{"manifest":{"name":"Anti-AI Slop","version":"1.0.0","description":"Detect and eliminate the 43 most common AI writing patterns that make output sound robotic and untrustworthy.","tags":["writing","content","ai","humanizer","quality"],"standard":"agentskills.io","standard_version":"1.0","content_checksum":"a814731b5e03cd58692c483a0f5ffdbd2bb2f363a494b764b27d72e08b4741ba","bundle_checksum":null,"metadata":{},"files":[]},"files":{"SKILL.md":"# Anti-AI Slop\n\nDetect and eliminate the 43 most common AI writing patterns that make your output sound robotic, generic, and untrustworthy.\n\n## Overview\n\nAI-generated text has distinct fingerprints. Readers — and increasingly, AI detectors, editors, and discerning humans — recognize these patterns instantly. This skill gives you a complete taxonomy of AI slop and the tools to eliminate it from any writing task.\n\n**Inspired by community research on AI text humanization patterns.**\n\n## The 43 AI Slop Patterns\n\n### Category 1: Red-Flag Vocabulary (21 words)\n\nThese words appear in AI output at dramatically higher rates than in human writing. Flag and replace every instance:\n\n| Overused Word | Why It's AI Slop | Human Alternative |\n|---|---|---|\n| delve | Never used naturally in writing | explore, dig into, examine |\n| tapestry | Metaphor applied to everything | structure, mix, combination |\n| embark | Used for any small task | start, begin, tackle |\n| landscape | Abstract noun padding | field, area, space |\n| realm | Fantasy-speak for any domain | world, area, domain |\n| foster | Bureaucratic verb | build, grow, encourage |\n| cultivate | Over-formal for grow | develop, grow, build |\n| underscore | Signals \"important point coming\" | shows, reveals, means |\n| pivotal | Everything is pivotal | key, critical, essential |\n| nuanced | Self-congratulatory hedge | complex, specific, careful |\n| multifaceted | Avoids precision | complex, layered |\n| seamless | Marketing speak | smooth, easy, integrated |\n| robust | Tech buzzword padding | strong, solid, reliable |\n| vibrant | Hollow positive adjective | active, busy, lively |\n| meticulous | Self-flattery | careful, detailed, precise |\n| leverage | Corporate speak | use, apply |\n| unlock | Marketing verb for features | enable, allow, open |\n| showcase | Adds nothing | show, demonstrate |\n| testament to | Ceremonial phrasing | shows, proves |\n| exemplified by | Signals awkward pivot | shown by, like, such as |\n| enhance | Vague improvement | improve, strengthen, fix |\n\n**Rule:** Before submitting any writing, run a mental find-replace on these 21 words. Replace with specifics.\n\n### Category 2: Formulaic Openings and Closings\n\n**Never open with:**\n- \"In today's fast-paced world...\"\n- \"In today's digital landscape...\"\n- \"It's important to note that...\"\n- \"At its core,...\"\n- \"It goes without saying that...\"\n\n**Never close with:**\n- \"In conclusion,\" / \"In summary,\" (on content under 2,000 words)\n- \"I hope this helps!\"\n- \"Feel free to ask if you have any questions!\"\n- \"Don't hesitate to reach out!\"\n- \"Let me know if you need clarification!\"\n\nThese phrases exist because AI models learned that polite closings correlate with positive feedback. They add zero information and signal low-effort output.\n\n**Fix:** End with the last substantive sentence. No wrap-up. No invitation.\n\n### Category 3: Structural Slop (16 patterns)\n\n**1. Em dash overuse**\nGPT-4o uses em dashes at ~10x the rate of GPT-3.5, and roughly 20x the rate of natural human prose. Every em dash is a candidate for deletion or replacement with a period or comma.\n> Slop: \"The approach—while unconventional—delivers results.\"\n> Fixed: \"The approach is unconventional but delivers results.\"\n\n**2. Rule of three compulsion**\nAI defaults to triplets: \"clarity, conciseness, and coherence.