Claude Projects locks your voice prompt into every conversation. Setup walkthrough, the template structure that works, and the batching workflow that ships 30 LinkedIn posts in 90 minutes.
Claude Projects (Claude Pro feature, £18/month) is the strongest tool for voice-consistent content production in 2026. Pastes your voice prompt as a system instruction, accepts reference files, and applies both to every conversation in the project. 30-45 minutes to set up, then 90-minute batching sessions that ship a month of voice-matched content.
Three structural reasons:
STEP 1 · 5 MIN
Claude Projects is a Pro-tier feature at £18/month. The free tier doesn't include it. Sign up at claude.ai. Subscription pays for itself within the first hour of use for any solopreneur producing weekly content.
STEP 2 · 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. Three months from now you'll have several projects and the description matters.
STEP 3 · 10 MIN
In the project's settings, find the "Custom Instructions" or "System Prompt" field. Paste your full voice prompt — voice essence, mechanical rules, banned words, tone by context, signature moves. The 500-800 word document built per how to build a voice prompt.
Add a one-line preamble at the top:
STEP 4 · 15 MIN
Reference files give Claude examples to anchor against. Upload:
STEP 5 · 10 MIN
Open a fresh conversation in the project. Run these three tests:
Read each output. If voice match is weak (under 70%), the voice prompt needs tightening. Iteration protocol.
Once the project is set up and tested, weekly content production runs like this:
Minute 0-15: Pillar selection. Pick one content pillar for the month or week. Examples: "voice systems for solopreneurs," "the asset-ownership shift in AI services," "what 30 voice builds taught me." One pillar = 8-12 posts.
Minute 15-30: Hook batching. In your LinkedIn Voice project, run: "Give me 12 hooks for this pillar: [pillar]. Use the seven LinkedIn patterns from the voice prompt. Vary structure across hooks." Pick the 8-10 strongest.
Minute 30-75: Post drafting. For each chosen hook, run: "Write a full LinkedIn post starting with this hook: [hook]. Apply the LinkedIn voice rules. 200-300 words. Include one specific named example." Claude produces the draft. Light edit.
Minute 75-90: Final pass + scheduling. Read each draft sentence by sentence. Tweak any rule violations. Load into Buffer or your scheduling tool. Done.
Total output: 8-10 LinkedIn posts ready to ship across 2-3 weeks. Repeat monthly.
System prompt: full voice prompt with LinkedIn-specific tone-by-context section emphasised.
Reference files: 5-10 LinkedIn posts in your voice, ICP description, hook library, the seven LinkedIn patterns from how to write LinkedIn posts AI doesn't suppress.
Best use: post drafting, hook batches, comment drafting, profile rewrites.
System prompt: voice prompt with email-specific tone (warmer, first-person, conversational).
Reference files: 5-10 emails you've sent that sound like you, your subscriber persona, your offer description.
Best use: welcome sequence drafts, broadcast emails, sales emails, sequence variants.
System prompt: voice prompt with sales-page-specific tone (sharper, direct, names objections).
Reference files: 2-3 sales pages you admire (with notes on what works), your offer details, common prospect objections, customer testimonials.
Best use: full sales page drafts, landing page copy, sales email sequences.
Mistake 1: Using one project for everything. Different content formats need different tone calibration. One mega-project that mixes LinkedIn, email, and sales page work produces blurred output. Three small projects work better than one big one.
Mistake 2: Skipping reference files. The system prompt alone produces 70-80% voice match. Adding 5-10 example posts as reference files lifts this to 80-90%. Most users skip this step. The 15 minutes to upload files compounds across every future conversation.
Mistake 3: Never updating the project. Voice evolves. ICP shifts. Quarterly: review the voice prompt, refresh reference files with recent posts, update banned words. 30 minutes per quarter.
Mistake 4: Treating Projects as conversations. Projects are workspaces, not conversations. Within one project, run multiple separate conversations for different content tasks. Don't try to do everything in one thread — Claude's context window will eventually fill.
Detailed answer: ChatGPT vs Claude for LinkedIn content. Short answer:
DFY Voice System includes the voice prompt, custom GPT, and Claude Project setup ready to import. The Voice Build methodology, delivered in 2-3 working days. £497 founder pricing.
See The Voice BuildA Claude Pro feature that locks a custom system prompt and reference files into a dedicated workspace. Every conversation in the project inherits the system prompt.
Yes, £18/month. The free tier doesn't include the feature.
Claude Projects follow long voice prompts more reliably for long-form content. Most serious users run both — ChatGPT for hooks, Claude for drafts.
Multiple. The pattern: one project per content area (LinkedIn voice, email voice, sales page voice).
Voice prompt (as system prompt), 5-10 example posts in your voice (as reference files), framework documents, ICP description.