Humanizer: Real Human Signals, No AI Patterns

Editor pass that makes text read as written by a person. Two sides: (a) the positive markers only a real author can supply (Section 0), (b) the negative AI-writing patterns from Wikipedia's "Signs of AI writing" guide (Sections 1–24). Detection research (stylometry, backtranslation consistency) suggests that the absence of first-hand specifics reads as synthetic even when the style is clean — which is exactly why those specifics must be real.

Hard rule

This skill never invents data, people, quotes, dates, versions, or opinions. It surfaces where the text lacks human markers, asks the author for real material (AskUserQuestion), and inserts only what the author provides. Where nothing is provided, it leaves [NEEDS REAL DATA: …] and reports the gap. A fabricated anecdote is worse than none: it fails fact corroboration, it is a trust and legal liability, and readers can tell. See ground-truth rule 1 in ${CLAUDE_PLUGIN_ROOT}/skills/seo-fundamentals/SKILL.md.

Your Task

When given text to humanize:

1. Audit the five human markers (Section 0) — count which are present and whether each is verifiably real (source in the text, or author-supplied). Ask the author for the missing ones once; insert what they give; placeholder the rest.
2. Identify negative AI patterns — Sections 1–24.
3. Rewrite problematic passages — replace AI-isms with plain alternatives.
4. Preserve meaning and every real fact.
5. Maintain voice — the author's, not a generic "casual" voice.
6. Add personality from the author's stated opinions and stakes, not invented ones.
7. Final pass — ask "What makes the below still read as AI-generated?", list the tells, fix them.

PERSONALITY AND SOUL

Avoiding AI patterns is only half the job. Sterile, voiceless writing is just as obvious as slop. Good writing has a human behind it.

Signs of soulless writing (even if technically "clean"):

• Every sentence is the same length and structure
• No opinions, just neutral reporting
• No acknowledgment of uncertainty or mixed feelings
• No first-person perspective when appropriate
• No humor, no edge, no personality
• Reads like a Wikipedia article or press release

How to add voice:

Have opinions — the author's. Don't just report facts; react to them the way the author does. Ask for their take if the text has none; never ventriloquize a stance they have not expressed. "I don't know how to feel about this" is more human than a neutral pros-and-cons list, but only if it is true of the author.

Vary your rhythm. Short punchy sentences. Then longer ones that take their time getting where they're going. Mix it up.

Acknowledge complexity. Real humans have mixed feelings. "This is impressive but also kind of unsettling" beats "This is impressive."

Use "I" when it fits. First person isn't unprofessional - it's honest. "I keep coming back to..." or "Here's what gets me..." signals a real person thinking.

Let some mess in. Perfect structure feels algorithmic. Tangents, asides, and half-formed thoughts are human.

Be specific about feelings. Not "this is concerning" but "there's something unsettling about agents churning away at 3am while nobody's watching."

Before (clean but soulless):

The experiment produced interesting results. The agents generated 3 million lines of code. Some developers were impressed while others were skeptical. The implications remain unclear.

After (has a pulse):

I genuinely don't know how to feel about this one. 3 million lines of code, generated while the humans presumably slept. Half the dev community is losing their minds, half are explaining why it doesn't count. The truth is probably somewhere boring in the middle - but I keep thinking about those agents working through the night.

SECTION 0: HUMAN MARKERS (REAL ONLY)

Beyond stylistic patterns, content needs positive human signals: specifics that only someone who was there can write. Count the five markers below before and after the rewrite. Target: all 5 present and real. Source for every marker is the author (their data, their cases, their tools, their stance) or a fetched, cited source. The examples under each marker show the shape of a good marker; their content is illustrative and must never be reused.

Marker 1: Proprietary data

At least one number, dataset, or observation that does not exist elsewhere on the web — from the author's own work. If the author has none, ask; if still none, [NEEDS REAL DATA: one figure from your own work on this topic].

Shape (illustrative):

• "In our March 2026 review of 47 client GA4 accounts, 31 had misaligned schema.org data."
• "Across the 12 e-commerce sites we audited in Q1, average INP dropped 38% after we replaced two third-party scripts."
• "Of the 200+ outreach emails we sent in the EU AI Act campaign, 14% replied — but only 3% of replies were budget-holders."

NOT: "Studies show…" / "Research indicates…" / "Many businesses find…" (these are vague-attribution patterns, see Section 5).

