muvon/octomind-tap/content-voice

Overview

Write clear, useful content in the actual author's voice. Apply this to drafting and revision across posts, replies, articles, and scripts; platform skills provide local format and participation rules. These are editorial practices, not a test of whether a human or AI wrote something.

Mental model

Professional voice comes from a definite point, credible detail, and respect for the reader. Natural writing can be formal, conversational, technical, or concise. It does not require slang, mistakes, dramatic rhythm, or a story about the speaker.

Establish the brief

• Identify the speaker, intended reader, task, venue, and source material. Use approved samples to understand register, terminology, humor, and level of detail; don't copy another person's identity or invent a persona.
• With no voice samples, use clear, restrained prose and the supplied facts. Missing samples do not block an ordinary draft. Missing evidence blocks the unsupported claim.
• Preserve numbers, names, quotations, dates, technical terms, conditions, and uncertainty. A polished sentence must not make the claim stronger.
• First person is appropriate for experience or decisions the speaker supplied. Researching a product does not authorize “I tested it.” Don't invent customers, struggle, excitement, vulnerability, or quotations to create personality.

Write for the reader

• Lead with the useful fact, observation, answer, or decision. A hook should tell readers why the content matters and be paid off by the content; don't manufacture suspense around an ordinary fact.
• Give enough context to understand the point without the planning brief. Explain unfamiliar terms once; preserve precise domain vocabulary when the audience needs it.
• Prefer concrete verbs and identifiable actors. Replace “a transformative solution” with what it actually does, if known. If specifics are unavailable, narrow the claim instead of supplying plausible details.
• Literal before clever: a headline, tagline, or opener must make sense to a stranger without context. A slogan or wordplay can sit beside a literal line, not replace it.
• A comparison names the alternative and the measure ("imports 1M rows in 40 s; Tool B takes 95 s on the same machine"). "Faster than alternatives" or "the closest competitor" needs both filled in from the source, or it goes. Prefer the like-for-like comparison (same model, hardware, or input) and state the condition.
• Name the members of a set when the source does ("PostgreSQL, MySQL, and SQLite"), instead of "major databases"; mark a partial list ("including", "such as") so it doesn't read as complete. Replace abstract nouns with the observable effect ("resumes the upload after the connection drops", not "ensures robust connectivity").
• Use the reader's standard term over a coined one; if the subject has its own term, map it once to the known one. Call the product by name, not "the engine" or "the platform".
• Define a term before you reuse it; don't make the reader meet an insider noun cold. Internal names (config keys, module or class names) appear only in code formatting where the reader needs them to find something; elsewhere use the plain word.
• Outcome before mechanism: what the reader gets, then how it works. Evidence goes near the top, next to the claim it supports, not in a closing section.
• Cut filler adverbs that assert ease or speed without measuring it: easily, simply, just, quickly, very, effectively (Google and Microsoft style guides).
• Remove throat-clearing, repeated claims, unsupported superlatives, and abstract modifiers. Words such as “innovative,” “seamless,” or “leverage” invite a meaning check, not a mechanical ban. Literal or technical uses may be correct.
• Keep terminology consistent. Don't rotate “users,” “customers,” and “buyers” for variety when they name different people.
• State an evidenced position clearly; keep real qualifications. “May,” “in this sample,” and “under these conditions” can be essential to accuracy, not weakness.

Rhythm and structure

• Read for comprehension and natural emphasis. Split an overloaded sentence or join choppy ones when it improves the logic. Don't enforce sentence-length alternation, contraction percentages, fragments, or punctuation quotas.
• Contractions, fragments, lowercase, and slang should fit the author's register and venue. Correct accidental errors; never add typos, fake edit notes, broken grammar, or invented dialect as camouflage.
• Use active voice when it clarifies responsibility; passive voice is useful when the actor is unknown or the process/result deserves focus.
• Treat these as review prompts, not bans: repeated "not just X, but Y" contrasts, forced groups of three, padded adjective lists, staccato paragraph stacks, compulsory morals, inflated significance ("pivotal," "testament to"), vague attributions ("experts say"), trailing "-ing" clauses that assert significance ("..., highlighting its importance"), "serves as" or "boasts" in place of "is" or "has," buzzword compounds ("-grade," "-native," "-first," "-powered"), outline-style "challenges and future outlook" endings, title-case headings, emoji bullets, bold scattered through running prose, bold inline headers on every bullet, and chatbot residue ("I hope this helps," "it's important to note"). Keep a real comparison, a meaningful list of three, or a useful summary.
• Use questions when their answers matter, not as automatic openers or “Agree?” endings. End after the point, a useful next action, or an unresolved question grounded in the subject; no mandatory twist or inspirational closer.
• Use headings, lists, and tables when they help readers navigate or compare. A technical reply may need steps or code; a brief conversational reply usually needs prose. Length and formatting follow the task, not an authorship stereotype.

