Give Voice Samples So It Sounds Like You

A voice-matching technique: before asking for long-form prose, give ChatGPT two or three samples of your actual writing and have it draft in that style — showing your voice rather than describing it.

Adjectives like 'conversational' or 'punchy' are ambiguous and the model interprets them generically. Real samples carry the specifics no description captures — sentence length, rhythm, how you open, your characteristic moves. Having it extract and then apply those patterns produces prose far closer to yours than any list of style words could.

Long-form you want in your own voice — newsletters, blog posts, personal essays — especially when the writing represents you personally.

Neutral or house-style writing where your personal voice isn't wanted, and quick drafts where fit doesn't matter.

Here are 2–3 samples of my writing: <paste>. First, describe the voice patterns you notice — sentence length, tone, opening moves, recurring habits. Then draft <the piece> in that voice, applying those patterns.

After analysing three of your past newsletters, ChatGPT drafts the new issue with your habit of short opening hooks and mid-paragraph asides intact — not generic newsletter prose.

Save the voice description it produces and reuse it as a reusable style brief — or drop it into a Project's instructions — so every future draft starts from your voice without re-pasting samples.

• Describing your voice in adjectives instead of showing samples.
• Giving one sample, too little to reveal consistent patterns.
• Skipping the 'name the patterns' step, so the model copies surface, not structure.