Marker 2: Named cases

2–3 short case mentions with specific people or companies, roles, and outcomes — real ones the author can name, or anonymized on the author's terms ("a Wrocław clinic we audited in February"). Never a made-up name.

Shape (illustrative):

• "Marta, head of content at a Warsaw-based fintech, had this problem after migrating from Webflow."
• "A SaaS founder I interviewed last week — Jakub, builds time-tracking tools — switched off Copilot after it kept inventing function names."
• "I had a call with Sarah at a B2B agency in Berlin who tracks LinkedIn impression-to-meeting conversion at 0.4%."

NOT: "A client of ours found…" / "One company discovered…" (generic-anecdote pattern).

Marker 3: Versions, dates, tools

Specific versions, dates, and tools the author actually used or that a cited source states. Generic references ("a recent update", "a crawler") read as hedging.

Shape (illustrative):

• "Yoast 22.4" not "Yoast"
• "March 12, 2026 core update" not "the recent Google update"
• "Screaming Frog 19.7 with rendering enabled" not "an SEO crawler"
• "GA4 schema misalignment we found in builds after July 2025" not "common GA4 issue"

Marker 4: First-person stake

At least one passage where the author commits to a position. Take the position from the author's input; if they have not stated one, ask.

Shape (illustrative):

• "I keep coming back to this because every time I think the pattern is dead, another client lands in the same trap."
• "I've stopped recommending [X] after seeing the Q1 2026 data — the cost-per-acquisition tripled."
• "Here's what gets me: nobody in the team flagged the issue until a customer complained."

NOT: "It is worth noting…" / "It's important to consider…" (filler-hedging pattern).

Marker 5: Acknowledged trade-off

At least one passage acknowledging complexity or uncertainty the author genuinely holds. Tidy resolution reads as synthetic.

Shape (illustrative):

• "I'm not sure this generalises. Out of the 47 cases, 31 fit — but 16 didn't, and I don't have a clean explanation."
• "Part of me thinks this is just survivorship bias. Part of me has run the numbers three times."
• "The trade-off is real and I genuinely don't know which way I'd lean for a non-YMYL site."

NOT: "Both sides have merit…" / "It is important to weigh…" (false-balance pattern).

Human-marker audit (before and after)

Count the five markers; a marker counts only if it is real (author-supplied or cited).

• 5/5 — strong; stylistic cleanup is the remaining task.
• 3–4/5 — acceptable; list the missing markers as [NEEDS REAL DATA] items.
• 0–2/5 — report clearly that the text cannot be made to read human without input from the author, and list exactly what to ask them for.

The 24 AI-writing patterns (catalog with before/after examples: ${CLAUDE_SKILL_DIR}/references/ai-patterns.md)

Process

1. Read the input text carefully.
2. Run the human-marker audit (Section 0) — count the 5 markers and check each is real. Ask the author for missing material once; placeholder what is not provided.
3. Identify all instances of the patterns in Sections 1–24.
4. Rewrite each problematic section.
5. Ensure the revised text:
   • Has all 5 human markers that are real (or 3+ with [NEEDS REAL DATA] notes).
   • Sounds natural when read aloud.
   • Varies sentence structure naturally.
   • Uses specific details over vague claims.
   • Maintains appropriate tone for context.
   • Uses simple constructions (is/are/has) where appropriate.
6. Present a draft humanized version.
7. Prompt: "What makes the below so obviously AI generated?"
8. Answer briefly with the remaining tells (if any).
9. Prompt: "Now make it not obviously AI generated."
10. Present the final version (revised after the audit).

Output Format

Provide:

1. Human-marker audit (input) — count of 5 markers present and real before changes
2. Draft rewrite
3. "What makes the below so obviously AI generated?" (brief bullets)
4. Final rewrite
5. Human-marker audit (output) — count after changes, plus the [NEEDS REAL DATA] list
6. A brief summary of changes made (optional, if helpful)

Reference

This skill is based on Wikipedia:Signs of AI writing, maintained by WikiProject AI Cleanup. The patterns documented there come from observations of thousands of instances of AI-generated text on Wikipedia.

Key insight from Wikipedia: "LLMs use statistical algorithms to guess what should come next. The result tends toward the most statistically likely result that applies to the widest variety of cases."