ai-writing

name	ai-writing
author	[ContractorKeith](https://github.com/ContractorKeith)
description	Write articles, blog posts, tutorials, and long-form content that reads like a human wrote it. Use this skill whenever the user asks to write an article, blog post, tutorial, guide, essay, or any long-form written content, even if they don't explicitly mention "humanized" or "anti-AI." Also trigger when the user asks to "write something that doesn't sound like AI," improve writing quality, or draft content for publication on a blog, newsletter, or website. This skill applies rigorous anti-AI-detection writing patterns derived from linguistic research to produce prose that is specific, grounded, and stylistically human.
aliases	["ai-writing","ai-writing-skill","AI Writing"]
type	skill
status	active
tags	["skill","writing","content","anti-ai","voice"]

AI Writing: Human-Voice Long-Form Content

Write articles, blog posts, and tutorials that read like they were written by an experienced human writer, not a language model. This skill encodes dozens of known AI writing tells (catalogued by Wikipedia editors, computational linguists, and AI detection researchers) as negative constraints (things to actively avoid) alongside positive craft principles that produce genuinely good prose.

Intent (read first)

This skill exists to help Keith write content that genuinely helps people and grows an authentic following. It does NOT exist to convert readers into consulting clients, drive booked calls, or sell services.

When in doubt while drafting, ask: Would a reader who couldn't hire Keith for anything still get value from this? If the honest answer is no, the piece is mis-aimed — restructure before continuing.

This intent shift matters because some of the older sample articles in writing-examples/ (specifically the contractor-targeting "Jack-of-All-Trades" and "Hiring Subs" pieces) were written when Keith was actively pitching consulting. Their structural patterns — prescriptive listicles, CTAs to "watch my video," dictionary-entry openings — are deprecated. Use them only as cautionary references for what to avoid. The canonical voice samples are the Rock Bottom article and thread.

How to use this skill

1. Read references/voice-anchor.md — this is non-negotiable for writing in Keith's voice. It contains the canonical voice markers (specificity, fragments, vulnerability, conjunction openers, named characters, open-ended endings) and the audience/intent framing distilled from his current canon samples.
2. Read the user's topic, audience, and any style/tone preferences.
3. Load references/ai-writing-antipatterns.md to refresh the full anti-pattern list.
4. If writing from material in an Obsidian vault project (20-Projects/<project>/research/), end every factual claim with a [[wikilink]] to its source note. Citation discipline is non-negotiable when drafting from the vault.
5. Draft the article following the Writing Rules below, calibrated to the genre (see Genre Adjustments).
6. Self-check the draft against the Anti-Pattern Checklist AND the voice-anchor checklist before delivering.
7. If invoked from a path under 20-Projects/<project>/, save the draft to 20-Projects/<project>/drafts/<slug>-<YYYY-MM-DD>.md with proper frontmatter (type: draft, status: draft, tags: [draft, <project-tags>]) instead of returning in chat.

Genre adjustments

The base rules below are tuned for medium-length blog posts (~1500–3000 words). Adjust per genre:

• Long-form essay (3000+ words): Double the hard requirements (2 fragments, 4 conjunction openers, 2 short/2 long sentences). Multiple opening scenes acceptable.
• Short blog post (under 800 words): One of each hard requirement is enough. Be ruthless about cuts.
• Newsletter: Looser, more conversational. More first-person ("I noticed this week..." framing welcomed). Open-door endings work especially well.
• LinkedIn long-form: Single-line paragraphs more aggressively. Lead with a hook line, then expand. Personal scene in opening is critical.
• X long-form / threads: Each post should pass the rules independently. Sentence fragments become structural, not occasional. White space does the pacing.
• Technical documentation / tutorial: Specificity and concrete examples matter even more. Self-deprecation sparingly. The reader wants to learn the thing, not be entertained, but humanized prose still reads better than AI prose.

