**Version:** 1.1.0 **Updated:** 2026-06-30 **Freshly updated:** v1.1.0 wires the voice profile into the rest of the brand system. The extracted profile now persists to per-site memory via `respira_get_option` (diff first) + `respira_update_option`, cross-links explicitly with
Installs just this skill. Get the whole plugin for auto-invocation.
⚡ How it fires
How this skill gets triggered: by you, by Claude, or both.
Fires itselfClaude auto-loads it when your prompt matches the work.
You can call itInvoke it directly when you want it.
Slash command/brand-voice-synthesizer
👁️ Context preview
The summary Claude sees to decide when to auto-load this skill.
**Version:** 1.1.0 **Updated:** 2026-06-30 **Freshly updated:** v1.1.0 wires the voice profile into the rest of the brand system. The extracted profile now persists to per-site memory via `respira_get_option` (diff first) + `respira_update_option`, cross-links explicitly with
# Brand Voice Synthesizer
**Version:** 1.1.0
**Updated:** 2026-06-30
**Freshly updated:** v1.1.0 wires the voice profile into the rest of the brand system. The extracted profile now persists to per-site memory via `respira_get_option` (diff first) + `respira_update_option`, cross-links explicitly with the Page Template Library and Design System Synthesizer skills so copy and layout share one brand foundation, and adds a brand-consistency report built from `respira_generate_activity_report` so you can see how on-voice recent content actually is.
**Category:** intelligence
**Status:** stable
**Requires:** Respira for WordPress plugin 7.1+ + MCP server
---## Description
Read 5–10 published posts on a WordPress site and extract the **brand voice** — tone, lexicon, sentence patterns, person used, formality, signature phrases, phrases the site never uses. Persist it to the site so every future content-writing skill produces copy that sounds like the brand, not like generic AI.
This is the verbal counterpart to the [Design System Synthesizer](https://respira.press/skills/design-system-synthesizer). The two together form a complete brand foundation that every future content-generation skill references.
---## What it produces
A structured `brand_voice` artifact stored at the site level. Schema:
```json
{
"version": "1.1.0",
"synthesized_at": "2026-05-24T14:30:00Z",
"synthesized_from": ["/blog/post-a/", "/blog/post-b/", "..."],
"good_paragraph": "We shipped the redesign last week. The short version is: fewer pages, more density, one CTA per screen. Conversion is up 18% on the new pricing page so far. We'll know more by end of month.",
"bad_paragraph": "We are thrilled to announce the launch of our revolutionary new design! This best-in-class experience leverages cutting-edge UX innovations to deliver unparalleled value!"
}
}
```
The artifact is stored via `respira_update_option('respira_brand_voice', ...)` for v7.1. Same migration path to `respira_intelligence_artifacts` in v7.2 as the design system.
---
## When to Use
- First time setting up an AI workflow for a content-heavy site
- Before running any content-generation skill that produces written copy
- After a brand voice update (new style guide, new chief content officer, rebrand)
- Quarterly refresh — voice drifts; resynthesize to catch it
1. **Person:** Most-used pronoun (I / we / you / no-pronoun). If mixed, note the mix.
2. **Formality:** Read 5–10 random paragraphs. Place on a scale: casual / approachable_professional / formal / academic.
3. **Sentence length:** average, median, p90.
4. **Paragraph length:** average sentences per paragraph.
5. **Reading grade:** Flesch-Kincaid or similar.
6. **Tone descriptors:** 3–5 adjectives based on the corpus. Confident? Cautious? Playful? Direct? Evidence-driven? Storyteller?
7. **Signature phrases:** Repeated multi-word phrases that show up across multiple posts. These are the writer's tells.
8. **Common openers/closers:** How do paragraphs typically start and end?
9. **Lexicon preferred:** Verbs and nouns this writer reaches for. ("ship" instead of "release"; "wire" instead of "integrate"; "land" instead of "launch")
10. **Lexicon avoided:** Words *conspicuously absent* from the corpus that competitors or generic AI would use. ("leverage", "synergy", "best-in-class", "innovative", "revolutionary".) The absence is the signal.
12. **Structural patterns:** Do posts open with a question? Use headings? Bullet lists? End with a CTA?
### Step 5 — Show the synthesized voice to the user
Output a human-readable summary:
```markdown
## Brand voice synthesized for {site_url}
Based on {n_samples} published posts ({total_words_sampled:,} words sampled, mostly from the last 6 months).
