SOCIAL

Match an existing social voice

Use when you are ghostwriting for a founder or colleague and the drafts keep sounding like you instead of them.

voice-match-from-samples.md

Download .md

You are analyzing a person's writing voice and then writing in it.

Voice samples, written by {{PERSON_NAME}}:
"""
{{VOICE_SAMPLES}}
"""

Step 1. Produce a voice profile with these fields, evidence-backed. Quote a short phrase from the samples for each:
- Average sentence length and range.
- Vocabulary level and any recurring words.
- How they open posts.
- How they close posts.
- Punctuation habits, including whether they use lists, parentheses, or one-line paragraphs.
- What they never do.
- Their default stance: teaching, arguing, reporting, or thinking aloud.

Step 2. Write a new post on this topic: {{NEW_TOPIC}}, length {{WORD_COUNT}} words, matching the profile.

Step 3. Self-audit: list every line in your new post that breaks the profile, and rewrite those lines.

Rules:
- Do not smooth their voice. If they write blunt fragments, write blunt fragments.
- Do not add opinions they have not expressed in the samples. Mark anything speculative as [CHECK WITH {{PERSON_NAME}}].
- No em dashes unless the samples use them. No "it's not just X, it's Y" unless it appears in the samples.
- Do not invent facts about their company, results, or history.

Fill in before running

Replace each placeholder with your own detail. The more specific you are, the less the model invents.

{{PERSON_NAME}}
{{VOICE_SAMPLES}}
{{NEW_TOPIC}}
{{WORD_COUNT}}

Getting a better result

Five to eight real posts is enough; more samples from different years muddies the profile.
Keep the voice profile in a file and paste it into future prompts instead of re-analyzing every time.
The "what they never do" field is worth more than all the others combined.

Questions about this prompt

When do I need this rather than describing the voice I want?

When you are ghostwriting and the drafts keep sounding like you. Telling a model to write in a warm, conversational style returns a generic voice that belongs to nobody. This builds a profile from quoted evidence in the samples first, which is why it can then audit its own draft against it.

What do I need in front of me?

Five to eight real posts by {{PERSON_NAME}}, per the tips, from roughly the same period rather than spread over years. Also the new topic and a word count. Samples that someone else edited heavily are worse than none, because the profile then describes the editor and not the person you are writing as.

What comes back, and what should I keep?

A seven field voice profile with a quoted phrase as evidence for each, the new post, then a self-audit rewriting every line that breaks the profile. The what they never do field is worth more than the rest combined. Save the profile to a file and you never pay for the analysis twice.

What is the mistake?

Letting it smooth the voice into something more polished than the person actually writes. Blunt fragments stay blunt fragments. The bigger risk is opinion drift: anything marked [CHECK WITH {{PERSON_NAME}}] is a view they have not expressed in the samples, and publishing one under their name ends the arrangement.