AI Writing Cleanup: Removing AI-Sounding Phrasing SKILL.md

This is the exact file agents read, including the credit header and licence notice. Save it as ai-writing-cleanup/SKILL.md.

---
name: ai-writing-cleanup
description: "Audit and rewrite drafts that sound machine-made: common AI tells from stock openers and inflated vocabulary to not-X-but-Y contrasts, forced triads, dramatic closers, em dash overuse and chatbot residue, adding specifics and first-hand evidence, matching an author or brand voice from samples, fact-checking and respecting AI disclosure policies. Use whenever someone says a draft sounds like AI or ChatGPT, asks to humanise text, remove AI slop, make writing sound human or like them, clean up AI-assisted content or edit generic agency copy."
license: CC-BY-4.0
metadata:
  author: Afeez Adebayo
  author_url: https://afeezadebayo.com
  author_linkedin: https://www.linkedin.com/in/afeez-adebayo/
  author_x: https://x.com/za_icon
  source: https://afeezadebayo.com/skills-md/marketing/content-copy-pr/ai-writing-cleanup/
  category: marketing
  subcategory: content-copy-pr
  version: "1.0.0"
  created: "2026-10-10"
  updated: "2026-10-10"
---

# AI Writing Cleanup: Removing AI-Sounding Phrasing

AI writing tools make drafting fast, and they leave fingerprints. Readers increasingly recognise the patterns: stock openers, inflated vocabulary, "it's not just X, it's Y" constructions, lists of three adjectives, dramatic one-line closers, em dashes everywhere and paragraphs that say little in many words. Content that sounds machine-made loses trust, reads as generic and rarely earns links, shares or citations. Google has said it rewards helpful content however it is produced, and treats mass-produced low-value content as spam, so the goal is quality and originality, not hiding the tool. [Confirmed, Google]

This skill audits and rewrites drafts to remove AI-sounding patterns, restore a specific human voice, add first-hand detail and evidence, and match the brand or author style.

## Operating principles
1. **Fix substance first.** Generic phrasing usually hides generic thinking; add specifics, examples and opinions.
2. **Cut patterns, keep meaning.** Remove tells without losing what the piece needs to say.
3. **Match a real voice.** Use the brand guide or samples of the author's own writing.
4. **Add what only you know.** Data, experience, customer stories and named examples make content human.
5. **Be open about AI use where policy requires.** Cleanup isn't about deception; it's about quality.

## Common AI tells
| Pattern | Example | Fix |
|---|---|---|
| Stock openers | `In today's fast-paced world`, `In the ever-evolving landscape of` | Start with the point or a specific fact |
| Inflated vocabulary | `delve`, `tapestry`, `seamless`, `leverage`, `unlock`, `elevate`, `robust` | Plain words: look at, mix, smooth, use, open, improve, strong |
| Not X but Y contrasts | `It's not just a tool, it's a revolution` | State the claim directly with evidence |
| Forced triads | `fast, efficient and reliable` | Keep the one quality that matters, with proof |
| Dramatic closers | `And that changes everything.` | End with a useful takeaway or next step |
| Em dash overuse | Several per paragraph | Use commas, full stops or brackets |
| Hedging filler | `It's important to note that`, `It's worth mentioning` | Delete; state the point |
| Generic examples | "A company might..." | Named or specific examples |
| Summary repetition | "In conclusion, as we've seen..." | Cut or replace with next steps |
| Chatbot residue | `Certainly!`, `I hope this helps`, `As an AI` | Delete |
| Uniform rhythm | Every sentence the same length | Vary length; mix short and long |
| Overuse of headers and bullets | Bullets for everything | Prose where reasoning matters |

## When to use, and when not to
Use it on AI-assisted drafts, agency content that reads generic, existing pages that underperform because they sound templated, and before publishing executive or thought leadership content.

Don't use it as a way to disguise low-value content at scale, or to remove required disclosures of AI use.

## Inputs to ask for, in one message
1. The draft and its purpose.
2. Brand voice guide, or three to five samples of the author's own writing.
3. Facts, data, experiences or examples the author can add.
4. Any AI disclosure policy.
5. Audience and channel.

## Workflow

### Step 1. Diagnose
Mark AI tells by type; note where content is generic or unsupported.

### Step 2. Fix substance
Add specifics: numbers, named examples, first-hand experience, opinions, customer quotes. Remove paragraphs that say nothing.

### Step 3. Rewrite openings and endings
Start with the point or a concrete scene; end with a useful takeaway.

### Step 4. Replace inflated words
Use the plain alternatives; cut hedges and filler.

### Step 5. Fix structure tells
Break triads, remove not-X-but-Y patterns, reduce em dashes, convert unnecessary bullets to prose.

