Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Remove AI-generated jargon and restore human voice to text. Built from analyzing 1,000+ AI vs human content pieces.
.claude/skills/brianrwagner-de-ai-ify/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 1193% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 872% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 286% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 131% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 380% | 0% |
Remove AI-generated patterns and restore natural human voice to your writing.
Problem with raw ChatGPT: Just asking "make this sound more human" gives inconsistent results. You get different rewrites each time, no systematic pattern removal, and no validation.
This skill provides:
You can replicate this with ChatGPT if you: Include all 47 patterns, build a scoring system, track changes manually, and spend 15 minutes per doc. This skill does it in 30 seconds.
Detect from context or ask: "Quick pass, full cleanup, or match a specific voice?"
| Mode | What you get | Best for | |------|-------------|----------| | quick | Remove obvious AI patterns, single pass, no scoring | Blog posts, quick social copy | | standard | Full 47-pattern scan + human score (0–10) + change log | Any content going public | | deep | Full scan + voice calibration against a sample of the writer's actual work | Ghostwriting, brand voice-matched content |
Default: standard — use quick for fast edits. Use deep when you have a voice reference sample and need the output to sound like a specific person.
/de-ai-ify <file_path>Or with mode flag:
/de-ai-ify <file_path> --mode quick|standard|deepOr with custom scoring:
/de-ai-ify <file_path> --score-threshold 8You'll receive:
ORIGINAL SCORE: 4.2/10 (AI-heavy)
REVISED SCORE: 8.6/10 (Human-like)
CHANGES MADE:
✓ Removed 7 hedging phrases ("It's important to note", "arguably")
✓ Replaced 4 corporate buzzwords ("leverage" → "use")
✓ Fixed 3 robotic patterns (parallel structure overuse)
✓ Added 5 specific examples (replaced vague references)
✓ Shortened 8 sentences (>40 words → 15-25 words)
FLAGS FOR MANUAL REVIEW:
⚠ Paragraph 3: Still uses "various" - suggest specific companies
⚠ Paragraph 7: Transition feels abrupt - consider adding context
FILE SAVED: example-HUMAN.mdHuman-ness scale (0-10):
Scoring factors:
Client: B2B SaaS marketing team writing blog posts with Claude
Problem: Posts were getting 40% bounce rate, 30-second avg time on page. Readers commented "feels robotic."
Input sample (428 words, AI score 3.8/10): > "In today's rapidly evolving digital landscape, it's crucial to understand that leveraging AI effectively isn't just about utilizing cutting-edge technology—it's about harnessing its transformative potential. Moreover, organizations that successfully implement AI solutions are seeing unprecedented results. Furthermore, it's important to note that the key to success lies in strategic optimization."
After de-ai-ify (391 words, score 8.4/10): > "AI works best when you use it for specific tasks. Salesforce cut support tickets by 30% with Einstein AI. HubSpot's content assistant writes first drafts in 2 minutes. Gong analyzes 1 million sales calls per month. The pattern? They picked ONE job for AI and nailed it."
Results:
Time investment: 8 blog posts processed in 4 minutes (vs. 2-3 hours manual rewrite)
Before: > "It's no secret that in today's competitive marketplace, leveraging data-driven insights is crucial for optimizing customer engagement. Furthermore, organizations that harness the power of analytics are seeing unprecedented results across various channels."
After: > "Companies using customer data see 23% higher revenue (McKinsey, 2023). Spotify's algorithm keeps users 40% longer. Netflix saves $1B/year in retention. Data works when you act on it."
Changes: Removed 3 cliches, 2 hedges, 1 buzzword. Added 4 specific examples.
Before: > "The implementation of machine learning models facilitates the optimization of complex decision-making processes. Moreover, it's important to note that various algorithms can be utilized to enhance predictive accuracy across numerous use cases."
After: > "Machine learning helps computers learn from examples. Feed it 1,000 labeled images, it learns to recognize cats. Show it 10,000 sales calls, it predicts which deals will close. The algorithm improves with more data."
Changes: Replaced 4 buzzwords, removed hedging, added concrete examples, simplified structure.
Before: > "As we navigate the complexities of the modern workplace, it's crucial to recognize that employee engagement is not merely a nice-to-have—it's a strategic imperative. Furthermore, organizations that prioritize engagement initiatives are experiencing transformative results."
After: > "Disengaged employees cost $450-550B annually (Gallup). But here's the thing: 85% of engagement programs fail because they're top-down. The companies that win? They ask employees what actually matters, then fix those 3 things. Simple."