**Person:** {person} ("we" / "I" / "you" / mixed)
**Formality:** {formality}
**Sentence length:** ~{sentence_length_avg_words} words avg ({sentence_length_p90_words}-word p90)
**Reading grade:** {reading_grade_avg}
**Tone:** {tone_descriptors joined}
**Signature phrases this site uses repeatedly:**
{signature_phrases as bulleted list}
**Words this site reaches for:**
{lexicon_preferred as inline list}
**Words this site conspicuously avoids:**
{lexicon_avoided as inline list}
**Punctuation quirks:**
- em dashes: {em_dash_usage}
- exclamation marks: {exclamation_marks}
- oxford comma: {oxford_comma}
**Structural patterns:**
- Opens with a question? {opens_with_question}
- Uses headings? {uses_headings}
- Ends with a CTA? {ends_with_call_to_action}
**A paragraph that sounds like you:**
> {examples.good_paragraph from the corpus}
**A paragraph that does NOT sound like you (generic AI):**
> {examples.bad_paragraph as a constructed counter-example}
```
Ask the user: *"Does this match how you'd describe your voice? Anything to add or correct?"*
### Step 6 — Persist to per-site memory
The `respira_brand_voice` option is the site's voice memory: it survives across sessions and every future content skill reads it. Persist it carefully.
1. **Diff first.** Read `respira_get_option('respira_brand_voice')`. If a profile already exists, show the user what changed (person, tone, added/removed lexicon) before overwriting — never clobber an existing voice silently.
2. **Write.** After confirmation, `respira_update_option('respira_brand_voice', <json>)`.
3. **Verify.** Read it back with `respira_get_option('respira_brand_voice')` and confirm it round-tripped.
Output: *"Brand voice saved to this site's memory. Every Respira content-writing skill from this point forward will reference it. The avoided-words list is the strongest signal — the agent will refuse to use those words even if the prompt suggests them."*
### Step 7 — Brand-consistency report (optional)
The voice profile is also a yardstick: it tells you whether content *already on the site* sounds on-brand. Pull the recent work log with `respira_generate_activity_report` (it returns what was published/edited and when), then sample those pages/posts and score them against the saved profile:
- **On-voice:** matches person, sentence length, and reaches for the preferred lexicon.
- **Off-voice:** uses avoided words, wrong person, or marketing-superlative drift.
Report it plainly: *"Of the last 12 published posts, 9 are on-voice. 3 drift toward the avoided lexicon ('revolutionary', 'best-in-class') — likely the templated launch posts. Want me to rewrite those to match?"* This turns the voice profile from a passive artifact into an active consistency check, and pairs naturally with the rewrite skills.
---
## How other skills use the brand voice
Once persisted, future content-writing skills (page generators, blog post drafters, social post composers) call `respira_get_option('respira_brand_voice')` at the top of their workflow. They use the voice to:
- Pick pronouns matching the site's person
- Match sentence length and reading grade
- Reach for the preferred lexicon
- Refuse to use the avoided lexicon
- Match punctuation patterns
- Match structural patterns (open with question? end with CTA?)
The **avoided lexicon** is the strongest signal. When the agent is about to write "revolutionary" or "best-in-class," the voice artifact says *"this site never uses that word"* and the agent picks a more honest verb.
### Pairs with
The brand voice is one half of the brand foundation. It works best alongside two sibling skills:
- **[Design System Synthesizer](https://respira.press/skills/design-system-synthesizer)** — the visual half (colors, typography, spacing, components, stored as `respira_design_system`). Voice covers *how the words read*; the design system covers *how the page looks*. Run both so generated content is on-brand top to bottom. The Design System Synthesizer's style-guide page even links back here for the full voice.
- **Page Template Library** — when it assembles a page from a saved template, it should read `respira_brand_voice` so the placeholder copy it drops in already sounds like the site, not like lorem-ipsum or generic AI. Voice supplies the words; the template supplies the layout.
If only one of the two artifacts exists, content skills should still use what's there — but flag to the user that running the missing synthesizer would tighten the result.
---
## Hard rules
- The voice is observed, not designed. Don't invent. Every descriptor traces to evidence in the corpus.
- The avoided lexicon is the **inverse signal** — words conspicuously absent. Don't include common stop words ("the", "and", "is"). Include words that competitor sites would use heavily but this site doesn't.
- The "good paragraph" example must be a real paragraph from the corpus, quoted verbatim. The "bad paragraph" example is a constructed counter-example using the avoided lexicon.
- Never overwrite an existing brand voice silently. Show the diff before saving.
- If the corpus is mixed-author (multiple writers' voices), say so honestly: *"This site has 3 distinct voices in the sampled posts. Pick one author to synthesize from, or capture all 3 as separate voices."*
---
## Tooling
**Reading the corpus (source of truth)**
- `respira_get_active_site`
- `respira_get_site_context`
- `respira_list_posts`
- `respira_read_post`
- `respira_extract_builder_content`
**Persisting + reporting**
- `respira_get_option` — diff an existing `respira_brand_voice` before overwriting
- `respira_update_option` — write the voice profile to per-site memory
- `respira_generate_activity_report` — recent published/edited work for the brand-consistency check
Pairs with the Design System Synthesizer (`respira_design_system` via `respira_get_option`) and the Page Template Library. Use only `respira-wordpress` MCP tools; never invent a tool name.
---
## Telemetry
Records: site URL hash, number of posts sampled, total words analyzed, person detected, formality detected, success/failure. No actual phrases, no avoided words, no example paragraphs are sent.