Platform adaptation and trust

• Adapt the same factual core to each venue's reader intent and medium. A narrated demo, a support reply, and a professional announcement need different context and pacing; changing a few synonyms is not adaptation.
• Read the parent post and relevant replies before answering. Add a specific answer or respectful correction; skip empty praise, imitation of the speaker, and unrelated promotion.
• Give a legitimate CTA only when useful. Asking a real question or inviting a relevant action is different from manufactured engagement, false urgency, or a promised reward that cannot be delivered.
• Keep commercial affiliation and required provenance disclosures visible. For US-facing endorsements, FTC guidance requires truthful experience and clear material-connection disclosures; platform controls alone may not communicate the relationship sufficiently. Check applicable local and platform rules.
• Keep alt text descriptive and faithful to visible information; retain material chart labels or provide equivalent explanation. Review captions for names, negation, numbers, and timing. Accessibility fields are not keyword dumps or extra sales copy.
• For another language, match local register and meaning rather than translating English catchphrases literally. Use contextual evidence or qualified review for uncertain idiom; low search frequency is not proof that a phrase is wrong.

Examples

Illustrative source facts: a product exports CSV; importing CSV is not supported.

Before:

This revolutionary update unlocks seamless data freedom.

After:

You can export your data as CSV. CSV import isn't supported.

The rewrite adds only the stated capability and limitation. It does not invent a user result.

Illustrative source: the author's team tested one repository; runtime was not measured.

Avoid:

We made every workflow dramatically faster.

Use:

We tested this on one repository. We haven't measured runtime yet.

Illustrative source: a 1M-row import benchmark on one laptop; the product took 40 s; Tool B took 95 s.

Avoid:

Blazing-fast imports that leave the competition behind.

Use:

Imports 1M rows in 40 s on a laptop; Tool B takes 95 s on the same machine.

Checklist

• [ ] The opening delivers a specific point for the intended reader.
• [ ] Every experience, result, and quotation belongs to the stated speaker and source.
• [ ] Edits preserve scope, uncertainty, and technical meaning.
• [ ] Headlines and openers are literal; every comparison names the alternative and the measure; sets name their members.
• [ ] Rhythm, formatting, and vocabulary serve comprehension without artificial quotas.
• [ ] Platform participation, disclosure, accessibility, and publication conditions are satisfied.
• [ ] The final copy is separate from internal source notes and unresolved questions.

References

Reviewed 2026-10-12. Apply source principles within their stated scope; they do not establish social-platform ranking factors. This is a review date, not a publication/update date for the sources below.

• GOV.UK clear-language guidance — a public-service style guide supporting audience-focused clarity, not a universal brand voice; its A–Z "words to avoid" entry lists terms such as drive, robust, leverage, and deliver used for abstract things.
• NN/g: concise, scannable, and objective writing — in a 51-user study, objective copy measured 27% more usable than promotional copy, and concise + scannable + objective 124%.
• Google developer style word list and Microsoft: simple words, concise sentences — filler adverbs, "use" over "utilize/leverage", one term per concept.
• FTC endorsement disclosures — US-facing endorsements and truthful experience.
• W3C image alternatives — communicate an image's purpose and relevant information.
• Google's AI-content guidance — accuracy and reader value, not guaranteed traffic from “human” style.
• Wikipedia: Signs of AI writing — descriptive catalogue of formulaic patterns; single signs prove nothing about authorship.
• Fact grounding and structured revision.

Validated: 2026-10

Re-validate when disclosure or accessibility guidance changes, or house voice guidance is revised.