Voice clone from samples
Teach the model your voice from 3 samples — then write anything in it.
Use when: Drafting emails, posts, or replies that need to sound like you.
Fill these placeholders
[SAMPLE_1][SAMPLE_2][SAMPLE_3][NEW_TASK]Replace each with your specifics — the more concrete, the better the model performs.
Want the whole thing?
Grab the full assembled prompt with section headers — paste it straight into ChatGPT or Claude.
Or copy block by block
You are a voice analyst. You can extract a writer's idiolect from 3 samples — sentence length distribution, signature openers, recurring metaphors, what they NEVER say.
Below are 3 samples of my writing. Treat them as the only ground truth for my voice.
Step 1: Build a voice fingerprint. Output it as: (a) avg sentence length and variance, (b) 5 signature words/phrases I reach for, (c) 3 things I never do, (d) my default emotional register. Step 2: Using ONLY that fingerprint, write: [NEW_TASK].
Do not insert any phrase or rhythm that doesn't appear in the samples. If a phrase is needed but not in my voice, leave a [BRACKETED PLACEHOLDER] for me to fill.
**Voice fingerprint:** - avg sentence length: ... - signature phrases: ... - never does: ... - register: ... **Draft:** <text>
Sample 1: [SAMPLE_1] Sample 2: [SAMPLE_2] Sample 3: [SAMPLE_3]
Why this works
Few-shot anchoring — explicit feature extraction
Just pasting samples and saying 'write in this voice' makes the model regress to its default. Forcing it to enumerate the fingerprint *first* anchors generation to your real patterns instead of generic 'professional' English.