95% Confidence Brand Voice Skill
Interview deeply, encode brand rules, and generate platform-specific content
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 5
- Confidence
- 99%
This method turns brand knowledge that normally remains implicit in a creator's head into a reusable AI writing skill. Begin with a request to create the skill, then have the AI conduct a structured interview until it is 95% confident that it understands the desired output. The interview captures platforms, content pillars, audience, post length, personal boundaries, calls to action, platform differences, positive traits, prohibited tendencies, and representative writing samples. Those inputs become persistent context that can be invoked whenever content is requested. Testing occurs in a new conversation, which checks whether the saved skill works independently of the original interview. The result is not a single polished prompt but a repeatable content-production asset that can produce differentiated drafts for LinkedIn, Facebook, Twitter, and other channels without restating the brand rules each time.
Origin
Extracted from Marketing Against The Grain, where Sabrina Romanov demonstrates how she teaches Claude Cowork her brand voice before using it for high-volume social publishing.
Core principles
- 01Treat initial prompting as discovery, not final configuration
- 02Capture concrete preferences instead of relying on generic style labels
- 03Use successful writing samples as evidence of the desired voice
- 04Model platform differences explicitly
- 05Continue interviewing until confidence reaches a defined threshold
How to run it
- 1
Define the Skill's Job
Ask the AI to create a reusable content-writing skill for social posts about the business and personal brand. Add any business context it cannot reliably infer.
Pro tip Reference relevant past conversations when they contain accurate business context.
Watch out Do not treat a vague request to write like you as sufficient training.
- 2
Set a Confidence Gate
Tell the AI to interview you until it is 95% confident that its outputs will reflect your brand. This creates an explicit stopping condition for discovery.
Pro tip Answer with concrete examples and exclusions rather than abstract adjectives.
Watch out Stopping after a few superficial questions leaves important voice rules unstated.
- 3
Map the Content System
Document platforms, content pillars, target audience, post lengths, calls to action, personal-disclosure boundaries, and platform-specific differences.
Pro tip Explain how the same idea changes from one platform to another.
Watch out A single universal format can erase meaningful channel differences.
- 4
Supply Voice Evidence
Describe what sounds like you, what never sounds like you, and provide successful writing samples. Let the AI derive patterns from real work alongside your stated preferences.
Pro tip Choose samples you would genuinely want future content to resemble.
Watch out Weak or unrepresentative samples can encode the wrong style.
- 5
Save and Test the Skill
Save the completed interview as a reusable skill, open a new conversation, and invoke it on a realistic content task. Confirm that the resulting drafts reflect both the shared voice and each platform's requirements.
Pro tip Explicitly invoke the skill during early tests so you know it was applied.
Watch out Do not assume a skill works merely because the setup conversation sounded accurate.
In the wild
Romanov asks Claude to find a receipts image in her downloads folder. Claude identifies a screenshot showing a new Facebook page's organic performance, analyzes its contents, and drafts distinct LinkedIn, Facebook, and Twitter posts using the stored brand preferences, including her platform-specific preference for very short tweets.
→ One local artifact becomes three differentiated drafts without re-explaining the creator's voice.
A founder defines audience, content pillars, preferred post length, personal boundaries, signature endings, and channel differences, then supplies two strong newsletter and LinkedIn examples. The AI interviews the founder until the confidence threshold is reached and stores the answers as a reusable writing skill.
→ Future drafts begin from a documented voice system rather than a generic prompt.
Common mistakes
Using Only Style Adjectives
Words such as professional or authentic are too ambiguous to reproduce a distinctive voice. Pair preferences with concrete examples, exclusions, and platform rules.
Ignoring Platform Differences
Applying one format everywhere can make posts feel copied rather than native. Define how length, structure, media, and calls to action differ by channel.
Skipping the Fresh-Chat Test
The setup conversation may hide missing context because its history remains available. Test the saved skill in a new chat to verify that the reusable artifact is complete.
Is it for you?
Best for
It is best for creators and small marketing teams that repeatedly publish across several platforms.
Not ideal for
It is not ideal for one-off content where the author cannot provide preferences, examples, or feedback.
From the transcript
“Interview me until you're 95% confident the outputs will reflect my brand.”
“What are my content pillars? How long are my posts? What are some uh some tidbits about how I sound, what sounds like me, what…”
“It's like a repeatable task that contains a bunch of information, context, and preferences so that you don't have to explain it again and again…”
From the episode
I Run 250+ Social Media Posts/Week… Alone (Claude Code Workflow)