---

Writing Rules

Voice and Tone

• Write in first person or direct second person unless the format requires otherwise. Real writers have a point of view. Use "I," "you," "we" naturally.
• Vary sentence length dramatically. Mix 5-word punches with 30-word complex sentences. AI text clusters around 15-22 words per sentence. Break that pattern. Include at least one sentence under 8 words AND at least one sentence over 25 words in every article.
• Use sentence fragments intentionally. Not a suggestion. A requirement. At least one per article. Fragments like "Not even close." or "Worth it." add rhythm that AI avoids.
• Start at least two sentences with "And," "But," "So," or "Because." These informal openers signal conversational fluency. AI text almost never begins a sentence with a conjunction. Doing so is one of the strongest human tells you can deploy.
• Deploy dry humor, understatement, or mild irreverence when appropriate for the audience. AI text is relentlessly earnest. Humans are not.

Word choice: the banned list

The following words and phrases are statistically overrepresented in AI output and must be avoided entirely unless quoting a source or using the word in its literal, non-figurative sense:

Significance/Legacy inflation:

• stands as, serves as (when "is" works), is a testament to, is a reminder that
• vital/significant/crucial/pivotal/key role/moment
• underscores/highlights its importance/significance
• reflects broader, symbolizing its ongoing/enduring/lasting
• setting the stage for, marking/shaping the, represents/marks a shift
• key turning point, evolving landscape, focal point, indelible mark, deeply rooted

AI vocabulary (high-frequency tells):

• delve, foster/fostering, bolster/bolstered, garner, showcase/showcasing
• tapestry (figurative), vibrant (describing anything abstract), meticulous/meticulously
• intricate/intricacies, interplay, underscore (as verb), pivotal, enduring
• enhance (prefer "improve," "strengthen," or a more specific verb)
• landscape (as abstract noun, e.g. "the competitive landscape")
• Additionally (especially sentence-initial), moreover (sentence-initial)
• align with, valuable insights, navigate (figurative), leverage (as verb)
• comprehensive, robust (describing anything non-physical)
• It's important to note, It's worth noting, It should be noted
• In today's [anything], In the realm of, When it comes to
• Whether you're a [X] or a [Y]

Promotional/travel-guide tone:

• nestled, in the heart of, renowned, groundbreaking, diverse array
• exemplifies, commitment to (corporate-speak), natural beauty (generic)
• rich (figurative, e.g. "rich cultural heritage"), profound (figurative)
• boasts a, captivates, dynamic hub, gateway to

Copula avoidance (AI replaces "is/are/was" with fancier verbs):

• Do NOT replace "is" with "serves as," "stands as," "marks," "represents"
• Do NOT replace "has" with "boasts," "features," "offers" when the simpler verb works
• Prefer "is," "are," "was," "has." Simple copulas are a human tell, not a weakness

Structure: anti-AI patterns

• No "Challenges and Future Outlook" sections. This is the single most identifiable AI structural pattern. If you must discuss challenges, weave them into the narrative naturally rather than quarantining them in a formulaic closing section.
• No "Despite its [positives], [subject] faces challenges..." formula. Ever.
• No preamble definitions. Do not open articles by defining the topic as if writing a dictionary entry. Jump into the story, the problem, the hook.
• No "Not just X, but also Y" parallelisms unless genuinely contrasting two things. AI uses this construction to sound balanced; humans just make the point directly.
• No rule-of-three lists disguised as prose. AI defaults to "adjective, adjective, and adjective" or "short phrase, short phrase, and short phrase" everywhere. Use lists of 2, 4, or 5 items. Break the triplet habit.
• No elegant variation for its own sake. If the subject is "fence estimating," call it "fence estimating" every time. Do not rotate through "this discipline," "the craft," "this essential practice," "the field." Repetition of the actual term is more human than thesaurus gymnastics.
• No superficial analysis via trailing participle phrases. Avoid sentences that end with "...highlighting the importance of X" or "...underscoring the need for Y" or "...reflecting broader trends in Z." These trailing -ing phrases are an AI fingerprint. If the point matters, give it its own sentence with evidence.
• No vague attributions. Avoid "experts say," "researchers note," "critics argue," "many believe" without naming specific people or sources. AI uses these to simulate authority. Humans cite by name or skip the attribution entirely.

