How to Teach Claude Your Exact Brand Voice (and Stop Sounding Like AI)
By Ashley Gross
·
·
Claude

Feeding Claude three blog posts gets you a decent initial draft, but the model quickly slips back into generic filler on longer pieces. You do not have to rewrite every paragraph by hand or settle for sounding like an algorithm. Building a structured style guide gives you drafts that sound like you on the first pass.
What You Need
Plan level: Pro, Team, or Enterprise account (Custom Styles, Projects, and Skills require paid tiers).
Source material: Five to ten pieces of your best writing.
Time: 45 minutes.
Step-by-Step Walkthrough
1. Extract Your Personal Voice Profile
Most operators try to describe their style from memory, which leads to vague directives. Instead, upload five of your best articles to Claude and run an extraction prompt. Ask the model to map your actual structural patterns, sentence rhythm, and rhetorical choices.
Why it matters: Grounding your style guide in real text prevents guesswork and gives Claude concrete rules to follow.

2. Build Your Banned List and Anti-AI Patterns
AI models rely heavily on predictable transition words, dramatic adjectives, and specific punctuation marks like em dashes. Create a explicit list of banned words, banned phrases, and forbidden punctuation. State clearly what Claude must use instead.
Why it matters: Telling Claude what to exclude eliminates ninety percent of generic AI slop before drafting begins.

3. Format the Voice Guide With Contrast Examples
Rules alone are easy for LLMs to ignore during long generation runs. Combine your rules with before-and-after contrast pairs. Show a weak AI sentence alongside your exact manual rewrite to ground the instructions in visual examples.
Why it matters: Contrast examples show Claude the exact visual boundary between AI patterns and your human voice.

4. Choose Where to Store Your Voice Profile
Select the right location in Claude based on your workspace setup and plan:
Styles: Create a custom Style from the style menu in the chat box for quick tone toggles across simple chats.
Projects: Paste your full style guide into Project Instructions and upload sample pieces to Project Knowledge. This works best for long-form content projects where team members need identical output.
Skills: Package your voice rules into a custom Skill if you want to invoke specific editing workflows across different team spaces.
Why it matters: Putting instructions in the right structural container ensures your rules stay active without re-pasting them every session.

5. Test on a Real Piece and Refine
Run a real article topic through your new setup. Read the draft aloud to find lingering AI habits. If you spot a generic phrase or an em dash, add that specific pattern directly to your banned list.
Why it matters: Real-world testing catches edge cases that your initial voice extraction missed.


6. Run an Automated Self-Check Pass
Never read a raw first pass. Include a final instruction that forces Claude to audit its own text against your rules before showing you the output.
Why it matters: Forcing an automated audit catches accidental rule breaks before you spend time editing.

Real Example: The Alliance System
At The Alliance, we maintain strict publishing standards across all long-form articles. Here is how our internal system transforms raw drafts.
The Input Draft (Before Rules)
"In today's fast-paced world, leveraging artificial intelligence is a game-changing strategy—allowing founders to seamlessly optimize their daily schedules and unlock amazing productivity gains."
Applied Voice Profile Rules
No em dashes.
No banned words (leverage, seamlessly, game-changing, amazing, in today's fast-paced world).
Active voice only.
Sentences under 25 words.
The Final Output (After Rules)
"AI saves time when you structure your prompts correctly. Set up your system once, eliminate manual work, and finish your tasks faster."
The revised version cuts the fluff, removes the forbidden punctuation, and delivers a clear statement in active voice.
Common Mistakes
Using vague adjectives: Rules like "be friendly but professional" give Claude zero actionable constraints. Use concrete rules, such as "keep sentences under 25 words" or "never use passive voice."
Skipping the banned list: Omitting a banned list allows AI models to default back to high-frequency training patterns like em dashes and dramatic transitions.
Omitting visual contrast pairs: Giving rules without before-and-after examples forces Claude to guess what compliance looks like.
Take It Further
Add a permanent "Voice Scan" line to your Project Instructions. Tell Claude: "If a draft contains an em dash or a banned word, pause immediately, delete the word, and rewrite the sentence before continuing." This persistent instruction keeps long generation runs aligned with your style guide.
