# HeyAda Sound Like You

Capture a brand's voice as a short, reusable voice profile, so anyone (a person or an AI)
can write new copy that sounds unmistakably like them. Built for HeyAda
client work: we create the voice profile once per client, then every page, article and email draws
from it, so the whole presence sounds like one coherent voice instead of five different ones.

**Core principle: describe the voice that is actually there, don't impose a generic one.**
The DNA is evidence-based. Read their real words first. If they have no voice yet (thin or
templated copy), say so and propose a voice that fits the business, clearly marked as a
proposal rather than an observation.

## When to use
- Starting a client build or content engagement — produce the voice profile before writing any copy.
- Keeping a long-running content stream (blog, emails, social) on-voice over time.
- Onboarding anyone new to a client's writing.

## Inputs
- The client's website URL (or pasted samples of their existing copy).
- Optional: who they are, who they serve, and how they want to come across.

## Method

### 1. Read the real voice
Fetch the homepage plus a couple of content-rich pages (about, a service page, an article).
Read the actual sentences. Notice how they open, how they address the reader, their rhythm,
the words they reach for, what they avoid, where they are warm vs formal, confident vs modest.

### 2. Profile the voice profile across these dimensions
1. **Essence** — one sentence capturing who this voice is, plus a short archetype (e.g. "the
   straight-talking expert", "the warm neighbour", "the quiet luxury house").
2. **Tone** — 3 to 5 adjectives, each with a one-line nuance so they are not generic.
3. **Formality & register** — where it sits (casual ↔ formal), and when it flexes.
4. **Point of view & relationship** — "we"/"you"/"I", and the stance toward the reader
   (peer, guide, host, adviser).
5. **Sentence & rhythm** — typical length, structure, cadence; short punchy vs flowing.
6. **Vocabulary** — signature words and phrases they use; jargon level; and a clear list of
   words/phrases to avoid.
7. **Rules (do / don't)** — concrete, enforceable lines (e.g. "no em dashes", "no vanity
   percentages", "always name the real benefit", "never say 'leverage'").
8. **Signature moves** — how they open, how they close, recurring rhetorical patterns.
9. **In-voice examples** — 2 to 3 sample lines written in this voice (and, where useful, a
   flat sentence rewritten into the voice) so the voice profile is demonstrated, not just described.

### 3. Make it usable
End with a one-paragraph "how to use" note a writer can follow, and keep the whole thing
short enough to paste at the top of any drafting task. The DNA is a working tool, not a
brand bible — tight, specific, and immediately applicable.

## Output contract
Return:
`{ brand, essence:{line, archetype}, tone:[{word, nuance}], formality, pov, sentence_style,
vocabulary:{signature:[..], jargon_level, avoid:[..]}, rules:{do:[..], dont:[..]},
signature_moves:[..], examples:[..], how_to_use, basis:"observed|proposed" }`

## Applying the voice profile (the point of it)
The DNA is only useful if it drives writing. To draft new copy in voice: load the brand's voice profile
and put it in front of the writing task as hard constraints — essence, tone, point of view,
the signature vocabulary to use, the avoid-list to never touch, and the do/don't rules — then
write to the brief against them. A reader who knows the brand should recognise the result; if
it reads generic, it failed. Keep every piece tight and usable, no preamble. The reusable
constraint-prompt builder lives at `scratchpad/apply-dna.mjs <dna.json> "<brief>"`. Good
practice: produce a "flat" (voiceless) draft and the "in-voice" draft side by side so the
difference is visible and the voice can be checked, not just claimed.

## Boundaries
- Base the voice profile on real, public copy. If you could not read enough, say so and mark it `proposed`.
- Do not copy another brand's voice onto them; fit the voice to who they actually are.
- Plain, specific language throughout — the voice profile itself should model good writing.