10 Rules to Eliminate “AI-Smelling” Technical Writing from LLMs | Technical Book Publisher’s SKILL.md

AI-generated technical documents often exhibit a characteristic “AI smell” — polished in form but lacking in substance. A practical set of rules to eliminate this has been published on GitHub Gist by Keiichiro Shikano, representative of technical book publisher Lambda Note. Featured in a TechnoEdge article, this SKILL.md provides actionable guidelines for producing high-quality Japanese technical writing with LLMs.

📑Table of Contents

1. What is the “AI Smell” in LLM-Generated Documents?
2. Background and Purpose of the Published SKILL.md
3. Overview of the 10-Chapter Writing Standards
4. Prohibiting LLM-like Expressions (Core Rules and Examples)
5. Practical Application of Formatting, Paragraph Structure, and Argument Rigor
6. Managing Reader Load and Perspective/Narration Techniques
7. Points on Suppressing Dramatization and Eliminating Redundancy
8. How to Use the Gist and Real-World Application Examples
9. Verification of Effects and Impact on Technical Book Quality
10. Frequently Asked Questions (FAQ)
11. Summary

---

What is the “AI Smell” in LLM-Generated Documents?

LLM outputs frequently display an “AI smell”: superficially well-structured text that fails to convey meaningful information. Common traits include meaningless previews, summary phrases, vague adjectives, and repetitive conjunctions. Readers struggle to grasp the core message.

In practice, this arises because LLMs generate pattern-based “plausible” text. Avoiding these patterns builds reader trust in technical documentation.

---

Background and Purpose of the Published SKILL.md

Lambda Note publishes numerous technical books. Shikano, its representative, shared the company’s internal writing standards as “japanese-tech-writing/SKILL” on Gist. The goal is to equip humans with skills to leverage LLMs effectively for quality technical writing, not to improve raw LLM output.

Source: GitHub Gist (as of June 22, 2026)

---

Overview of the 10-Chapter Writing Standards

The SKILL.md comprises 10 chapters: 1. Formatting rules 2. Paragraph and argument structure 3. Rigor of argumentation 4. Managing reader load 5. Perspective and narration 6. Suppressing dramatization 7. Prohibiting LLM-like expressions 8. Eliminating redundancy 9. Heading practices 10. Honesty toward readers

Applying these sequentially improves document quality.

---

Prohibiting LLM-like Expressions (Core Rules and Examples)

The core focus is banning LLM-like phrasing. Key examples are summarized below.

Prohibition Category	Example	Reason
Meaningless previews	“The important point is…” “This chapter covers…”	Foreshadows without substance
Summaries	“In summary…” “To put it simply…”	Creates redundancy
Vague adjectives/verbs	“Key…” “Deep dive…” “Verbalize…”	Lacks specificity
Conjunction overload	“Furthermore…” “Also…” “In addition…”	Obscures logical flow
Weak qualifiers	“It could be said…” “Very…” “Extremely…”	Dilutes assertions

Avoiding these makes writing direct and persuasive.

---

Practical Application of Formatting, Paragraph Structure, and Argument Rigor

Formatting emphasizes one sentence per line and one topic per paragraph. Explicitly state causal mechanisms and avoid unnecessary proper nouns. Arguments must detail “why” with concrete evidence.

---

Managing Reader Load and Perspective/Narration Techniques

Reduce cognitive load by maintaining consistent perspective and explaining terms on first use.

---

Points on Suppressing Dramatization and Eliminating Redundancy

Focus on facts and logic; cut superfluous expressions for tighter prose.

---

How to Use the Gist and Real-World Application Examples

Incorporate the rules into prompts to improve LLM output quality. One-sentence-per-line discipline proved effective in actual technical book writing.

---

Verification of Effects and Impact on Technical Book Quality

Applying these rules led to better reader feedback and increased trust in publications. The TechnoEdge article highlights their relevance in the LLM era.

Source: TechnoEdge article

---

Frequently Asked Questions (FAQ)

Q1: Can these rules be applied directly to LLM output?

Yes. Adding them to prompts significantly reduces AI smell, though human review is recommended.

Q2: Can they be used outside technical books?

Effective for blogs, documentation, proposals, and any logical writing.

Q3: Won’t avoiding prohibited expressions make text too short?

It may shorten text, but essential information remains, improving readability.

Q4: Are they actually used in Lambda Note books?

Yes, as internal standards contributing to publication quality.

Q5: Will the Gist be updated?

It is public; feedback via forks or comments is welcome.

Q6: Can beginners practice this?

Start by applying one category from the table at a time.

---

Related articles:

• Boltz Bio、BoltzMol-1 / BoltzProt-1 をリリース — 創薬向け新モデル、Claude Code / Codex 統合対応
• 画面操作を“録画”→AIが作業代行 Codexに新機能「Record & Replay」
• 【完全版】AIで誰でもポートフォリオを無料計算できる方法（REITやゴールドもOK!）

Summary

This SKILL.md is a practical guide for writing high-quality technical documents while leveraging LLMs. Eliminate AI smell and prioritize reader-centric writing. Refer to the GitHub Gist and TechnoEdge article for details.

As a next step, apply these rules to your own writing or prompts.

Related new article:

• 4285 – This published update adds current operational context for 「AI臭い文章を生成させない」ルール集。LLMに“質の高い技術文書”を書かせるスキルを技術書出版社代表が公開（生成AIクローズアップ） | テクノエッジ TechnoEdge.
• 4308 – This published update adds current operational context for 「AI臭い文章を生成させない」ルール集。LLMに“質の高い技術文書”を書かせるスキルを技術書出版社代表が公開（生成AIクローズアップ） | テクノエッジ TechnoEdge.
• PHOTON LLM Architecture Claims 475x Transformer Throughput — Major GPU Efficiency Breakthrough – This published update adds current operational context for 10 Rules to Eliminate “AI-Smelling” Technical Writing from LLMs | Technical Book Publisher’s SKILL.md.
• Boogu-Image-0.1 and Krea 2 Shake Up AI Gravure Creation — Challenging Z-Image – This published update adds current operational context for 10 Rules to Eliminate “AI-Smelling” Technical Writing from LLMs | Technical Book Publisher’s SKILL.md.
• Baidu Releases Free Local OCR Model “Unlimited OCR” for One-Shot Multi-Page PDF Processing, Commercial Use Allowed – This published update adds current operational context for 10 Rules to Eliminate “AI-Smelling” Technical Writing from LLMs | Technical Book Publisher’s SKILL.md.
• Beyond RAG: Implementing Agent Search with LangGraph for Knowledge Operations – This published update adds current operational context for 10 Rules to Eliminate “AI-Smelling” Technical Writing from LLMs | Technical Book Publisher’s SKILL.md.