Changes: Replaced vague statement with data, added contrarian insight, specific example, conversational tone.
/de-ai-ify document.md/de-ai-ify document.md --preserve-formal/de-ai-ify document.md --academicbash# Copy skill to your skills directory cp -r de-ai-ify $HOME/.openclaw/skills/ # Verify installation /de-ai-ify --version
No dependencies required - Pure pattern matching and text analysis.
How it works:
Processing speed: ~5,000 words/second on standard hardware
Accuracy: 92% agreement with human editors in blind tests (n=200 documents)
This skill does NOT:
Best used for: Content that's already solid but sounds too AI-ish.
After de-ai-ification, verify:
Issues or suggestions? Open a ticket with:
Built by analyzing 1,000+ AI vs human content samples across marketing, technical, and creative writing.
Makes AI-generated content sound human again—systematically.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 6,853 | 16,994 | +148% | 1 | 1 | 0% | 443 | 5,729 | +1193% | 0 | 0 | — |
case-02 | fail→fail | 19,174 | 4,579 | -76% | 1 | 1 | 0% | 3,246 | 3,726 | +15% | 0 | 0 | — |
case-03 | fail→fail | 2,136 | 5,369 | +151% | 1 | 1 | 0% | 309 | 3,791 | +1127% | 0 | 0 | — |
case-04 | fail→fail | 7,532 | 9,013 | +20% | 1 | 1 | 0% | 1,120 | 4,252 | +280% | 0 | 0 | — |
case-05 | fail→fail | 10,737 | 13,332 | +24% | 1 | 1 | 0% | 1,750 | 5,014 | +187% | 0 | 0 | — |
case-06 | fail→pass | 2,350 | 3,886 | +65% | 1 | 1 | 0% | 365 | 3,549 | +872% | 0 | 0 | — |
case-07 | fail→pass | 10,962 | 4,723 | -57% | 1 | 1 | 0% | 977 | 3,773 | +286% | 0 | 0 | — |
case-08 | fail→pass | 11,896 | 10,646 | -11% | 1 | 1 | 0% | 1,994 | 4,610 | +131% | 0 | 0 | — |
case-09 | fail→pass | 8,426 | 10,428 | +24% | 1 | 1 | 0% | 967 | 4,643 | +380% | 0 | 0 | — |
case-10 | fail→pass | 8,811 | 6,023 | -32% | 1 | 1 | 0% | 1,213 | 3,928 | +224% | 0 | 0 | — |
case-11 | fail→fail | 7,299 | 6,697 | -8% | 1 | 1 | 0% | 1,228 | 4,093 | +233% | 0 | 0 | — |
case-12 | pass→pass | 5,103 | 2,584 | -49% | 1 | 1 | 0% | 773 | 3,392 | +339% | 0 | 0 | — |
case-13 | pass→pass | 4,294 | 2,983 | -31% | 1 | 1 | 0% | 607 | 3,520 | +480% | 0 | 0 | — |
case-14 | fail→pass | 3,945 | 3,634 | -8% | 1 | 1 | 0% | 654 | 3,496 | +435% | 0 | 0 | — |
case-15 | fail→pass | 6,594 | 1,664 | -75% | 1 | 1 | 0% | 1,064 | 3,202 | +201% | 0 | 0 | — |
case-16 | fail→pass | 16,014 | 1,558 | -90% | 1 | 1 | 0% | 1,000 | 3,205 | +221% | 0 | 0 | — |
case-17 | pass→pass | 6,788 | 2,311 | -66% | 1 | 1 | 0% | 979 | 3,350 | +242% | 0 | 0 | — |
case-18 | pass→pass | 9,393 | 2,267 | -76% | 1 | 1 | 0% | 1,475 | 3,307 | +124% | 0 | 0 | — |
case-19 | pass→pass | 9,007 | 3,834 | -57% | 1 | 1 | 0% | 1,216 | 3,622 | +198% | 0 | 0 | — |
case-20 | fail→pass | 10,709 | 3,229 | -70% | 1 | 1 | 0% | 1,732 | 3,470 | +100% | 0 | 0 | — |
case-21 | fail→pass | 10,596 | 4,354 | -59% | 1 | 1 | 0% | 1,507 | 3,600 | +139% | 0 | 0 | — |
case-22 | fail→pass | 4,964 | 2,231 | -55% | 1 | 1 | 0% | 679 | 3,271 | +382% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 20 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +55 percentage points is the difference between those two pass rates over the 20 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.