Formatting: anti-AI patterns

• Do not use em dashes. AI overuses em dashes as a crutch to punch up clauses and insert asides. This is one of the most recognizable AI formatting tells. Use commas, parentheses, colons, or periods instead. Every em dash in a draft should be replaced with one of those alternatives. Zero tolerance.
• Use sentence-case headings, not Title Case. AI strongly defaults to capitalizing every major word in headings. Sentence case (only first word and proper nouns capitalized) reads as more human.
• Do not decorate headings or bullets with emoji. This is a strong AI tell, especially in instructional content.
• Do not create unnecessary tables. AI generates small tables where prose would be clearer. Use tables only when comparing 3+ items across 2+ dimensions.
• Use straight quotes and apostrophes, not curly/smart quotes (mainly a ChatGPT tell, but worth noting).
• Bold sparingly. AI bolds key terms aggressively. Bold only for genuine emphasis, not as a substitute for clear writing.

Content quality: positive principles

• Rhythm check: short sentences are mandatory, even in technical articles. After every 2-3 paragraphs, insert a sentence of 7 words or fewer. "That matters." "It wasn't." "Worth knowing." Technical topics especially need these punches to stay readable. If every sentence in a section is 12-25 words, the prose has gone flat.
• Break up data-dense passages with rhythm shifts. When listing facts, numbers, or arguments in sequence, never let more than 2 consecutive sentences be similar length. After two medium-length fact sentences, insert a very short sentence (under 8 words) or a very long one (over 25 words) before continuing. This is especially important in short opinion pieces where data points stack up fast.
• Lead with the specific, not the general. Instead of "Construction is a complex industry," open with "The first time I priced a 6-foot cedar privacy fence wrong, it cost me $2,400." Concrete details before abstractions.
• Include at least one "only-I-would-know-this" detail per section. Personal anecdotes, specific numbers, named tools, real places. These details are impossible to hallucinate and signal authentic experience.
• Let the reader draw conclusions. Present evidence and let the implication land. Do not append "This demonstrates the importance of..." after every anecdote.
• End sections with a forward pull, not a summary. Instead of restating what you just said, end with a question, a transition to the next idea, or a half-open thought that makes the reader keep going.
• Cite real sources by name when making claims. "A 2024 study in Science Advances" is better than "research shows." "Karpathy's autoresearch repo" is better than "recent developments in autonomous optimization."
• Include appropriate hedging. Real writers say "I think," "in my experience," "from what I've seen." AI text presents everything as settled fact. Calibrate confidence to match actual certainty.

---

Titles, Hooks, and Reader Psychology

Writing titles that don't sound like AI

• Avoid the "[Number] [Plural Noun] That Will [Verb] Your [Noun]" formula. AI defaults to listicle-style titles. "7 Strategies That Will Transform Your Workflow" is an AI tell. Write titles the way a magazine editor would: short, specific, slightly incomplete so the reader has to click.
• Use lowercase unless it's a proper noun. "How fence estimating actually works" reads more human than "How Fence Estimating Actually Works."
• Prefer statements or questions over promises. "I stopped using spreadsheets for estimates" beats "Why You Should Stop Using Spreadsheets for Estimates." The first sounds like a person. The second sounds like content marketing.
• Keep titles under 10 words when possible. Short titles feel confident. Long titles feel like they're trying to SEO their way into a click.
• Never use "The Ultimate Guide to" or "Everything You Need to Know About." These are the title equivalents of banned AI vocabulary.

Opening hooks

• The first sentence determines whether anyone reads the second. Do not waste it on context-setting or definitions. Open with a scene, a number, a question, a confession, or a claim that makes the reader want to disagree.
• Name something specific in the first two sentences. A dollar amount, a date, a place, a person's name. Specificity in the opening signals to the reader that this was written by someone who was actually there.
• Avoid throat-clearing openers. "In today's fast-paced world..." or "When it comes to [topic]..." or "If you've ever wondered about..." are all AI openers. Cut them. Start with the thing itself.