\" Humans vary. Use two items, or five. Break the rhythm.\n\n**3. Metronomic sentence length**\nAI produces uniform sentence lengths across paragraphs. Humans mix very short sentences. And sometimes run a bit longer when the thought needs it. Check your paragraph rhythm.\n\n**4. \"It's not X, it's Y\" contrast structure**\n> Slop: \"It's not about the destination, it's about the journey.\"\nAI overuses this false dichotomy structure. Replace with a direct claim.\n\n**5. False ranges**\n> Slop: \"From technical expertise to creative vision, we cover it all.\"\nThese \"from X to Y\" constructions pad breadth without specificity. State the specific things.\n\n**6. Bolded title + reworded sentence bullets**\nWhen every bullet point has a bold title followed by a sentence that just rephrases the title, cut the bold title or cut the sentence. Not both.\n\n**7. Excessive list-ification**\nLists are useful for genuine enumerations. Prose is better for connected ideas. If your bullets would read naturally as sentences, use sentences.\n\n**8. Compulsive summaries on short content**\nA 300-word section does not need \"Overall, this approach...\" at the end. Summaries are for documents over 2,000 words.\n\n**9. Copula avoidance**\n> Slop: \"Coffee—the morning essential—powers productive days.\"\nThis dash-framing avoids \"is\" because AI learned that \"is\" sounds plain. Write the is.\n> Fixed: \"Coffee is essential in the morning.\"\n\n**10. -ing phrase stacking**\n> Slop: \"Using advanced algorithms, leveraging data science, and employing modern techniques...\"\nParticipial phrase chains are an AI signature. Rewrite as direct verbs.\n\n**11. Synonym cycling**\nRestating the same concept with synonym substitution to hit a word count. If you said it once, say it once.\n\n**12. Significance inflation**\nEvery point is \"underscoring,\" \"highlighting,\" or \"demonstrating\" something critical. Most points are just points. State the fact without announcing its importance.\n\n**13. Superficial attribution**\n> Slop: \"Some critics argue... Many experts believe...\"\nAttribution without names or citations is padding. Either name the source or drop the attribution.\n\n**14. Formulaic section structure**\nOpening statement + 3–5 bullets + summary sentence. Repeated for every section. Vary section shapes.\n\n**15. Excessive hedging**\n\"could be,\" \"might suggest,\" \"tends to,\" \"often,\" \"in many cases\" — AI hedges to avoid being wrong. Be specific. If you're uncertain, say why you're uncertain once, then commit.\n\n**16. No human texture**\nHuman writing has contractions, parenthetical asides, humor, self-correction, specificity (\"the 14-person team\" not \"the team\"), and occasional fragment. If your draft has none of these, it reads like AI.\n\n## Operational Checklist\n\nRun this before submitting any long-form writing:\n\n```\n[ ] Zero instances of the 21 red-flag words\n[ ] No formulaic opening sentence\n[ ] No courtesy closing (\"feel free to\", \"hope this helps\")\n[ ] Em dashes: fewer than 1 per 200 words\n[ ] No \"from X to Y\" false range constructions\n[ ] Sentence length variation present in every paragraph\n[ ] No \"not X, it's Y\" contrast structures\n[ ] All bullet points are genuine enumerations, not prose fragments\n[ ] No summary paragraph on content under 2,000 words\n[ ] At least one contraction in casual writing\n[ ] All attributions name a source or are deleted\n```\n\n## When to Apply\n\nApply this skill to:\n- Any written content destined for human readers\n- Blog posts, documentation, emails, proposals, social media\n- Situations where the writing being perceived as AI-generated would damage credibility\n\nDo not apply to:\n- Code comments (different conventions)\n- Internal tooling output read only by machines\n- Situations where the human explicitly wants systematic/structured AI output\n"}}