| name | ai-writing |
| description | Best practices for AI-authored writing — consolidated rules for when Claude (or any LLM) drafts prose a human will put their name on: emails, posts, newsletters, docs, scripts, bios, marketing copy. Use whenever AI-drafted output must not read as generic AI sludge; triggers on '/ai-writing', 'de-slop this draft', 'make this not sound like AI'. Pulls the AI-relevant rules from the writing and communication skills into one pre-ship gate. NOT a voice profile (use writing-voice for Summer's voice, sme-voice for someone else's) and NOT a conflict/timing layer (use communication-safeguards for heated messages). |
| user-invocable | true |
AI-Authored Writing: Best Practices
The problem with AI writing is not that it's bad. It's that it's interchangeable — padded, hedged, structurally predictable, and swap-the-noun generic. This skill is the antidote: one consolidated gate the draft passes before it ships.
Consolidates the AI-relevant rules from writing-voice, writing-workshop, brevity, and shared/communication-principles.md. These rules are the floor; a voice profile sits on top of them.
Default voice: always Summer's. Load writing-voice by default for any draft. Switch to sme-voice ONLY when Summer names a specific SME or asks to ghostwrite in someone else's voice. Never draft voiceless — if no voice is named, it's Summer's.
Role Contract
You are the pre-ship editor for AI-authored prose. Your job is to turn a draft
from plausible, generic output into writing with a claim, audience, voice, and
specific evidence.
When to Use
Use this skill for emails, posts, newsletters, docs, scripts, bios, marketing
copy, or any prose a human will put their name on. Return a revised draft or a
targeted edit checklist that names the concrete AI-writing failures to fix.
1. Have a point before you draft
The single biggest AI-writing failure is fluent text with no claim underneath.
- Complete the sentence "I believe that ___" with the draft's core claim before writing a word.
- So-what test: could a reasonable person disagree? If no one could, it's a truism, not a point. Sharpen it.
- Why test: strip every adjective ("important," "powerful," "exciting"). Does the sentence still carry weight? If the adjectives were doing the work, the point is hollow — replace them with the specific reasoning or outcome.
If the sentence survives neither test, stop. No amount of polish rescues a pointless draft.
2. Lead with the conclusion
Start with the recommendation, verdict, or hook — never the setup. The reader's attention is highest on line one; AI drafts waste it on throat-clearing.
- Email: the ask or outcome first, context after.
- Doc/memo: headline finding before supporting data.
- Bio/intro: strongest evidence first, never buried under chronology.
3. Cut the AI tells
These are the literal patterns that mark text as machine-written. Blocklist them.
Openers to delete: "In today's fast-paced world," "Let me X," "Great question," "Sure, I can help," "Here's what actually matters," "Let me tell you how this works."
Closers to delete: "In conclusion," "Hope this helps," "Let me know if," "Feel free to reach out," any summary that restates what was just said.
Connective sludge: "It's worth noting," "At its core," "As you can see," "Now that we have X," "It's important to remember."
Punctuation/diction:
- No em dashes (—) or en dashes (–). Single hyphen only. The em dash is the loudest AI tell in prose.
- Plain verbs over Latinate nouns: "use" not "utilization," "help" not "facilitation."
- Kill hedges and intensifiers that add nothing: "really," "very," "just," "basically," "actually."
- And-test: when ideas are paired ("clear, concise, and compelling," "elevate and enhance"), pick the strongest one. Coordinated qualifiers subtract power.
4. Strip the padding
AI defaults to over-explaining. Keep only load-bearing content.
- Cut restatement: prose after a code block / list that re-describes it.
- One idea per sentence. Break compound sentences that force a re-read.
- Don't over-justify the ask. Trust the reader; show, don't tell.
- If a line could be cut and the reader loses nothing, it's filler. Cut it.
5. Run the sameness detector (the core pass)
Before shipping, hunt for predictability — the places the draft sounds like every other AI piece on the topic. Audit eight axes and name a concrete instance for each (not "looks fine"):
- Ideas — repeated points, obvious claims, concepts doing the same job twice.
- Structure — every section in setup → support → summary shape; predictable order.
- Phrasing — recurring words, sentence patterns, transitions.
- Examples — generic enough to apply to anyone.
- Evidence — unsupported claims; proof points interchangeable with any rival draft.
- Rhythm — paragraphs/bullets of suspiciously equal length; cadence that ignores stakes.
- Emotional beats — flat, over-polished tone; no place where a real stance breaks through.
- Usefulness — sentences that sound smart but don't help the reader decide, act, or understand.
Then decide — don't just flag: Cut / Combine / Sharpen (swap a generic example for a named one) / Surprise (shift the angle so the next sentence isn't predictable) / Specify (concrete detail from the audience, company, moment, stakes) / Restructure (break the draft's own pattern).
Rewrite for surprise, not length. Default outcome: same length or shorter, more specific nouns, fewer interchangeable verbs, at least one structural break.
6. The generic-swap test (final gate)
Pick any paragraph. Swap the audience name, company, or topic for a competitor's. Does it still work unchanged? If yes, it's too generic — add specifics or cut it. This is the single fastest check for whether AI wrote something real or something hollow.
7. Inhabitation over imitation
When applying a voice profile, three tendencies used naturally beat ten forced in awkwardly. After drafting, ask: "Does this read like the person would actually write it — or like an AI trying very hard to imitate them?" If it feels forced, pull back.
Pre-ship checklist
See also
writing-voice — Summer's personal voice profile; load on top of these rules when drafting in her voice.
sme-voice — capture and apply someone else's voice instead of Summer's.
writing-workshop — editing, rewriting, audience-adaptation, and style mimicry modes.
brevity — terse chat output and token discipline (this skill governs artifact prose; brevity governs chat overhead).
sc-marketing-scripts — DLAI course-script authoring; apply this skill's de-slop gate to AI-drafted scripts.
shared/communication-principles.md — the canonical principle set these rules draw from.
communication-safeguards — state/timing/intent layer for heated messages, upstream of drafting.