Step-by-step process to make Claude produce on-voice content using Projects, voice prompts, and reference files. Why Claude beats ChatGPT for long-form drafts and how to set it up in 45 minutes.
You can't literally "train" Claude on your writing samples — no persistent memory between conversations. What you can do is engineer it to produce on-voice content every time using Claude Projects + a 500-800 word voice prompt. With a properly-built voice prompt, Claude produces 80-90% on-voice output on first draft for long-form content. Setup time: 45 minutes once the voice prompt exists.
Quick clarification: Claude can't be fine-tuned on individual writing samples in the consumer interface. The fine-tuning APIs exist for developers but are aimed at high-volume use cases, not solopreneur content production.
The practical equivalent: a structured voice prompt loaded as the system instruction in a Claude Project. The project becomes a workspace where every conversation starts with your voice rules already loaded. Claude follows those rules across long content drafts more reliably than ChatGPT does — which is why this setup is the recommended path for voice-critical long-form content.
STEP 1 · 5 MIN
Claude Projects is a Pro feature at £18/month. Sign up at claude.ai. The free tier doesn't include Projects. The Pro subscription pays for itself within the first hour of use for any solopreneur producing weekly content.
STEP 2 · 4-6 HOURS (FIRST TIME) OR PASTE EXISTING
If you don't already have a voice prompt: build one using the five-section structure. Voice essence, mechanical rules, banned words, tone by context, signature moves. 500-800 words. 4-6 hours of focused work.
Alternative: DFY Voice System at £497 builds the voice prompt for you in 2-3 working days using The Voice Build methodology.
If you already have a voice prompt built for ChatGPT, paste the same one into Claude — voice prompts are plain text and tool-agnostic.
STEP 3 · 5 MIN
Click "Projects" in the Claude sidebar, then "Create project." Name it after the content area:
Add a clear description so future-you knows what the project is for.
STEP 4 · 5 MIN
In the project's custom instructions field, paste the full voice prompt with this preamble at the top:
You are a writing assistant for [Your Name]. Every response must follow the voice rules below. Before producing any output, mentally check the rules. Never deviate from the mechanical rules or banned words list. Voice rules: [paste full voice prompt here]
STEP 5 · 15 MIN
Reference files give Claude examples to anchor against. Upload:
Claude references these across all conversations in the project, eliminating the need to re-paste them.
STEP 6 · 15 MIN
Open a fresh conversation in the project. Run these three tests:
Read each output sentence by sentence. Apply this test: would I actually write this?
If voice match is weak, identify which rules failed and tighten the voice prompt. Common failure modes: vague mechanical rules, missing banned words, wrong tone for the format. Three iteration rounds typical to land at 80-90% first-draft voice match.
Across the 30+ voice builds we've shipped, Claude consistently outperforms ChatGPT on these specific tasks:
Honest caveat: Claude isn't strictly better than ChatGPT. The tasks where ChatGPT outperforms:
Most serious solopreneurs run both — ChatGPT for hooks and short-form, Claude for long-form drafts. Detailed comparison: ChatGPT vs Claude for LinkedIn.
Once your project is set up:
Min 0-15: Pillar selection. Pick one content pillar for the week. Examples: "voice systems," "asset ownership in AI services," "what 30 voice builds taught me."
Min 15-30: Hook batch (in ChatGPT). Quick batch of 12 hooks for the pillar. Pick the 8-10 strongest.
Min 30-75: Post drafts (in Claude Project). For each chosen hook, run: "Write a full LinkedIn post starting with this hook: [hook]. Apply LinkedIn voice rules. 200-300 words. Include one specific named example." Claude produces strong drafts; light edit.
Min 75-90: Final pass + scheduling. Read each draft. Tweak any rule violations. Load into Buffer or scheduling tool.
Mistake 1: Vague voice prompts. "Friendly tone" produces default Claude output. Specific instructions ("12-18 word sentences, banned words: leverage / synergy / cutting-edge, opens with reader's pain") produce on-voice output.
Mistake 2: Skipping reference files. System prompt alone gives 70-80% voice match. Adding 5-10 example posts as reference files lifts to 80-90%. Most users skip this step.
Mistake 3: One mega-project. Different formats need different tone calibration. Three small projects beat one big project.
Mistake 4: Treating projects as conversations. Projects are workspaces. Run multiple separate conversations within one project for different tasks. Don't try to do everything in one thread.
DFY Voice System ships your voice prompt + Claude Project setup + Custom GPT setup + hook library. £497 founder pricing. The Voice Build methodology, applied to your existing writing.
See The Voice BuildNot literally — no persistent memory. But you can engineer it to produce on-voice output every time using a structured voice prompt in a Claude Project.
Claude follows long voice prompts more reliably for long-form content. Stronger first-draft voice match. ChatGPT wins on hooks and short-form.
45 minutes once the voice prompt exists. The voice prompt itself takes 4-6 hours DIY or 2-3 working days DFY.
Projects require Pro (£18/month). Without Pro, paste the voice prompt as the first message of each conversation.
Yes. Voice prompts are plain text and tool-agnostic. Same prompt works in Claude, ChatGPT, Gemini, anywhere.