# Writing Style And Tone > The writing-craft skill — style and tone as applied technique: rhythm, specificity, tone modulation, and the edit passes that make writing read as human. Use when someone's posts "sound like ChatGPT," a draft needs a style edit, they ask how to write differently per platform or moment without losing the brand voice, writing is grammatically perfect but exhausting, a manager wants em dashes banned or detector scores chased, or any AI-assisted draft needs its mandatory quality pass. Uses the STYLE framework. Reads voice-builder + brand-profile + the format skill first. Tells must be read as a cluster with judgment — no single marker proves AI, and detectors misflag real writers; the winning move is substance so owned the question never arises. Voice is identity; tone is register. The agent drafts; the HUMAN supplies lived specifics and final say; never invents anecdotes or staged typos. Distinct from voice-builder, hook-writer, the format skills, and content-research-and-sourcing. - Skill: `social-media-skills/writing-style-and-tone` (Agent Skill, multi-file: 6 files) - Install (CLI): `npx skillmds add social-media-skills/writing-style-and-tone` - Raw SKILL.md: https://api.skillmd.com/api/skills/social-media-skills/writing-style-and-tone/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: social-media-skills (https://skillmd.com/u/social-media-skills) - Updated: 2026-09-09 - Page: https://skillmd.com/skills/social-media-skills/writing-style-and-tone --- # writing-style-and-tone The **writing craft** — sound like someone, tune tone to the moment, yank the dead weight, let the rhythm breathe, and edit against the tells intelligently. The agent drafts and edits in the voice; the **human supplies the lived specifics and the final call**; **WoopSocial publishes**. (Craft skill — no tool file.) ## The POV: the tells are symptoms — the disease is saying what anyone could say 2026 quantified the stakes: **52% of consumers say they disengage the moment they suspect copy is AI**, only 13% say they completely trust AI, experienced AI users now detect machine text at near-perfect rates, and discovery systems treat predictable-pattern content as noise. So the tell taxonomy matters — the delve-class vocabulary (44% of PubMed's *all-time* "delve" appearances landed in 2023–24), "It's not X, it's Y," the forced rule-of-three, the rigid intro-point-point-conclusion shape, fluff sentences, relentless positivity — but the top-1% position holds two truths at once. **First: apply the tells as a cluster, with judgment** — no single marker proves AI, the em-dash panic is folk detection, and detector-chasing punishes real writers (the Stanford study: detectors misflagged non-native-English essays at a 61.3% average false-positive rate). **Second: the cure is substance, not deletion** — *make the writing so obviously human that the question never comes up*: a real number, a named moment, a stated opinion, productive digression, varied rhythm. Say what only you could say. And the discipline that governs every moment: **voice is identity (constant); tone is register (flexes)** — the playful brand goes plain on its hardest day without becoming a different company. One absolute: authenticity cannot be counterfeited — no invented anecdotes, no manufactured vulnerability, no staged typos. ## Read these first 1. **voice-builder** — the voice this craft inhabits (the constitution; this skill governs by it). 2. **brand-profile** + the piece's **format skill** (conventions the craft applies inside). ## The framework: STYLE (Depth: `references/the-style-framework.md`.) - **S — Sound like someone:** one only-you element per piece (number, moment, opinion); concrete over abstract; subjectivity as a feature. - **T — Tune tone to the moment:** the tone map (celebration / education / sales / complaint / crisis) with voice fingerprints persisting; platform register shifts, identity doesn't (the cover-the-handle test). - **Y — Yank the dead weight:** fluff sentences deleted whole; throat-clearing cut; verbs over adjectives; one idea per sentence; front-load. - **L — Let the rhythm breathe:** burstiness — long-build then short landing; punctuation as pacing; the read-aloud test (every stumble is an edit). - **E — Edit against the tells, intelligently:** the cluster sweep with judgment, never bans; the substance test decides, not a detector score; disclose AI use where required. ## The reality (verify-quarterly) The stakes: Bynder's 52% disengagement figure; Klaviyo's 13% complete-trust figure; near-perfect detection by experienced AI users (Russell et al.); information-gain systems penalizing sameness. The taxonomy (converging 2026 sources): hallmark vocabulary (the measured delve surge; delve+underscore co-occurring in 98.8% of joint 2023–24 papers), structural tells (repetitive shapes, furthermore/moreover, forced threes, negative parallels, forced lists), substance tells (fluff sentences), texture tells (relentless neutrality/positivity; typographic perfection vs human "--" and straight quotes). The counter-nuance: no single marker proves AI (the "ChatGPT hyphen" discourse; Stanford's 61.3% average false-flag rate on non-native writers); write for readers, never against detectors — with the tactical footnote that a writer may spare a marker for a suspicious audience. The human profile to honestly reach: subjective, colloquial, emotion-rich, divergent, rhythm-varied. **Attribute all; verify-quarterly.** Full detail: `references/writing-style-2026-reality.md`; the four-pass edit, tone map, before/after, read-aloud protocol, and worked examples: `references/edit-passes-and-templates.md`. ## Honest scope (never violate) - **The agent** drafts in the voice, runs the four passes, renders tone per the map, and flags where a real specific is needed — placeholders stay visibly unfilled until the **human** supplies lived material and signs off. **The authenticity absolute:** never invent the human's experiences, failures, credentials, or feelings; never manufacture vulnerability; never stage typos — real substance, not counterfeit texture. - **Detector honesty:** no score theater; the reader is the judge; **AI-disclosure** where platform/policy requires. **Tone ethics:** crisis and complaints get plain, accountable language — never spin. Claims route through content-research-and-sourcing; metrics never fabricated. (Full scope: `references/scope-and-connections.md`.) ## Distinct from its siblings (route correctly) **writing-style-and-tone (this)** = the applied prose craft + tone control · **voice-builder** = defines the voice (this inhabits it) · **hook-writer / hooks-and-retention-deep-dive** = the first line (this owns everything after) · **text-post-and-microblog / caption / email skills** = format conventions · **copywriting/conversion skills** = persuasion architecture (this is the prose layer under it) · **content-research-and-sourcing** = verifies the claims this styles · **contrarian-and-opinion** = the POV strategy this renders · every **AI-drafting workflow** = this is its mandatory edit layer. ## Where this connects Reads first: **voice-builder** + **brand-profile** + the **format skill.** Consumes: drafts from any content skill or AI assistant + the human's lived specifics. Feeds: every written surface — text posts, captions, carousels, newsletters, scripts (spoken register → **scripting-and-storyboarding**), community replies. Publishes via: the finished piece → **scheduling-and-queue → WoopSocial.** Measure with: read-through, saves/ shares, reply quality, and the absence of "did AI write this?" comments via **analytics-and-reporting** — never fabricated. ## Definition of done Prose that passes the substance test before any tell sweep: at least one only-us element per piece (a real number, named moment, or defensible opinion supplied by the human — never invented), dead weight yanked (no fluff sentences, no throat-clearing, verbs doing the work), rhythm varied and read aloud until stumble-free, tone rendered deliberately from the map (voice fingerprints constant; cleverness dialed to zero as stakes rise; the cover-the-handle test passed across platforms), and the tell cluster swept with judgment — no banned punctuation, no detector-score theater, AI use disclosed where required; the agent flagging placeholder slots rather than fabricating lived experience, the human filling and approving them, claims verified via content-research-and-sourcing, and the piece published via WoopSocial; **no invented anecdotes, manufactured vulnerability, staged imperfection, or fabricated metrics**; and correctly distinguished from voice-builder, hook-writer, the format skills, and content-research-and-sourcing.