### Step 6. Match voice
Compare with author samples or the brand guide; adjust rhythm, vocabulary and humour.

### Step 7. Fact-check
AI drafts can include confident errors; verify every claim.

### Step 8. Final read aloud
Does it sound like a specific person who knows the subject?

## Before and after
```
Before: In today's fast-paced digital landscape, businesses must leverage cutting-edge tools to unlock seamless growth. It's not just about efficiency, it's about transformation.
After: Most finance teams we work with lose two days every month to manual reconciliation. Automating three checks gets most of that time back.
```

## Voice matching from samples
- Note average sentence length, favourite phrases, humour and how the author opens and closes.
- Note words they never use.
- Rewrite one paragraph and ask the author whether it sounds like them; adjust.
- Build a short voice profile for future drafts.

## Substance checklist
Before polishing words, check that the draft includes at least some of these:
- A number from your own data or a cited source.
- A named example, customer or case.
- A first-hand experience (what we tried, what happened).
- An opinion the author would defend.
- A detail a generic writer wouldn't know (a tool setting, a common mistake, a cost).
If none can be added, the piece probably shouldn't be published.

## Rhythm fixes
AI drafts often use sentences of similar length and paragraphs of similar shape. Mix a short sentence after a long one, let some paragraphs be one line and others four, and allow the occasional question. Read the result aloud to check it flows like speech.

## Team checklist for AI-assisted content
- Human editor assigned before drafting starts.
- Draft checked for tells and substance before line editing.
- Every fact verified.
- Voice checked against the guide or author samples.
- Disclosure applied where policy requires.
- Published pieces sampled monthly for quality.

## Prompting to reduce tells upstream
When AI tools are used for first drafts, give them the voice guide, banned phrases, real examples and data, and ask for plain language with specific claims. Better inputs reduce cleanup time, though a human edit is still required.

## Edge cases and judgement calls
| Situation | What to do |
|---|---|
| Draft has no real substance | Return it with questions; cleanup can't add expertise |
| Technical content | Keep precise terms; remove only filler and hype |
| Regulated industries | Don't change approved claims; flag wording issues |
| Very short copy (ads, subject lines) | Remove clichés; keep it specific |
| Policy requires AI disclosure | Keep disclosure; cleanup improves quality, not concealment |

## Handoffs to related skills
- Voice guide: `brand-voice-guide`
- Full edit: `editing-proofreading`
- Thought leadership: `thought-leadership`
- Workflow and AI policy: `editorial-operations`
- Content quality for search: the Content SEO `eeat-content-quality` skill

## Worked example
A SaaS blog had published dozens of AI-assisted posts that ranked poorly and drew no links. Review found the same stock openers, inflated vocabulary and generic examples across posts. The team rewrote the top 20 posts: each gained a customer example, a number from product data and an opinion from the head of product; openers went straight to the problem; hype words were replaced; bullets became short explanations where reasoning mattered. Engagement time rose, several posts earned links, and the team added an AI-tell check to its publishing checklist.

## Output format
```
# AI writing cleanup: <piece>, <date>
## Diagnosis (tells by type, generic sections)
## Substance added (facts, examples, opinions)
## Rewritten draft
## Voice notes
## Fact-check log
```

## Common mistakes
- Swapping words without adding substance.
- Replacing one set of clichés with another.
- Losing precision in technical content.
- Skipping the fact-check.
- Treating cleanup as a way to hide mass-produced content.

## Questions to ask before you rewrite
- What does the author know that a generic draft doesn't?
- Whose voice should this sound like?
- What is the policy on disclosing AI use?

## Self-check before you answer
- Were AI tells removed and substance added?
- Does it sound like a specific person or brand?
- Are all facts verified?
- Are disclosure policies respected?

## Further reading
- Google on using generative AI content: https://developers.google.com/search/docs/fundamentals/using-gen-ai-content
- Wikipedia, Signs of AI writing (editor guidance): https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing

---
Created by Afeez Adebayo, Marketing, SEO & Growth Expert
Website: https://afeezadebayo.com
LinkedIn: https://www.linkedin.com/in/afeez-adebayo/
X: https://x.com/za_icon
Source: https://afeezadebayo.com/skills-md/marketing/content-copy-pr/ai-writing-cleanup/

Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). You may copy, share and adapt this skill, including for commercial use, as long as you credit Afeez Adebayo and link to the source URL above. Keep this notice in any copy or adapted version.