeners that do no work. Delete and rewrite with the actual condition: 'If you are doing X,' 'When your audience is Y.'

Search for 'delve.' It is the single most statistically overrepresented word in AI-generated text. Replace with 'look at,' 'examine,' 'cover,' or cut entirely.

Search for 'nuanced.' If the word appears, ask whether you have described the nuance, or just claimed one exists. If you have not shown it, cut the word and either show it or drop the point.

Replace 'comprehensive' with a specific number or scope. 'Comprehensive guide' tells a reader nothing. '14-step guide covering X, Y, and Z' tells them everything.

These three verbs are management-speak that AI uses as defaults. Replace 'foster' with 'build,' 'cultivate' with 'grow,' and 'leverage' with 'use.'

Search for both. They are used as inflated substitutes for 'market,' 'industry,' 'tools,' or 'environment.' Replace with the specific thing you mean.

If you have used either word, you need to either prove the claim in the next sentence with a concrete before-and-after, or replace with plain language. 'It changed how I approach client onboarding' is more credible than 'transformative.'

Delete every instance. Nothing integrates seamlessly. Describe what the actual experience is instead.

Find every paragraph starting with 'That said,' and rewrite the opening. AI uses this phrase to introduce a qualification, but it often signals a contradiction that could be resolved by choosing one position rather than hedging two.

Search for these transitions and delete the phrase. The section subheading already signals the move. These phrases are AI narrating its own structure.

If your subheadings or paragraphs follow the pattern First, Second, Third, Finally, you have an AI-generated list wearing a prose costume. Either use an actual numbered list or restructure to remove the scaffolding words.

Delete both. They signal a closing thought but add no information. Start the closing sentence with the actual closing thought.

Read the piece and highlight every claim. For each claim, ask: is there a specific example, a named tool, a real number, or a personal observation backing it up? If a claim has none of these, either add one or cut the claim. Vague claims are the most reliable signal of AI padding.

Read the whole piece and ask: does anything here reflect a specific point of view, a disagreement with conventional wisdom, or a detail that only someone who has done this work would know? If the answer is no, add at least one sentence that does. That sentence is the difference between AI-shaped content and content with a person in it.