Keeping the reader moving

• Every section should end with a reason to read the next one. A question, a half-open thought, a transition that pulls forward. Never end a section by summarizing what you just said. Summaries tell the reader they can stop.
• Vary paragraph length as deliberately as sentence length. A one-sentence paragraph after a dense block of text creates visual breathing room and emphasis. AI text produces uniform 3-5 sentence paragraphs. Break that pattern.
• Bury your best insight in the middle, not the introduction. Readers who make it to the middle of an article are committed. Reward them. AI front-loads every important point in the intro and then restates it. Human writers build toward something.
• Use tension and unresolved questions. State a problem before offering the solution. Describe what went wrong before explaining what you learned. The gap between "what happened" and "what it means" is what keeps people reading.

---

Anti-Pattern Self-Check

Before delivering any article, run this mental checklist against the draft. Hard requirements (non-negotiable, even on short articles under 500 words): Verify the draft contains: (a) at least one sentence fragment, (b) at least two sentences starting with "And," "But," "So," or "Because," (c) at least one sentence under 8 words, (d) at least one sentence over 25 words. If any are missing, add them before proceeding.

1. Banned words scan: Does the draft contain any words from the Banned List used in their AI-typical figurative sense? → Rewrite those sentences.

2. Copula check: Count uses of "is/are/was/has." If they feel scarce and everything is "serves as" or "represents," → simplify the verbs.

3. Em dash scan: Any em dashes in the draft? → Replace all with commas, parens, colons, or periods. Zero allowed.

4. Trailing -ing phrases: Any sentences ending with "...highlighting/underscoring/ reflecting/emphasizing X"? → Rewrite as standalone claims with evidence.

5. Rule-of-three scan: More than one "X, Y, and Z" list in a single section? → Vary the list lengths.

6. Opening sentence test: Does the article open by defining its own topic? → Rewrite with a hook, anecdote, or provocative claim.

7. Legacy/significance bloat: Any sentences about the topic's "enduring legacy," "broader significance," or "lasting impact"? → Delete or replace with specific evidence.

8. Heading case check: Are headings in Title Case? → Convert to sentence case.

9. Specificity test: Does every section contain at least one specific name, number, date, or place? → Add concrete details where missing.

10. Conclusion check: Does the article end with a "Challenges and Future Outlook" section or a formulaic "In conclusion..." paragraph? → Rewrite the ending.

11. Promotional tone scan: Does any section read like a press release, travel brochure, or sales page? → Rewrite with flat, factual language.

12. Voice-anchor checks (from references/voice-anchor.md):

    • Could only Keith have written this? Specific names, numbers, places, dates that make it unfakeable? → If generic, add concrete detail.
    • Did the opening drop the reader into a scene, or define a topic? → If the latter, restructure with a scene or specific moment.
    • Is there at least one moment of honest hedging or admission? → If everything is stated as settled fact, calibrate confidence.
    • Does the ending open a door (story, question, "reach out") rather than close with a sales CTA? → If it tries to convert, soften to an open-door pattern.
    • Is there ZERO "watch my video / book a call / hire me" framing? → Strip if present.
    • Would a reader who couldn't hire Keith for anything still get value? → If no, restructure to lead with the useful insight, not the sales hook.

---

Reference Files

• references/voice-anchor.md : Read first. Keith's canonical voice (distilled from the Rock Bottom article + thread) plus current intent framing (genuine help / authentic following, NOT consulting conversion). Includes voice markers, anti-markers, audience definition, and a publish-ready checklist.
• references/ai-writing-antipatterns.md : Complete catalogue of AI writing tells with examples, organized by category (Content, Language, Style, Formatting). Read for the full list with contextual notes beyond the summary above.

Sample files (writing-examples/)

• CANON (mimic these): From Rock Bottom to Resilience..., Hitting Rock Bottom at 50
• DEPRECATED (avoid mimicking): Why Being a Jack-of-All-Trades..., What To Look For When Hiring Subs_
• See references/voice-anchor.md for the full breakdown of which samples are canon vs. deprecated and why.

Workflow integration

When working inside the Obsidian vault, this skill pairs with:

• Before this skill: [[research-synthesis]] to ground the draft in cited source material; [[content-cornerstone-outline]] to outline structure first
• After this skill: [[content-shorts-from-cornerstone]] to derive distribution posts from the finished cornerstone