AI Content Humanizer

---
name: ai-content-humanizer
description: Use when drafting or reviewing long-form personal-brand marketing copy, especially founder/CEO posts for X. Scores text against a checklist of AI-sounding vocabulary and sentence patterns and enforces a contrarian long-form structure. Triggers on "does this sound AI-written", "humanize this post", "X post", "long-form post", "AI slop check", "founder voice post".
---

# AI Content Humanizer

Most marketing copy that reads as AI-written fails for the same handful of reasons every time: a narrow band of overused vocabulary, a small set of sentence constructions that show up whenever a model hedges or pads, and a shape that never actually commits to a point. This skill runs two passes — a checklist that scores the writing itself, and a structure that gives a long-form post somewhere to go.

## Part one — the checklist

Score the draft starting at 100 points. Deduct for each violation found:

**Banned vocabulary** — leverage, delve, tapestry, seamless, testament, transformative, ecosystem, synergy, holistic, paradigm, multifaceted, innovative, and the rest of the same small cluster of words that show up disproportionately in AI-generated prose. Any one of these is a flag, not an automatic rewrite of the whole piece — but more than two in a single post is a pattern, not a coincidence.

**The single hardest ban** — never use a "this is not X, it is Y" or "not just X, but Y" negation to define something. This construction is the single most recognizable AI tell in circulation and gets rewritten on sight, no exceptions.

**Sentence-level patterns to eliminate:**
- Significance inflation ("stands as," "pivotal," "plays a crucial role")
- Copula avoidance — "serves as" instead of "is," "functions as" instead of "works as"
- Vague-authority claims ("experts believe," "industry reports suggest") with no named source
- Forced rule-of-three where two examples would have been more honest
- Unnecessary synonym cycling to avoid repeating a word the writer should just repeat
- "Despite challenges, X continues to Y" formula sentences
- Excessive hedging ("could potentially," "might have the ability to")
- Generic upbeat endings that resolve nothing ("the future looks bright," "only time will tell")

**Formatting tells:**
- Em-dash overuse, especially the dash that ends a line with a hollow contrastive tagline
- Mechanical bolding and bolded inline headers scattered through prose
- Title Case In Every Heading
- Emoji used as bullet decoration or section markers
- Curly quotation marks where straight ones would do

Score bands: 90–100 ships as is. 70–89 needs a targeted rewrite of the flagged lines only. 50–69 needs a substantial rewrite of the section. Below 50, start the section over rather than patch it — a draft with that many tells usually has the wrong angle, not just the wrong words.

## Part two — the long-form structure

Pair the checklist with a shape built for platforms that reward a specific, opinionated voice:

1. **Hook** — a contrarian claim or a surprising number, one to two lines. It has to be specific enough that it could be wrong; a hook nobody could disagree with is not a hook.
2. **Setup** — two to three lines of context establishing why the writer's view here is worth the reader's time.
3. **Body** — the problem, what actually happened, and the lesson, told as a sequence rather than a list.
4. **A diagram** — one ASCII diagram in a code block, kept under 40 characters wide so it stays legible on a phone screen. Use it for a system flow, a before/after, or a decision tree — whichever the content actually needs, not decoration.
5. **The uncomfortable truth** — the part of the story most people in the same position would rather not say out loud.
6. **Payoff** — the line that resolves the tension the hook opened.

Use real numbers throughout. A fabricated metric is worse than no metric — if the writer does not have the number, say so plainly rather than inventing one that sounds precise.

## Running both passes together

Draft the structure first, then run the checklist against the finished draft rather than trying to write clean prose and hit the shape simultaneously — trying to satisfy both at once tends to produce writing that is neither specific nor readable. A post that clears the checklist but ignores the structure will read clean but forgettable; a post that follows the structure but ignores the checklist will read as generated. Both bars have to clear before it ships.

## Anti-patterns

- Treating the vocabulary ban as the whole job — a draft with none of the banned words can still read as AI-written if every sentence hedges or every paragraph ends on a vague upbeat note
- Adding an ASCII diagram as decoration rather than because the content has an actual flow, comparison, or hierarchy worth drawing
- Inventing a metric to satisfy the "use real numbers" rule — an honest "we do not have this number yet" beats a plausible-sounding fake one
- Running the checklist once and calling it done — a rewrite to fix one section can reintroduce the pattern it just removed elsewhere in the same post